Abstract:

新型电力系统逐渐呈现出高比例新能源和高比例电力电子设备的“双高”趋势。以风、光为代表的新能源存在出力功率波动性、间歇性和随机性以及控制非线性等特点，叠加用电侧新兴电力电子化负荷的多样性与冲击性，容易引发配电网多频率的谐波振荡，从而影响用电端的电能质量。为有效提升新型电力系统用电侧的安全稳定，亟须开展面向新型电力系统的谐波抑制与电能质量提升技术研究。

Abstract:

The impedance-based method is favored by engineering because it can analyze system stability under conditions with the unknown device control structure or parameters. Considering that the impedance characteristics of AC grid-connected equipment represented by power electronic converters are easily affected by the AC steady-state operating point,quickly deriving an impedance model for any operating condition of the converter using black-box identification can greatly improve the efficiency of stability analysis. The neural network-based can eliminate the limitations of the least squares method-based identification,this paper further improves the neural network design to significantly improve its interpretability. In the data collection stage,the frequency sweep method is used to obtain the frequency response of the closed-loop impedance model under enough operating conditions. In the model training stage,taking into account the latent features of the converter impedance model,a neural network with the same number as the disturbance frequency was designed,and the Levenberg-Marquardt algorithm with Bayesian regularization integrated is used to enhance the generalization ability of the network trained with a small dataset. In the model verification phase,the network is fed with set operating conditions,achieving highly accurate identification of stable operating conditions and offline prediction.

Abstract:

In weak grid,a multi-inverter grid-connected system may be simultaneously affected by large grid impedance and background harmonics. The interaction between inverter and grid is enhanced by the large grid impedance, resulting in resonance problems. Furthermore,the grid-connected voltage and current waveforms are distorted by the background harmonics, which makes the system unable to meet the grid-connected requirements. In view of this,firstly,an impedance model for multi- inverter grid-connected system is established,and the mechanism behind waveform distortion and resonance in weak grid is clarified. Subsequently,a control strategy combining improved grid voltage feedforward with parallel adaptive active damping is proposed. The improved grid voltage feedforward is used to reshape the impedance of the multi-inverter grid-connected system in order to mitigate background harmonics effects, while the active damper is employed to synthesize virtual resistance for suppressing system-grid resonance. When there are changes in working conditions of the system, the damping effect can be further improved by adaptively adjusting virtual resistance values through compensation. Simulation results show that background harmonics and resonance problems can be effectively suppressed by the proposed strategy,and the adaptability of multi-inverter grid-connected systems in weak grid is enhanced.

Abstract:

In existing estimation methods for utility harmonic impedance,it is commonly assumed that the harmonic impedance remains invariant,which often diverges from actual conditions. In practice,both the utility harmonic impedance and background harmonics typically time-varying with operating conditions. For the time interval between two sample points,large numerical value probably gives rise to more conspicuous difference between the impedance and the background harmonics at the corresponding time. Based upon the information of the sample points with a far gap,it is difficult to estimate the impedance value of the sample points of concern. As a consequence,a brand new time-varying utility harmonic impedance estimation method is put forward based on locally-geographically weighted regression. Firstly,a weight matrix is constructed based on time interval,assigning smaller weights to sample points with larger intervals from the points of interest. Locally weighted regression (LWR) is then applied to initially estimate the utility harmonic impedance and background harmonic reference values. Secondly,the impedance reference value is used to modify the regression equation to reduce the under determination of the original regression equation. To screen out the sample points that are similar to the background harmonics of the sample points of concern,the background harmonic reference value is simultaneously utilized as the prior information. On the basis of the screened samples,the background harmonic voltage and the utility harmonic impedance at each point are coped well with by geographically weighted regression (GWR). Under strong background harmonic fluctuations,the recommended method can not only identify abrupt changes in impedance,but also estimate the trend of utility harmonic impedance. Lastly,simulation and case studies demonstrate that the proposed method improves estimation accuracy by approximately 40% compared to traditional constant harmonic impedance estimation methods,and by around 30% compared to existing time-varying impedance estimation methods.

Abstract:

Mining different over-limit patterns of the low voltage is very important to guide the management of low voltage issues in users. Due to the complexity and the ever-changing nature of voltage,the over-limit patterns of low voltage are always inherently unknown in users. A pattern mining method for low voltage in users based on hierarchical affinity propagation clustering (HAP) is proposed in this paper. Firstly,large-scale voltage data is clustered into several clusters using the HAP clustering algorithm,and these clusters are regarded as the different over-limit patterns of low voltage. Then,four indices are defined from two aspects of the duration and amplitude to characterize the features of the clusters. The features of the over-limit patterns are then drived by calculating the indices for each cluster. Finally,the proposed method is applied to a real dataset,effectively mining four over-limit patterns of low voltage. The characteristics of different patterns provide the important information for the supervision and analysis of low voltage issues in users,and the priorities of low voltage problems management in users can be well leveled.

Abstract:

The voltage crossing limit problem is gradually highlighted and the stable operation of the system is becoming increasingly complex in the context of building a new distribution system. In this paper,based on the holomorphic embedding method,the offset characteristics of node voltage indicator trajectories under topology changes are studied,and a weak node identification method considering fault reconfiguration in distribution networks is proposed. Firstly,the node voltage index and distribution network voltage visible safety region are derived and constructed based on the idea of decoupling power system. Then,the node voltage index trajectory is solved based on the holomorphic embedding method. The concept of voltage index offset distance characterizing the trajectory characteristics is defined,and the probabilistic voltage index trajectory solution method is proposed taking into account the topology changes after the fault in distribution network. Finally,the weak node evaluation index system and identification method are proposed according to the relative position relationship between the voltage visible security region of distribution network and the node voltage index trajectory by considering the normal state and N-1+1 fault state operation conditions of the distribution network. Using the IEEE 33-node distribution system as an example for arithmetic analysis,the results show that the proposed method can realize the visual monitoring of the node voltage status and accurately identify the weak nodes.

Abstract:

During the operation of group trains,the trains frequently pass through the neutral zone. Then,the generated closing overvoltage will cause serious insulation damage to the traction power supply system and the trains' high-voltage equipment box. In order to qualitatively analyze the closing overvoltage characteristics of group trains,the equivalent simplified dynamic circuit of the traction power supply system of group trains is established according to the principle of circuit equivalent simplification. The expression of the train terminal voltage at the moment of closing is solved by reasonable assumptions,which reveals the formation reasons of the closing overvoltage in terms of the qualitative relationship. Based on the transient model of the group train traction power supply system,the influence of different train numbers,distribution locations and operating power on the closing overvoltage of group trains is investigated. Finally,a simulation model is built in PSCAD/EMTDC to verify the correctness of the theoretical analysis results. The results show that with the increase of the number of group trains,the centralization of the operation location,and the traction power,the peak closing overvoltage decreases. The results can also effectively guide the closing overvoltage mitigation of group train neutral zone,which reduces the closing overvoltage when the group trains pass through the neutral zone.

Abstract:

In order to improve the performance of dynamic voltage restorer (DVR) in controlling asymmetrical voltage sag fault,a fractional order plus-negative sequence decoupling control strategy based on instantaneous symmetric component method is proposed and applied to three-phase four-bridge inverter DVR system. Firstly,the partial sequence decoupling mathematical model of the DVR system is obtained by using the symmetric component method. Secondly,to solve the problem that the traditional symmetric component method is not suitable for the analysis of the transient process of the system,the positive and negative zero sequence components of the instantaneous value of the asymmetric voltage and current signal are obtained by using the instantaneous symmetric component method. Then,the integer order plus-negative order decoupling control strategy is extended to fractional order,and the frequency domain method is combined with the robustness condition when the gain changes to correct the relevant parameters. Finally,the simulation results verify the feasibility of the fractional order plus-negative sequence decoupling control strategy applied to the DVR system,and the proposed strategy has better dynamic response speed and anti-interference performance than the traditional plus-negative sequence decoupling control startegy.

Abstract:

A lightweight power quality disturbances (PQDs) recognition model that integrates convolutional neural network (CNN) and Transformer (CaT) is proposed to address the high number of parameters and computational complexity in existing deep learning-based models. Depthwise separable convolutions are first employed to extract local features from the disturbance signals. An efficient softthreshold block is then introduced to reduce noise and redundant features without significantly increasing the model's parameters or complexity. The Transformer model is used to capture global features of the disturbance signals. Finally,pooling layers,fully connected layers,and Softmax are applied to complete the recognition PQDs. Simulation experiments demonstrate that the CaT model effectively recognizes PQDs with fewer parameters and floating point operations,achieving high accuracy and strong noise robustness. Its lightweight,end-to-end design also results in shorter inference times compared to other deep learning models.

Abstract:

Power quality control based on grid-connected converter can effectively utilize the surplus capacity of the converter and improves the benefits. However,existing methods tend to only consider grid-following (GFL) control and fail to fully utilize the control potential of multiple converters. In this paper,a cooperative control strategy for harmonic compensation and voltage drop suppression is proposed based on a parallel system of GFL and grid-forming (GFM) converters. Firstly,the basic control principle and mathematical model of grid-connected converter are introduced. Secondly,to compensate the harmonics at the point of common coupling (PCC),a harmonic sub-compensation method is proposed. This approach can improve the flexibility of harmonic compensation and reasonably apportion harmonic between grid-connected converter of different capacities. To support the PCC voltage,the parallel system can operate in three conditions:normal operation condition,condition of GFL support voltage independently and co-supported operating condition. Through compensation capacity calculation and reactive power allocation,the voltage at the PCC can maintain near the rated voltage. Finally,the feasibility and superiority of the proposed strategy are verified through simulations.

Abstract:

Dual power supply cascaded-type power electronic transformers (DPSC-PET) is connected to two power supplies with high operational reliability and flexible operation mode,and it can be widely used in medium and low voltage distribution networks. It is of great importance to conduct in-depth research on its voltage sag tolerance and regulation method for maintaining efficient energy transmission of DPSC-PET as well as high quality power supply during voltage sag. Firstly,the topological structure and control strategy of DPSC-PET are analyzed. Secondly,the influence factors of voltage sag tolerance of DPSC-PET are analyzed for three-phase symmetrical voltage sag which has the most serious power shortage. Then,from the perspective of power balance,a real-time analysis method of voltage sag immunity of DPSC-PET as well as power coordination method between dual input ports when voltage sag occurs on different power supply are proposed to achieve a perfect recovery of low-voltage DC bus voltage,which means the significant improvement of DPSC-PET to cope with transient disturbances. Finally,the simulation model of DPSC-PET is established and the simulations for voltage sag occurred with different magnitudes on different power supply are cited out. The results show that the voltage sag tolerance of DPSC-PET is significantly improved with the proposed method.

Abstract:

Because of the complex control strategy and weak feed characteristics of soft normal open points (SNOP),traditional fault location methods of distribution networks are no longer applicable to flexible distribution network (FDN). A method for fault location in FDN using waveform similarity of positive-sequence current component is proposed. Firstly,considering the typical control strategy of SNOP,the short-circuit fault characteristics of FDN are analyzed. Secondly,Tanimoto coefficients of positive sequence current component at different locations are calculated. By comparing the waveform similarity at different locations of positive sequence current component,a short circuit fault location criterion for FDN is constructed. The Teager energy operator (TEO) is used to accurately calibrate the fault time,and smart terminal units (STU) are used to transmit information. Finally,the proposed method is analyzed and verified through modeling and simulation. The results show that the proposed method can accurately locate the fault section,and has the ability to resist the influence factors such as fault location,fault type,transition resistance,sampling frequency and communication delay,which verifies the feasibility and effectiveness of the proposed method.

Abstract:

Against the dual background of frequent occurrence of typhoon disaster and increasing penetration of new energy generation,a rolling optimal scheduling strategy of multiple flexibility resources including scaled electric vehicle (EV) is proposed to enhance the resilience of the distribution networks under typhoon disaster. Firstly,the fault scenarios of lines and photovoltaic (PV) in the region are simulated by Monte Carlo method based on typical meteorological characteristics,and the typical scenarios are screened by using system information entropy to get the temporal fault states of lines and PV. Secondly,a multiple flexibility resources optimization regulated model is established with the objective of minimizing the weighted load loss rate. Based on the spatio-temporal characteristics of EVs,they are expropriated and regulated,and the network is reconfigured and coordinated with mobile emergency generators (MEG) to maximize the use of resources in the network. Finally,in order to adapt to the evolution of the system fault state and adjust the regulated scheme in real time,a two-stage rolling solution method is proposed to reduce the problem solving complexity. A simulation analysis is conducted on an improved actual power supply unit network in a certain area of Jiangsu province,and the results show that the proposed strategy can effectively reduce load losses and enhance the resilience of the distribution networks in extreme scenarios.

Abstract:

In recent years,a fractal model that can reflect the tortuosity and dispersion of lightning has been gradually used in the research of lightning shielding performance for overhead power lines,based on electrical geometric models and pioneering development models. In this paper,a 1:40 scaled experimental platform based on the equivalence between simulation experiments and natural lightning is constructed,with the ZB6T type tower commonly used in 500 kV transmission lines as the research object. A negative polarity impulse wave of 160/2 500 μs is applied to investigate the relationship between flashover probability,spatial distribution,and protection angle. Based on the WZ model and finite difference method,a lightning fractal model is established,and the characteristics of line flashover are simulated and analyzed under the configuration of the scaled experimental gap. Both the simulation experiments and the simulation results indicate that the points with high flashover probability are concentrated near the transmission line,and the overall flashover space presents a parabolic shape,conforming to the flashover law. As the protection angle decreases,the flashover probability decreases correspondingly,and the flashover space slightly decreases. The simulation results verify the reliability of the simulation experiments and the correctness of the model. Comparing the simulation results with the experimental data,the model shows a high level of reliability.

Abstract:

The online estimation of the state of health (SOH) is an essential part of a lithium battery management system. Most data-driven lithium battery SOH estimation methods are computationally intensive and difficult to use in real-time in battery management system microcontrollers. Therefore,a rapid estimation method of lithium battery SOH based on novel health feature is proposed in this paper. The charging data of the battery is firstly analyzed in the method,and based on the existing health characteristics of time interval of an equal charging voltage difference (TIECVD) in the constant current charging process of the battery,constructs a new health feature,that is,the health feature of charging voltage at the same starting point and charging time interval. Then, a fast estimation method of lithium battery SOH based on the novel health feature and multiple linear regression (MLR) is proposed. Next,by analyzing the oxford battery aging dataset and the random usage dataset of lithium ion batteries used by NASA,the method traverses the constant current charging voltage range in steps of 0.01 V and determines the optimal starting voltage of the lithium battery by maximizing the Pearson correlation coefficient. Finally,considering different time intervals,the method uses the ordinary least squares (OLS) regression analysis method to determine the optimal time interval parameter of the lithium battery. The training set divided by two datasets is used to establish a multiple linear regression model,and the validation set divided by two datasets is used to verify the method. The experimental results show that the proposed method and novel health feature can greatly reduce the calculation volume,and can achieve fast estimation of lithium battery SOH while ensuring prediction accuracy.

Abstract:

Temperature monitoring of converter valves can aid in fault prediction,and it is one of the key technologies to ensure the safe and stable operation of converter stations. However,there are few suitable solutions to the problem of reliable energy supply. In this paper,an accurate output power characteristic model based on magnetic field induction energy harvesting is proposed,utilizing a superimposed magnetic flux calculation method. The impact of key core parameters on energy harvesting power is explored step by step,and an optimization method is proposed. To address the space constraints of the converter valve hall,a lightweight design for the magnetic core is proposed. On this basis,the energy harvesting characteristics of the chip-based energy harvesting device under deflection condition,influenced by the application environment,are studied. An optimization analysis method of energy taking power characteristics under the non-optimal position is proposed. The chip-based energy harvesting device designed in this paper can drive the sensor stably across a wide range of current fluctuations and realize long-distance wireless communication. At the same time,it can meet the needs of long-term maintenance-free detection of power equipment in narrow and high temperature space such as converter stations.

Abstract:

In response to challenges posed by existing transient stability feature selection methods,which often encounter limitations in searching for the optimum combination of critical features and lack an objective criterion for determining the optimal number of key features,this paper introduces a novel approach. A transient power angle stability key feature selection method that seamlessly integrates multi-head self-attention (MHSA) and the Boruta algorithm. A deep neural network (DNN) with an MHSA model is initially constructed to execute transient stability assessments directly on the input grid features. The model dynamically adjusts attention weights during training,focusing on key features. Subsequently,the Boruta algorithm is employed to determine the number of key features. It generates a combination of real and virtual features,which the MHSA model trains to select the actual features that are higher than the maximum virtual feature weight,and the model autonomously determines the optimal number of key features. An analysis is conducted on the IEEE 39-node and 118-node systems to validate the proposed method. The results demonstrate that this approach ensures evaluation accuracy while significantly reducing the number of input features. Moreover,the key features identified exhibit higher evaluation accuracy than traditional methods.

Abstract:

Hybrid cascaded ultra high voltage direct current (UHVDC) integrates the advantages of line commutated converters (LCC) and voltage source converters (VSC). However,there is surplus power on the direct current (DC) side during alternating current (AC) faults at the receiving end,which may cause overvoltage issues in VSC sub-modules. Based on the topological features of the hybrid cascaded UHVDC system,a multi-level adaptive coordinated control strategy is proposed,utilizing operation modes and electrical signals. Multi-level control strategies,namely controlling the negative sequence voltage of the faulty VSC converter,adopting the auxiliary DC voltage control of the sound VSC converters,switching on the controllable energy dissipation devices at the receiving end and retarding of the LCC converters at the sending end,adjusting the DC power of the converters at the sending and receiving ends,are adaptively coordinated within converters,between converters and between stations to effectively mitigate the overvoltage issues in VSC sub-modules and realize AC fault ride-through of the hybrid cascaded UHVDC system. A closed-loop real-time digital simulation test platform is built based on actual control and protection devices and the topology used on site. The test results show the validity of the proposed strategy. The proposed strategy has been successfully applied in the Baihetan-Jiangsu hybrid UHVDC project.

Abstract:

The escalating prevalence of aging distributed photovoltaic (PV) systems has given rise to challenges such as voltage fluctuations and curtailed energy production in large-scale grid-connected systems lacking energy storage. These issues can be effectively mitigated through the strategic deployment of energy storage solutions. In order to enhance the cost-effectiveness of energy storage configuration,an economic optimization method for distribution network energy storage that takes into account the attenuation of aging photovoltaic output is proposed. The impact of aging PV generation decay rate and operational lifespan on actual PV output is thoroughly analyzed,leading to the development of a mathematical relationship between PV output and rated energy storage power and capacity. Building upon this analysis,an optimization model for energy storage configuration is established with a focus on minimizing life cycle costs and reducing voltage deviations at distribution network nodes. This comprehensive approach considers factors such as construction operating costs,network losses,and peak-valley arbitrage income. Finally,particle swarm optimization is employed to solve this complex problem using the IEEE 33-node system. The results demonstrate that the proposed method can achieve significant reductions in both operating cost (33.91%) and life cycle cost (6.01%),thereby validating its economic effectiveness.

Abstract:

The intermittency and randomness of photovoltaic power present different characteristics due to seasonal variations,so it is important to consider seasonal characteristics to improve the accuracy of photovoltaic power prediction. Therefore,a short-term photovoltaic power prediction combination model considering seasonal characteristic and data window is proposed in the paper. Firstly,the Pearson correlation coefficient method is adopted to determine suitable meteorological factors with high contribution to photovoltaic power and reduce the input feature dimensions of the prediction model. Secondly,the prediction error of different photovoltaic power models is compared,and the two models with the lowest photovoltaic power prediction error and the lowest correlation are selected to construct the combination model,i.e.,gated recurrent unit (GRU) model and extreme gradient boosting (XGboost) model. Thirdly,the effects of different input windows in the historical meteorological data on the prediction accuracy of GRU-XGboost model are analyzed to determine the optimal data window. Finally,on this basis,GRU and XGboost predict the photovoltaic power respectively. The final prediction is obtained by weighted combination of the two predictions. Simulation results show that the proposed model has stronger adaptability and higher prediction accuracy than other models.

Abstract:

Virtual power plants (VPP) efficiently aggregate small-capacity and large-volume distributed energy resources through advanced control technologies to participate in electricity market transactions. With the increase in the number of distributed energy sources,the volatility of their power output and the problem of their returns after aggregation still need to be solved. Based on this,a cooperative game scheduling model is proposed for multi-type distributed energy sources aggregated in a virtual power plant under the day-ahead power market. Firstly,the operation framework of multi-type distributed energy aggregation in virtual power plant is proposed. Then,a combined prediction model based on variational modal decomposition (VMD) and improved bidirectional multi gated long short-term memory (Bi-MGLSTM) network is established because the uncertainty of wind power output seriously affects the operation of the system. Secondly,the same type of distributed energy sources form alliances and aim to maximize the revenue from power sales,and construct a cooperative game scheduling model for multiple alliances of virtual power plants. In order to realize the fairness of revenue distribution among alliances and members,a multifactor improvement shapley value method and a two-stage refinement of the revenue distribution scheme based on the parity cycle kernel method are designed. Finally,the example results show that the proposed method effectively improves the prediction accuracy of wind power,realizes the cooperative and complementary operation among alliances within the virtual power plant,and ensures the fairness and reasonableness of the revenue distribution among multiple subjects.

Abstract:

With the large-scale application of electrochemical energy storage,the safety of energy storage prefabricated cabin has become increasingly prominent. The study of the differences in energy storage prefabricated cabin fires under different thermal runaway positions in the energy storage prefabricated cabin can help to design more effective monitoring and fire extinguishing systems,and to improve the safety performance of the energy storage prefabricated cabin. Therefore,it is of great significance to simulate and study the change rule of smoke dispersion and temperature evolution in energy storage prefabricated cabin under different thermal runaway positions. Based on direct numerical simulation and vortex simulation,a numerical model of lithium iron phosphate energy storage prefabricated cabin based on the size of the actual energy storage prefabricated cabin is studied,and the t² model which is more in line with the development of the fire situation compared with the stable fire model,and investigated the change rule of the smoke dispersion and temperature inside the prefabricated cabin by simulating the fire situation with different thermal runaway positions. The simulation results demonstrate that smoke displays distinct dynamic behaviors depending on the thermal runaway positions within the energy storage prefabricated cabin. When thermal runaway positions occur closer to the bottom,smoke exhibits swifter movement and the cabin fills up in a shorter time. Additionally,as the thermal runaway positions above 1.85 meters approach the top,accompanied by significant temperature fluctuations. Notably,there is a noticeable amplification in the temperature disparity along the horizontal axis of the energy storage prefabricated cabin. Moreover,the design scheme of the fire extinguishing system for a standard energy storage prefabricated cabin with a rated capacity of 1.2 MW·h is analyzed. This perfluorohexane fire extinguishing system,with a sprinkler intensity of 20 L/(min·m²), a sprinkler angle of 120ånd a particle size of 50 μm,can successfully controls fires and mitigates fire damage. The research results in the article can provide theoretical guidance for the distributed deployment strategy in energy storage prefabricated cabin and fire safety design of monitoring and warning devices.

Abstract:

Given the complexity of the unified power quality conditioner (UPQC) system and the difficulty in its control,a single control strategy is no longer sufficient to address various faults in the power grid system. Consequently,a hybrid control strategy combining linear active disturbance rejection control (LADRC) and model predictive control (MPC) is employed in this paper. The LADRC strategy is used in the voltage outer loop control to enhance the system's rapid response and disturbance rejection capabilities,providing a accurate reference current signal for the current inner loop. In the current inner loop,the current MPC strategy further improves tracking performance and system robustness. The partitioning of the space voltage vector in model prediction is optimized to reduce the controller's computational load,thereby improving computation speed while ensuring the quality of the output current. Finally,the system is modeled and simulated on the MATLAB/Simulink simulation platform. The results validate that the LADRC-MPC control strategy effectively compensates for comprehensive power quality issues such as grid voltage sags/surges,current distortion,and harmonic pollution caused by load imbalance. It also enhances the voltage support capability of the grid.

Abstract:

Non-intrusive load monitoring (NILM) of residential houses is an important research content of the user demand side of smart grids,and the energy consumption analysis and power consumption management of residential loads are key steps in achieving energy conservation,emission reduction,and sustainable development. Aiming at the problems of poor recognition performance of traditional algorithms and difficulty in adapting to the current complex electricity environment,a NILM load recognition method integrating convolutional neural network (CNN)-self-attention mechanism is proposed from the optimization idea of enhancing the feature extraction performance of classification algorithms. Firstly,the power data of eight different household appliances are collected to establish a U-I trajectory curve database. Secondly,the feature aggregation ability of CNN is improved by using squeeze-and-excitation network (SENet) attention mechanism to complete the feature extraction and load identification of U-I trajectory curves of different electrical appliances. Finally,the private dataset and PLAID dataset are tested,and the example results show that the proposed method has high recognition accuracy and good generalization performance in different operational scenarios.