Wind-solar power generation and electric vehicle load cooperative control method based on user behaviors

Abstract

The invention discloses a wind-solar power generation and electric vehicle load cooperative control method based on user behaviors, and relates to the technical field of electric power system automation, and the method comprises the steps: obtaining historical power generation and real-time power generation data of a wind-solar power generation system according to an electric vehicle power load demand curve in a future time period, predicting a generating capacity curve of the wind-solar power generation system in the time period; performing matching analysis on the electric load demand curve of the electric vehicle and the wind-solar power generation curve to obtain a matching difference degree, and triggering cooperative control when the matching difference degree is greater than a preset threshold value; in the cooperative control, the generated power of the wind-solar power generation system and the actual power load of the electric vehicle are monitored in real time, the matching difference degree is calculated, and the generated power of the wind-solar power generation system is dynamically adjusted according to the change trend of the matching difference degree. The running cost of the electric vehicle is reduced, the utilization efficiency of wind and light power generation is improved, the energy utilization efficiency and the renewable energy permeability of the rail transit system are improved, and the purposes of energy conservation, emission reduction and green development are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system automation, and in particular to a method for coordinated control of wind and solar power generation and tram load based on user behavior. Background Art

[0002] Wind and solar power generation is intermittent and volatile, while tram load is also uncertain and random, resulting in a power mismatch between wind and solar power generation and tram load, affecting the utilization efficiency of wind and solar power generation energy and the operating economy of the tram system.

[0003] In order to solve this problem, the traditional control method mainly uses power electronic equipment to smoothly control the wind and solar power generation, or balances the power difference between wind and solar power generation and tram load by optimizing scheduling and energy storage systems. For example, Chinese patent CN105305659A discloses a method for smooth control of renewable energy power generation in a power system. By predicting wind and solar power generation, combined with the operating status of the power system, power electronic equipment is used to smoothly control renewable energy power generation, thereby reducing the impact of the fluctuation of renewable energy power generation on the power system. For another example, Chinese patent CN107394824A discloses a method for optimizing the scheduling of wind and solar power generation in a microgrid based on energy storage. By coordinating wind and solar power generation, energy storage systems and load demand, the operation scheduling of the microgrid is optimized, and the utilization efficiency of wind and solar power generation energy is improved.

[0004] However, the above traditional control methods do not fully consider the uncertainty and randomness of tram loads, especially the impact of user driving behavior on tram loads, resulting in limited control effects. In addition, traditional control methods lack real-time and adaptability, making it difficult to cope with the dynamic changes in wind and solar power generation and tram loads, affecting the control performance of the system. Summary of the invention

[0005] In order to solve the above problems, the present invention provides a method for coordinated control of wind and solar power generation and electric vehicle load based on user behavior, comprising the following steps:

[0006] Based on the historical and real-time driving data of trams, the demand curve of tram electricity load in the future is predicted;

[0007] Based on the historical power generation data and real-time power generation data of the wind and solar power generation system, the wind and solar power generation curve in the future period is predicted;

[0008] According to the matching difference between the tram power load demand curve and the wind and solar power generation curve, and based on the distribution characteristics of the tram power load demand, the tram power load demand within the forecast time period is classified, and based on the classification results, the wind and solar power generation and tram load are coordinated and controlled.

[0009] Preferably, in the process of acquiring historical driving data and real-time driving data, the historical driving data includes the tram's GPS trajectory data, accelerator pedal opening data, brake pedal opening data and battery pack charge state data, and the real-time driving data includes the tram's current GPS position, speed, accelerator pedal opening, brake pedal opening and battery pack charge state parameter data.

[0010] Preferably, in the process of obtaining the electric load demand curve of the tram, the historical driving data and the real-time driving data are preprocessed, and the preprocessed historical driving data are input into the long short-term memory neural network in the form of a time series. The input features are mapped to the vehicle speed, acceleration and battery pack power as the output of the network through the vehicle dynamics model and the energy consumption model, and the encoder-decoder network architecture is adopted. The network weights are trained by the cross entropy loss function and the adaptive moment estimation optimization algorithm to obtain the driving behavior model; which is used to predict the electric load demand curve of the tram in the future; wherein the preprocessing process includes: removing outliers and invalid values ​​through data cleaning, aligning the data collected by different sensors on the time axis through data alignment, and uniformly scaling the data of different physical dimensions to the same numerical range through data normalization.

[0011] Preferably, in the process of obtaining the electric load demand curve of the tram, based on the driving behavior model and according to the tram driving data collected in real time, the speed, acceleration and energy consumption change trends of the tram in the future period are predicted, and the predicted speed and acceleration changes are converted into power demand changes of the battery pack through the energy consumption model and the battery model, so as to obtain the electric load demand curve of the tram in the predicted time period; when predicting the electric load, the weather conditions and road conditions information at the current location of the tram are introduced as auxiliary inputs, and the weather conditions including temperature, humidity and wind speed are obtained, and the road conditions information including road type, traffic flow and speed limit are obtained, and the weather conditions and road conditions information are encoded and input into the driving behavior model, so that the driving behavior model can adapt to different external conditions and dynamically adjust the electric load demand curve of the tram.

[0012] Preferably, in the process of obtaining the wind and solar power generation curve, a combined model of a convolutional neural network and a long short-term memory neural network is used to train the model through historical power generation data, the convolutional neural network is used to extract the spatial characteristics of meteorological parameters, and the long short-term memory neural network is used to extract the time series characteristics of the generator set state, and a wind and solar power generation prediction model is constructed to predict the wind and solar power generation curve in the future.

[0013] Preferably, in the process of constructing a wind and solar power generation prediction model, the meteorological feature vector extracted by the convolutional neural network and the generator set state feature vector extracted by the long short-term memory neural network are spliced ​​into a long vector, and then the spliced ​​long vector is mapped to the predicted value of the wind and solar power generation power through one or more fully connected layers, and a loss function is used to measure the difference between the predicted value of the wind and solar power generation power and the actual power generation power; the wind and solar power generation prediction model is optimized in an end-to-end manner, the back propagation algorithm and the gradient descent optimization algorithm are used to update the wind and solar power generation prediction model parameters, and the wind and solar power generation prediction model is optimized through data enhancement, regularization and early stopping strategies.

[0014] Preferably, in the process of obtaining the matching difference, the matching difference is obtained according to the average absolute deviation of the two curves within a given time range.

[0015] Preferably, before the collaborative control process is performed, the collaborative control process is triggered according to the matching difference, including the following steps:

[0016] Considering the tram operating conditions, wind and solar power generation forecast results and energy storage system status factors, a multi-objective optimization model is established with minimizing the matching difference and tram energy consumption as the optimization objectives, and tram operating constraints and energy storage system constraints as constraints.

[0017] Solve the multi-objective optimization model to obtain a set of Pareto optimal solutions, from which the optimal solution of balanced matching difference and tram energy consumption is selected, and its corresponding tram operation parameters and energy management strategy are used as the collaborative control strategy;

[0018] The formulated coordinated control strategy is sent to the tram control system and energy storage management system. By coordinating and controlling the tram's traction system, braking system and power converter of the energy storage system, the tram's operating parameters and energy management strategy are optimized and adjusted, reducing the imbalance between the tram's power load demand and wind and solar power generation.

[0019] According to the real-time changes in the tram power load demand and wind and solar power generation as well as the feedback information on the collaborative control effect, the collaborative control strategy is updated and optimized in real time to ensure the operation of the tram system.

[0020] Preferably, in the process of constructing the multi-objective optimization model, the speed curve of the tram, the traction force curve and the charging and discharging power curve of the energy storage system are selected as optimization variables of the multi-objective optimization model, the power demand of the tram power load is calculated through the power transfer model of the tram traction system, and the wind and solar power generation power is predicted through the wind and solar power generation prediction model;

[0021] Construct a matching difference objective function, expressed as the average absolute deviation between the electric load demand power of the tram and the wind and solar power generation power within a given time range, and construct a tram energy consumption objective function, expressed as the integral of the electric load demand power of the tram within a given time range; set tram operation constraints, including speed constraints, acceleration constraints, and traction constraints; set energy storage system constraints, including state of charge constraints and charge and discharge power constraints, where the state of charge is calculated by the initial state of charge and the charge and discharge power of the energy storage system;

[0022] The matching difference objective function, tram energy consumption objective function, tram operation constraints and energy storage system constraints are combined to form a multi-objective optimization model.

[0023] Preferably, in the process of performing coordinated control, according to the changing trend of the matching difference between the electric vehicle power load demand curve and the wind-solar power generation curve, the power generation power of the wind-solar power generation system is dynamically adjusted based on the proportional-integral-differential controller, the wind-solar power generation power adjustment amount output by the proportional-integral-differential controller is superimposed on the original wind-solar power generation power, and the adjusted wind-solar power generation power is realized by controlling the power converter of the wind-solar power generation system;

[0024] The actual tram power load data is fed back into the driving behavior model for online learning. The prediction error is calculated by comparing the actual measured tram power load with the tram power load predicted by the driving behavior model, and the weights and biases of the driving behavior model are updated through the back-propagation algorithm.

[0025] Based on the driving behavior model predictive control, the collaborative control strategy is dynamically optimized. At the current moment, the predicted value of wind and solar power generation and the predicted value of tram power load in the future period are obtained. Taking the current state as the initial condition, the matching difference degree and tram energy consumption in the future prediction time domain as the optimization target, and the tram operation constraints and energy storage system constraints as the constraints, the optimal control sequence in the future prediction time domain is obtained. The first element of the optimal control sequence is used as the control quantity at the current moment and implemented in the tram control system and energy storage management system. The above steps are repeated at the next moment to achieve rolling optimization.

[0026] The present invention also discloses a wind-solar power generation and tram load collaborative control system based on user behavior, which is used to implement the above collaborative control method, including:

[0027] A first prediction module (first unit), used for predicting the electric load demand curve of the tram in a future period of time based on the historical driving data and real-time driving data of the tram;

[0028] The second prediction module (the second unit) is used to predict the wind and solar power generation curve in the future based on the historical power generation data and real-time power generation data of the wind and solar power generation system;

[0029] The collaborative control module (the third unit) is used to classify the tram electricity load demand within the forecast time period according to the matching difference between the tram electricity load demand curve and the wind and solar power generation curve and the distribution characteristics of the tram electricity load demand, and to collaboratively control the wind and solar power generation and the tram load according to the classification results.

[0030] The present invention discloses the following technical effects:

[0031] The present invention introduces a dynamic optimization strategy based on model predictive control to solve the optimal control problem of wind and solar power generation and tram operation in the rolling time domain, comprehensively considers multiple optimization objectives such as energy matching, tram energy consumption, and operation quality, and generates a collaborative control strategy that takes into account both economy and reliability, thereby reducing tram operating costs and improving the utilization efficiency of wind and solar power generation.

[0032] The present invention is based on real-time monitoring data and online learning, and is adaptive to changes in tram operating conditions and wind and solar resource conditions. It can be widely used in wind and solar power generation and tram load coordinated control scenarios in different environments, improving the energy utilization efficiency and renewable energy penetration rate of rail transit systems, and achieving the goals of energy conservation, emission reduction and green development.

[0033] The present invention intelligently optimizes the coordinated control process of wind and solar power generation systems and tram loads, maximizes the use of renewable energy, and adopts advanced control technologies such as model prediction and online learning. Compared with manual scheduling or simple feedback control, it greatly improves the absorption capacity and power utilization rate of wind and solar power generation, reduces power transmission losses, and reduces the peak and frequency regulation pressure of the power grid. While saving energy costs, it ensures the safety and reliability of tram operation, and has significant economic and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0036] Figure 2 It is a schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0037] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0038] like Figure 1 As shown, the present invention provides a method for coordinated control of wind and solar power generation and electric vehicle load based on user behavior, which specifically includes the following contents:

[0039] S1. Obtain the historical driving data and real-time driving data of the tram, train the historical driving data based on the machine learning algorithm to obtain the user's driving behavior model, use the driving behavior model to analyze the real-time driving data, and obtain the tram power load demand curve in the future period; obtain the historical power generation data and real-time power generation data of the wind and solar power generation system, the wind and solar power generation system includes a wind power generation system and a photovoltaic power generation system, the historical power generation data includes historical meteorological parameters and generator set status parameters, and the real-time power generation data includes meteorological parameters and generator set status parameters measured in real time; use a deep learning algorithm to train the historical power generation data to obtain a wind and solar power generation prediction model, the deep learning algorithm is a combination model of a convolutional neural network and a long short-term memory neural network, the spatial features of meteorological parameters are extracted by the convolutional neural network, and the time series features of the generator set status are extracted by the long short-term memory neural network; use the wind and solar power generation prediction model to analyze the real-time power generation data, and combine it with numerical weather forecast data to predict the power generation curve of the wind and solar power generation system in the future period;

[0040] S2. Match and analyze the electric vehicle load demand curve and the wind and solar power generation curve to obtain the matching difference between the two. When the matching difference is greater than the preset threshold, the collaborative control process is triggered. In the collaborative control process, according to the distribution characteristics of the electric vehicle load demand, the electric vehicle load demand in the forecast period is divided into peak demand, average demand and valley demand. When the electric vehicle load is at the peak demand, the wind and solar power generation system is controlled to increase the power generation, and the grid power is started to supplement when necessary. When the electric vehicle load is at the average demand, the wind and solar power generation system is controlled to output power steadily according to the predicted power generation. When the electric vehicle load is at the valley demand, the wind and solar power generation system is controlled to reduce the power generation and store the surplus power in the electric vehicle battery pack or the grid.

[0041] S3. During the coordinated control process, the power generation of the wind and solar power generation system and the actual power load of the tram are monitored in real time, the matching difference is calculated, and the power generation of the wind and solar power generation system is dynamically adjusted according to the changing trend of the matching difference. At the same time, the actual power load data of the tram is fed back to the user's driving behavior model for online learning, the driving behavior model is updated, and the prediction accuracy is improved, thereby realizing precise coordinated control of wind and solar power generation and tram load.

[0042] In an optional embodiment,

[0043] The steps of obtaining the historical driving data and the real-time driving data of the tram, training the historical driving data based on the machine learning algorithm to obtain the user's driving behavior model, analyzing the real-time driving data using the driving behavior model, and obtaining the tram power load demand curve in the future period include:

[0044] The historical driving data and real-time driving data of the tram are collected by sensors and data recorders installed on the tram, wherein the historical driving data includes GPS track data, accelerator pedal opening data, brake pedal opening data and battery pack charge state data of the tram, and the real-time driving data includes the current GPS position, speed, accelerator pedal opening, brake pedal opening and battery pack charge state parameter data of the tram; the collected historical driving data and real-time driving data are preprocessed, abnormal values ​​and invalid values ​​are removed by data cleaning, data collected by different sensors are aligned on the time axis by data alignment, and data of different physical dimensions are uniformly scaled to the same numerical range by data normalization;

[0045] The preprocessed historical driving data is input into the long short-term memory neural network in the form of time series. The input features are mapped into vehicle speed, acceleration and battery pack power as the network output through the vehicle dynamics model and energy consumption model. The encoder-decoder network architecture is used to train the network weights through the cross entropy loss function and adaptive moment estimation optimization algorithm to obtain the driving behavior model.

[0046] The tram driving data collected in real time is input into the trained driving behavior model. The driving behavior model predicts the speed, acceleration and energy consumption trend of the tram in the future period based on the current driving status and historical driving data. The predicted speed and acceleration changes are converted into power demand changes of the battery pack through the energy consumption model and the battery model, and the electric load demand curve of the tram in the predicted time period is obtained. When predicting the electric load, the weather conditions and road conditions at the current location of the tram are introduced as auxiliary inputs. The weather conditions including temperature, humidity and wind speed are obtained through the APIs interface, and the road conditions including road type, traffic flow and speed limit are obtained. These environmental factors are encoded and input into the driving behavior model, so that the driving behavior model can adapt to different external conditions and dynamically adjust the electric load demand curve.

[0047] For example, in order to achieve precise coordinated control of wind and solar power generation and tram load, it is necessary to accurately predict the tram load demand in the future. This paper proposes a tram load prediction method based on the user driving behavior model. By obtaining the historical and real-time driving data of the tram, the driving behavior model is trained using a machine learning algorithm, and the future tram load is dynamically predicted in combination with environmental factors.

[0048] First, the historical and real-time driving data of the tram are collected through various sensors and data recorders installed on the tram. The historical driving data includes the GPS trajectory data, accelerator pedal opening data, brake pedal opening data and battery pack state of charge (SoC) data of the tram in the past period of time. The real-time driving data includes the GPS position, speed, accelerator pedal opening, brake pedal opening and battery pack SoC data of the tram at the current moment.

[0049] After data collection is completed, the raw data needs to be preprocessed. First, outliers and invalid values ​​are removed through data cleaning. For example, data with an accelerator pedal opening greater than 100% or less than 0% is considered invalid data and is removed. Secondly, since the data collection frequencies of different sensors may be different, it is necessary to align these data on the time axis through data alignment to ensure that all types of data collected at the same time correspond to the same timestamp. Finally, since the numerical ranges of different physical quantities vary greatly, they need to be uniformly scaled to the interval of [0,1] through data normalization to improve the comparability of the data and the convergence speed of the model.

[0050] After data preprocessing is completed, the historical driving data is input into the long short-term memory (LSTM) neural network in the form of time series for training to obtain the driving behavior model. The LSTM network can effectively capture the long-term dependencies in time series data and is suitable for modeling time series patterns such as driving behavior.

[0051] Specifically, the historical driving data is divided into fixed time windows (such as 10 seconds), and the data in each time window constitutes a sample. For each sample, the GPS position, speed, accelerator pedal opening, brake pedal opening, and battery pack SoC in the first 9 seconds are used as input features, and the speed, acceleration, and battery pack power in the 10th second are used as labels. Through the vehicle dynamics model and energy consumption model, the speed and acceleration can be mapped to the battery pack power. The label acquisition process is as follows:

[0052] Speed: directly obtained from the driving data at the 10th second;

[0053] Acceleration: calculated by dividing the speed change between the 9th and 10th seconds by the time interval (1 second);

[0054] Battery pack power: Based on the speed and acceleration at the 10th second, the traction is calculated through the vehicle dynamics model, and the battery pack output power is estimated based on the energy consumption model.

[0055] After the sample is constructed, the LSTM network is trained using an encoder-decoder network architecture. The encoder encodes the input features and extracts the driving behavior pattern. The decoder predicts the speed, acceleration, and battery pack power at future moments based on the encoder output. The network weights are updated using the adaptive moment estimation (Adam) optimization algorithm by minimizing the cross entropy loss function between the predicted value and the true value, and the training is iterated until the model converges.

[0056] Assume that 1000 training samples are constructed, each sample contains 10 time steps, each time step contains 5 input features and 3 output labels. The input features are normalized GPS position longitude, GPS position latitude, speed, accelerator pedal opening, brake pedal opening and battery pack SoC, and the output labels are normalized speed, acceleration and battery pack power. The samples are divided into training set and validation set in a ratio of 8:2, the LSTM hidden layer dimension is set to 64, the number of layers of the encoder and decoder are both 2, the learning rate is 0.001, and the batch size is 32. After training 50 epochs, the mean absolute error (MAE) of the model on the validation set is 0.05, and the root mean square error (RMSE) is 0.08, indicating that the driving behavior model has achieved good training results.

[0057] Using the trained driving behavior model, the electric load of the tram in the future can be dynamically predicted. The specific steps are as follows:

[0058] First, the real-time collected tram driving data is input into the driving behavior model, including the current GPS position, speed, accelerator pedal opening, brake pedal opening and battery pack SoC.

[0059] Secondly, the driving behavior model predicts the speed, acceleration and battery pack power change trends of the tram in the future (such as the next 10 minutes) based on the input real-time driving data and historical driving data. Through the vehicle dynamics model and energy consumption model, the predicted speed and acceleration changes are converted into battery pack power demand changes to obtain the electric load demand curve of the tram in the predicted time period.

[0060] Finally, in order to improve the adaptability of power load forecasting, the weather and road conditions at the current location of the tram are introduced as auxiliary inputs in the forecasting process. By calling the weather API interface, the temperature, humidity, wind speed and other weather data of the current location are obtained; by calling the map API interface, the road type, traffic flow, speed limit and other road conditions data of the current location are obtained. These environmental factors are encoded and input into the driving behavior model, so that it can adapt to different external conditions and dynamically adjust the power load demand forecast results.

[0061] For example, assuming the current time is 09:00:00, and the tram is located at Zhongguancun Street, Haidian District, Beijing, the driving data at this time is:

[0062] GPS location: (39.9805,116.3296)

[0063] Speed: 56.7km / h (normalized to 0.189)

[0064] Accelerator pedal opening: 60% (normalized to 0.600)

[0065] Brake pedal opening: 0% (normalized to 0.000)

[0066] Battery pack SoC: 78% (normalized to 0.780)

[0067] The weather API is called to obtain the current temperature of 25°C (normalized to 0.417), the humidity is 60% (normalized to 0.600), and the wind force is level 3 (normalized to 0.125); the map API is called to obtain the current road type as a city trunk road (coded as 1), the traffic flow is smooth (coded as 0), and the speed limit is 80km / h (normalized to 0.667).

[0068] The above real-time driving data and environmental factors are input into the driving behavior model to predict the speed, acceleration and battery pack power change trends of the tram in the next 10 minutes.

[0069] Through the above method, dynamic prediction of tram power load can be achieved, providing important decision-making basis for subsequent coordinated control of wind and solar power generation.

[0070] In an optional embodiment,

[0071] The preprocessed historical driving data is input into the long short-term memory neural network in the form of time series. The input features are mapped into vehicle speed, acceleration and battery pack power as the network output through the vehicle dynamics model and energy consumption model. The encoder-decoder network architecture is used to train the network weights through the cross entropy loss function and adaptive moment estimation optimization algorithm. The steps to obtain the driving behavior model include:

[0072] The LSTM network adopts an encoder-decoder architecture. The encoder part is composed of several LSTM network units, which is responsible for encoding the input historical driving data sequence into a feature vector of fixed length. The decoder part is also composed of several LSTM network units, which decodes the driving behavior sequence in the future period according to the feature vector generated by the encoder.

[0073] Through the vehicle dynamics model and energy consumption model, the future speed component output by the long short-term memory network is mapped into vehicle speed, acceleration and battery pack power. Based on the future vehicle speed component output by the long short-term memory network, the vehicle speed and acceleration are calculated. Based on the longitudinal dynamics equation of the vehicle, the traction of the tram at the future moment is estimated. It is assumed that the tram uses a single permanent magnet synchronous motor as the drive motor and the motor efficiency is constant, and the output power of the motor is calculated. Furthermore, considering the power electronic conversion efficiency between the battery pack and the motor, the output power of the battery pack is obtained.

[0074] A mean square error loss function is designed to measure the difference between the LSTM network output and the actual driving data. Each physical quantity is normalized when calculating the mean square error loss, and an L2 regularization term is introduced into the loss function to constrain the size of network parameters and avoid overfitting. During the training process of the driving behavior model, an adaptive moment estimation optimization algorithm is used to update the network parameters, and data enhancement, early stopping mechanism and parameter fine-tuning strategies are used to optimize the driving behavior model. Finally, a LSTM network driving behavior model with good generalization performance and strong robustness is obtained.

[0075] For example, after obtaining the preprocessed historical driving data of the tram, it is necessary to use these time series data to train the driving behavior model so that it can accurately predict the electric load demand of the tram in the future. This paper uses the long short-term memory (LSTM) neural network with an encoder-decoder architecture to construct the driving behavior model, and maps the network output to vehicle speed, acceleration and battery pack power through the vehicle dynamics model and energy consumption model. The specific steps of model training are as follows:

[0076] The driving behavior model adopts an LSTM network with an encoder-decoder architecture, in which the encoder part is composed of several LSTM units, which is responsible for encoding the input historical driving data sequence into a feature vector of fixed length; the decoder part is also composed of several LSTM units, which decodes the driving behavior sequence in the future period according to the feature vector generated by the encoder.

[0077] Specifically, the encoder contains 3 layers of LSTM units, and the hidden state dimension of each layer is 128; the decoder contains 2 layers of LSTM units, and the hidden state dimension of each layer is 256. At the last time step of the encoder, its hidden state is used as the initial hidden state of the decoder. At each time step, the decoder generates the output of the current step based on the output and hidden state of the previous step until a complete future driving behavior sequence is generated.

[0078] In order to map the future speed components output by the LSTM network into vehicle speed, acceleration, and battery pack power, the vehicle dynamics model and energy consumption model need to be used for calculation.

[0079] First, based on the future vehicle speed component output by the LSTM network, the vehicle speed and acceleration at the corresponding time can be directly calculated. The vehicle speed can be obtained by integrating the speed component, while the acceleration can be calculated by dividing the speed change between two adjacent time steps by the time interval.

[0080] Secondly, based on the longitudinal dynamics equation of the vehicle, the traction of the tram at the future moment can be estimated. This equation takes into account the influence of factors such as the mass of the tram, the output torque of the motor, the wheel radius, air resistance, rolling resistance and slope on the longitudinal motion of the vehicle. Substituting the future moment speed and acceleration output by the LSTM network into the longitudinal dynamics equation, the traction at the corresponding moment can be solved.

[0081] Again, assuming that the tram uses a single permanent magnet synchronous motor as the drive motor and the motor efficiency is constant, the motor output power can be calculated based on the motor output torque and the tram speed.

[0082] Finally, considering the power electronic conversion efficiency between the battery pack and the motor, the output power of the battery pack can be estimated based on the motor output power.

[0083] At this point, the mapping process from LSTM network output to vehicle speed, acceleration, and battery pack power is completed.

[0084] For example, suppose that the speed component output by the LSTM network at a certain moment is 60km / h, the speed at the previous moment is 58km / h, the time interval is 1 second, and the various parameters of the tram are known, such as mass, motor torque, wheel radius, air resistance coefficient, windward area, air density, rolling resistance coefficient, slope angle, motor efficiency and power electronic conversion efficiency.

[0085] First, the acceleration is calculated based on the velocity change and the time interval.

[0086] Then, the speed, acceleration and other parameters are substituted into the longitudinal dynamics equation to solve the traction force at that moment.

[0087] Next, the motor output power is calculated using the traction and speed.

[0088] Finally, considering the influence of power electronic conversion efficiency, the output power of the battery pack is calculated.

[0089] Therefore, at this moment, the speed component output by the LSTM network corresponds to a vehicle speed of 60 km / h. The acceleration and battery pack output power can also be obtained through the above steps.

[0090] In order to train the LSTM driving behavior model, it is necessary to design a suitable loss function to measure the difference between the network output and the real driving data. This paper uses the mean square error (MSE) as the loss function to calculate the average squared difference between the model output value and the real value.

[0091] When calculating the MSE loss, in order to balance the differences in the numerical ranges of different physical quantities, the vehicle speed, acceleration, and battery pack power are normalized to their maximum and minimum values, respectively, so that their value ranges are scaled to the interval [0,1].

[0092] In addition, in order to avoid overfitting of the model, the L2 regularization term is introduced in the loss function to constrain the weight parameters of the LSTM network. L2 regularization adds a penalty term proportional to the weight size to the loss function, making the model tend to learn smaller weight values ​​and improve the generalization ability of the model.

[0093] Therefore, the final loss function is the weighted sum of the MSE loss and the L2 regularization term. By adjusting the weight coefficients of the two, a balance can be achieved between fitting the training data and avoiding overfitting.

[0094] During the training process of the driving behavior model, a series of optimization strategies are used to improve the performance and generalization ability of the model, including:

[0095] Adaptive Moment Estimation (Adam) Optimization Algorithm: Adam is an adaptive learning rate optimization algorithm that can assign different learning rates to different parameters to accelerate the convergence of the model. Set the initial learning rate of the Adam optimizer and decay the learning rate after each epoch.

[0096] Data enhancement: In order to increase the diversity of training samples, the original historical driving data is enhanced before each epoch. Data enhancement methods include: randomly adding Gaussian noise to the speed sequence, randomly adding bias to the acceleration sequence, randomly changing the starting time point of the sample, etc. Through data enhancement, the generalization performance and robustness of the model can be improved.

[0097] Early stopping mechanism: During model training, if the loss on the validation set does not decrease for a certain number of epochs, the early stopping mechanism is triggered to stop training and save the current optimal model parameters. This can prevent model overfitting and alleviate the performance degradation problem that occurs in the later stages of training.

[0098] Parameter fine-tuning: A batch of new driving data close to the actual operating conditions of the tram is selected as the fine-tuning dataset. Based on the trained driving behavior model, a smaller learning rate is used to fine-tune the model parameters to make it better adapt to the actual application scenarios.

[0099] By comprehensively applying the above optimization strategies, we finally obtain an LSTM driving behavior model with excellent performance and strong generalization ability, which can be used for subsequent tram power load prediction tasks.

[0100] In an optional embodiment,

[0101] The wind and solar power generation prediction model is obtained by training the historical power generation data using a deep learning algorithm. The deep learning algorithm is a combination model of a convolutional neural network and a long short-term memory neural network. The steps of extracting the spatial characteristics of meteorological parameters through the convolutional neural network and extracting the time series characteristics of the generator set status through the long short-term memory neural network include:

[0102] Data preprocessing is performed on the meteorological parameters and generator set status parameters of wind and solar power generation. The maximum and minimum normalization method is used for the meteorological parameters to scale the data to the [0,1] interval. The zero-mean normalization method is used for the generator set status parameters to transform the data into a distribution with a mean of 0 and a variance of 1. The preprocessed meteorological parameters are organized in the form of a two-dimensional matrix, and the preprocessed generator set status parameters are organized in the form of a time series.

[0103] The convolutional neural network is used to extract features from the preprocessed meteorological parameter matrix, and a compressed meteorological feature vector is obtained through a series of convolutional layers and pooling layers. The convolutional layer extracts the local spatial features of the meteorological parameters through convolution operations, and the pooling layer reduces the size of the feature map through downsampling operations.

[0104] The long short-term memory neural network is used to extract the features of the preprocessed generator set state parameter time series, and the compressed generator set state feature vector is obtained through a series of long short-term memory layers. The long short-term memory layer realizes the long-term memory of the generator set state historical information through the gating mechanism and memory unit.

[0105] The meteorological feature vector extracted by the convolutional neural network and the generator state feature vector extracted by the long short-term memory neural network are concatenated into a long vector, and then the concatenated long vector is mapped to the predicted value of wind and solar power generation through one or more fully connected layers, and the loss function is used to measure the difference between the predicted value of wind and solar power generation and the actual power generation;

[0106] The wind and solar power generation prediction model is optimized in an end-to-end manner, and the parameters of the wind and solar power generation prediction model are updated using the back propagation algorithm and gradient descent optimization algorithm. The generalization performance and robustness of the wind and solar power generation prediction model are improved through data enhancement, regularization and early stopping strategies.

[0107] For example, after obtaining the preprocessed historical data of wind and solar power generation, the data is trained using a deep learning algorithm to obtain a wind and solar power generation prediction model. This paper adopts a combined model of a convolutional neural network (CNN) and a long short-term memory (LSTM) neural network, where CNN is used to extract the spatial features of meteorological parameters and LSTM is used to extract the time series features of the generator state. The specific steps of model training are as follows:

[0108] Data preprocessing is performed on meteorological parameters (such as wind speed, wind direction, irradiance, temperature, etc.) and generator set state parameters (such as speed, blade angle, torque, etc.) of wind and solar power generation. For meteorological parameters, the maximum and minimum value normalization method is used to scale the data to the interval [0,1]; for generator set state parameters, the zero mean normalization method is used to transform the data into a distribution with a mean of 0 and a standard deviation of 1. The preprocessed meteorological parameters are organized in the form of a two-dimensional matrix, with each row representing the meteorological state of a time step; the preprocessed generator set state parameters are organized in the form of a time series and arranged in chronological order.

[0109] For example, suppose a wind farm has recorded meteorological parameters and generator status parameters every 10 minutes in the past year, with a total of 52,560 data points. Meteorological parameters include wind speed (0-25m / s), wind direction (0-360°), and temperature (-20-40°C), and generator status parameters include speed (0-20rpm) and blade angle (0-90°).

[0110] For wind speed, the maximum value is 25 m / s and the minimum value is 0 m / s. The normalized wind speed v'=(v-0) / (25-0), that is, v'=v / 25.

[0111] For the wind direction, the maximum value is 360° and the minimum value is 0°. The normalized wind direction d'=(d-0) / (360-0), that is, d'=d / 360.

[0112] For temperature, the maximum value is 40° C. and the minimum value is −20° C. The normalized temperature t′=(t−(−20)) / (40−(−20)), that is, t′=(t+20) / 60.

[0113] For the rotation speed, the mean is 10 rpm, the standard deviation is 5 rpm, and the normalized rotation speed s'=(s-10) / 5.

[0114] For the blade angle, the mean is 45°, the standard deviation is 15°, and the normalized blade angle a'=(a-45) / 15.

[0115] After preprocessing, the meteorological parameter matrix X_weather is obtained, with a shape of (52560,3), and each row is the normalized wind speed, wind direction, and temperature; the generator set state parameter time series X_state has a shape of (52560,2), and each row is the normalized speed and blade angle.

[0116] CNN is used to extract features from the preprocessed meteorological parameter matrix. CNN is composed of multiple convolutional layers and pooling layers stacked alternately, and local spatial features of meteorological parameters are extracted layer by layer through convolution and downsampling operations.

[0117] Specifically, a CNN with 3 convolutional layers and 2 pooling layers is designed. The first convolutional layer contains 16 3x3 convolution kernels, and the activation function is ReLU; the first pooling layer is 2x2 maximum pooling. The second convolutional layer contains 32 3x3 convolution kernels, and the activation function is ReLU; the third convolutional layer contains 64 3x3 convolution kernels, and the activation function is ReLU; the second pooling layer is 2x2 average pooling. The output of CNN is a 256-dimensional meteorological feature vector.

[0118] For example, for the meteorological parameter matrix X_weather with a shape of (52560,3):

[0119] After the first convolutional layer, 16 feature maps (52560, 3, 16) are obtained;

[0120] After the first pooling layer, 16 feature maps (26280, 1, 16) are obtained;

[0121] Through the second convolutional layer, 32 feature maps (26280, 1, 32) are obtained;

[0122] Through the third convolutional layer, 64 feature maps (26280, 1, 64) are obtained;

[0123] After the second pooling layer, 64 feature maps (13140, 1, 64) are obtained;

[0124] The pooled feature maps are flattened and concatenated to obtain a meteorological feature matrix F_weather with a shape of (13140, 256).

[0125] LSTM is used to extract features from the preprocessed generator state parameter time series. LSTM can effectively capture long-term dependencies in time series data through gating mechanisms and memory units.

[0126] Design a recurrent neural network with 2 layers of LSTM, each layer of LSTM contains 128 neurons. LSTM processes the generator state sequence time step by time step and outputs the hidden state as the generator state feature vector at the last time step.

[0127] For example, for the generator state parameter time series X_state with a shape of (52560,2):

[0128] Divide X_state into subsequences of length 24, and get a total of 2190 subsequences;

[0129] For each subsequence, encode it through 2 layers of LSTM to obtain a 128-dimensional hidden state vector;

[0130] The hidden state vectors of all subsequences are concatenated to obtain the generator set state feature matrix F_state with a shape of (2190,128).

[0131] The meteorological feature matrix F_weather extracted by CNN and the generator state feature matrix F_state extracted by LSTM are aligned in the time dimension and concatenated into a long vector along the feature dimension. Then, the concatenated long vector is mapped to the predicted value of wind and solar power generation through one or more fully connected layers.

[0132] The number of output nodes of the last fully connected layer is set to 1, corresponding to the predicted value of wind and solar power generation. The mean square error (MSE) loss function is used to measure the difference between the predicted power and the actual power. The MSE loss function calculates the mean of the square of the difference between the predicted value and the actual value.

[0133] For example, for F_weather with shape (2190,256) and F_state with shape (2190,128):

[0134] Align the two in the time dimension and concatenate them in the feature dimension to obtain a feature matrix F with a shape of (2190,384);

[0135] Through a 512-node fully connected layer and ReLU activation, F is mapped to an intermediate feature of (2190,512);

[0136] Through a 256-node fully connected layer and ReLU activation, the intermediate features are mapped to high-level features of (2190, 256);

[0137] Through a 1-node fully connected layer, the high-level features are mapped to power prediction values ​​of (2190,1).

[0138] Assuming that at a certain time step, the actual wind and solar power generation power is 300 kW, and the model's predicted power is 320 kW, then the MSE loss at this time step is (320-300)2=400.

[0139] The wind and solar power generation prediction model (including CNN, LSTM and fully connected layers) is jointly optimized in an end-to-end manner. The back-propagation algorithm is used to calculate the gradient of the loss function to the parameters of each layer, and the gradient descent optimization algorithm is used to update the model parameters to minimize the loss function.

[0140] In order to improve the generalization performance and robustness of the model, the following strategies are adopted during the training process:

[0141] Data enhancement: Randomly crop, flip, rotate and other transformations are performed on the meteorological parameter matrix and generator state sequence to increase the diversity of training samples.

[0142] Regularization: Add L2 regularization term to the loss function to constrain the size of model parameters and reduce the risk of overfitting.

[0143] Early stopping: Evaluate the performance of the model on the validation set after each epoch. If the performance of the validation set does not improve within multiple epochs, stop training and return the model parameters with the best performance.

[0144] Assume that the Adam optimizer is used, the initial learning rate is 0.001, and the learning rate is reduced to 0.5 times the original value every 10 epochs. The L2 regularization coefficient is 0.01. The early_stopping parameter is 5, which means that the training is stopped if the performance of the validation set does not improve after 5 consecutive epochs. In each epoch, 256 samples are randomly sampled from the training set to form a batch. In each batch, the meteorological parameter matrix is ​​randomly cropped and flipped, and the generator state sequence is randomly cropped and rotated, and then the model is trained with the enhanced data.

[0145] After 50 epochs of training, the optimal MSE loss on the validation set was 18.65 (kW)2, and the corresponding root mean square error (RMSE) of wind and solar power generation prediction was 4.32kW, with a prediction accuracy of 95.6%. The trained model was applied to the test set, and the MSE loss was 20.41 (kW)2, the RMSE was 4.52kW, and the prediction accuracy was 94.9%, indicating that the model has good generalization performance and can effectively predict the future power generation of wind and solar power generation systems.

[0146] In an optional embodiment,

[0147] The electric vehicle load demand curve and the wind and solar power generation curve are matched and analyzed to obtain the matching difference between the two. When the matching difference is greater than the preset threshold, the steps of triggering the coordinated control process include:

[0148] Calculate the matching difference between the electric vehicle load demand curve and the wind and solar power generation curve within a given time range. The matching difference calculation formula is the average absolute deviation of the two curves within a given time range. Compare the calculated matching difference with the preset threshold. When the matching difference is greater than the preset threshold, trigger the collaborative control process.

[0149] Taking into account the tram operating conditions, wind and solar power generation forecast results and energy storage system status factors, a multi-objective optimization model is established with minimizing the matching difference and tram energy consumption as the optimization objectives, and tram operating constraints and energy storage system constraints as constraints.

[0150] Solve the multi-objective optimization model to obtain a set of Pareto optimal solutions, from which the optimal solution of balanced matching difference and tram energy consumption is selected, and its corresponding tram operation parameters and energy management strategy are used as the collaborative control strategy;

[0151] The formulated coordinated control strategy is sent to the tram control system and energy storage management system. By coordinating and controlling the tram's traction system, braking system and power converter of the energy storage system, the tram's operating parameters and energy management strategy are optimized and adjusted, reducing the imbalance between the tram's power load demand and wind and solar power generation.

[0152] According to the real-time changes in the tram power load demand and wind and solar power generation as well as the feedback information on the collaborative control effect, the collaborative control strategy is updated and optimized in real time to ensure the operation of the tram system.

[0153] For example, after obtaining the electric vehicle load forecast curve and the wind and solar power forecast curve, it is necessary to match and analyze the two, and decide whether to trigger the collaborative control process based on the matching difference. If the collaborative control is triggered, the corresponding electric vehicle operation parameter optimization strategy and energy management strategy are generated, and the efficient collaborative operation of the electric vehicle-wind and solar system is achieved by coordinating and controlling the electric vehicle's traction system, braking system and energy storage system. The specific steps are as follows:

[0154] Calculate the matching difference between the electric load demand curve of the electric vehicle and the wind and solar power generation curve within a given time range (such as the next hour). The matching difference is measured by the average absolute deviation of the two curves within a given time range. The calculation formula is: for each time step, calculate the absolute value of the difference between the electric load demand of the electric vehicle and the wind and solar power generation, and then average the absolute deviations of all time steps.

[0155] The calculated matching difference is compared with the preset threshold. When the matching difference is greater than the preset threshold, the collaborative control process is triggered; otherwise, the independent operation of the tram and the wind and solar system is maintained. The selection of the preset threshold needs to comprehensively consider factors such as the system's tolerance, control flexibility, and optimization potential, and can be determined through empirical values ​​or simulation tests.

[0156] For example, assume that there are 60 time steps in the next hour with a time step of 1 minute. At a certain time step, the electric load demand of the tram is 80kW, and the wind and solar power generation is 100kW. The absolute deviation between the two is |100-80|=20kW. The average of the absolute deviations of all time steps is 25kW. Assume that the preset threshold is 20kW. Since 25kW>20kW, the collaborative control process is triggered.

[0157] Taking into account factors such as tram operating conditions, wind and solar power generation forecast results, and energy storage system status (such as SOC), a multi-objective optimization model is established with minimizing matching difference and tram energy consumption as optimization goals, and tram operation constraints and energy storage system constraints as constraints.

[0158] Among them, the tram operation constraints include the tram speed limit, acceleration limit, mileage requirements, etc.; the energy storage system constraints include the capacity limit, power limit, SOC upper and lower limits, etc. of the energy storage device.

[0159] In the optimization objectives, matching difference and tram energy consumption are usually two contradictory indicators. Reducing the matching difference requires the tram to frequently adjust the operating power to adapt to the fluctuations of wind and solar power generation, which often increases the tram's energy consumption; on the contrary, if the tram energy consumption is reduced blindly, the tram operating power curve and the wind and solar power generation curve may deviate greatly, and the matching difference is difficult to reduce. Therefore, it is necessary to seek a balance between the two objectives to obtain the optimal control strategy that takes into account both matching performance and energy-saving effects.

[0160] Since the two optimization objectives of matching difference and tram energy consumption are usually contradictory, the multi-objective optimization model often does not have a unique optimal solution, but a set of Pareto optimal solutions. A Pareto optimal solution is a solution in which any objective in the current solution must be further optimized to make another objective worse.

[0161] In order to obtain the Pareto optimal solution set, some multi-objective optimization algorithms can be used, such as genetic algorithms (NSGA-II, MOEA / D, etc.), particle swarm optimization algorithm (MOPSO), etc. These algorithms can find multiple Pareto optimal solutions with relatively uniform distribution in one operation by maintaining a population and iteratively evolving the population, providing decision makers with diverse choices.

[0162] From the Pareto optimal solution set obtained, the optimal solution that balances the matching difference and the tram energy consumption is selected as the final collaborative control strategy. This requires formulating selection criteria based on actual needs, such as selecting the solution with the smallest matching difference, the solution with the smallest tram energy consumption, or selecting the solution with the best comprehensive performance through weighted summation and other methods.

[0163] For example, suppose that the multi-objective optimization model is solved by the NSGA-II algorithm, and a Pareto optimal solution set containing 20 solutions is obtained. Among them, the solution with the smallest matching difference is 18kW, and the corresponding tram energy consumption is 120kWh; the solution with the smallest tram energy consumption is 100kWh, and the corresponding matching difference is 28kW. If you want to take into account both matching performance and energy-saving effects, you can choose a solution with a matching difference of 22kW and a tram energy consumption of 105kWh as the final collaborative control strategy. The tram operating parameters (such as speed curve, braking curve, etc.) and energy management parameters of the energy storage system (such as charging and discharging power, SOC scheduling curve, etc.) corresponding to this strategy will be issued as control instructions.

[0164] The formulated collaborative control strategy will be sent to the tram control system and energy storage management system. By coordinating and controlling the power converters of the tram's traction system, braking system and energy storage system, the tram's operating parameters and energy management strategy can be optimized and adjusted, reducing the imbalance between the tram's electricity load demand and wind and solar power generation.

[0165] Specifically, the tram control system controls the working state of the traction motor and brake resistor according to the optimized speed curve and braking curve, and adjusts the traction power and braking power of the tram; the energy storage management system controls the bidirectional DC / DC converter of the energy storage device according to the optimized charging and discharging power and SOC dispatching curve, and adjusts the charging and discharging power of the energy storage device. Through the coordinated cooperation of the tram and the energy storage device, the tram's power load curve will be matched with the wind and solar power generation curve as much as possible, reducing power fluctuations and spare capacity requirements on the grid side.

[0166] For example, at a certain moment, the wind and solar power generation is 120kW, while the electric load demand of the tram before optimization is 80kW, a difference of 40kW. According to the coordinated control strategy, the tram speed curve is optimized and adjusted, and the electric load demand of the tram is increased to 110kW; at the same time, the charging power of the energy storage device is set to 10kW to absorb the surplus power of wind and solar power generation. After coordinated control, the instantaneous power difference between the electric load of the tram and the wind and solar power generation is reduced to 10kW, which greatly reduces the imbalance of the system.

[0167] Since both the electric load demand of trams and the power generation of wind and solar power have certain uncertainties and fluctuations, it is difficult for a single coordinated control strategy to continuously ensure good control performance. Therefore, it is necessary to update and optimize the coordinated control strategy in real time based on the real-time feedback information of the system operation.

[0168] Specifically, during the operation of the tram and wind and solar systems, data such as tram power load, wind and solar power generation, and energy storage system status are collected in real time, and compared with the predicted values ​​to evaluate the execution effect of the current collaborative control strategy. If the actual operation deviation exceeds the predetermined range, or the operating parameters of the tram / energy storage system trigger the warning threshold, the collaborative control strategy will be re-optimized.

[0169] When re-optimizing, the current system state is used as the initial condition, the optimization time domain is updated and the tram power load and wind and solar power generation are re-predicted, a new multi-objective optimization model is constructed and solved, and an updated collaborative control strategy is obtained. By optimizing and updating the control strategy regularly or by event triggering, the real-time changes of the system can be responded to in a timely manner, and the efficiency and economy of the coordinated operation of trams, wind and solar can be continuously guaranteed.

[0170] For example, in the 30th minute of the execution of the collaborative control strategy, the actual wind and solar power generation was 15% lower than the predicted value, causing the matching difference between the tram power load and the wind and solar power generation to rise to 35kW, exceeding the predetermined threshold of 30kW, triggering the re-optimization of the collaborative control strategy. Taking the actual system state at the 30th minute as the initial condition, the tram power load and wind and solar power generation for the next 30 minutes are predicted, and the multi-objective optimization model is updated accordingly. The optimization model is solved to obtain the updated collaborative control strategy, which is then sent down for execution. The updated control strategy reduces the running speed of the tram and increases the discharge power of the energy storage device, reducing the matching difference to less than 25kW, meeting the performance requirements of collaborative control again.

[0171] In an optional embodiment,

[0172] Taking into account the tram operating conditions, wind and solar power generation forecast results and energy storage system status factors, the steps of establishing a multi-objective optimization model with minimizing matching difference and tram energy consumption as optimization goals and tram operating constraints and energy storage system constraints as constraints include:

[0173] The speed curve, traction curve and charging and discharging power curve of the tram are selected as the optimization variables of the multi-objective optimization model. The power transfer model of the tram traction system is used to calculate the tram load demand power, and the wind and solar power generation prediction model is used to predict the wind and solar power generation power.

[0174] Construct a matching difference objective function, expressed as the average absolute deviation between the electric load demand power of the tram and the wind and solar power generation power within a given time range, and construct a tram energy consumption objective function, expressed as the integral of the electric load demand power of the tram within a given time range; set tram operation constraints, including speed constraints, acceleration constraints, and traction constraints; set energy storage system constraints, including state of charge constraints and charge and discharge power constraints, where the state of charge is calculated by the initial state of charge and the charge and discharge power of the energy storage system;

[0175] The matching difference objective function, tram energy consumption objective function, tram operation constraints and energy storage system constraints are combined to form a multi-objective optimization model.

[0176] For example, in order to optimize the coordinated operation of the tram-wind-solar system, it is necessary to establish a multi-objective optimization model that comprehensively considers tram operation, wind-solar power generation, and energy storage system. The model takes minimizing the matching difference between tram power load and wind-solar power generation and tram operation energy consumption as the optimization goal, while satisfying various constraints of tram operation and energy storage system. The steps to construct the multi-objective optimization model are as follows:

[0177] The speed curve, traction curve and charging and discharging power curve of the tram are selected as the optimization variables of the multi-objective optimization model. The speed curve and traction curve determine the operating conditions and power load demand of the tram, and the charging and discharging power curve of the energy storage system determines the energy interaction between the energy storage device and the tram and the wind and solar system.

[0178] Specifically, the optimization time domain is discretized, and the speed curve, traction force curve and charge-discharge power curve are represented by piecewise constant or piecewise linear method. Each time step corresponds to a speed value, traction force value and charge-discharge power value, and the variable values ​​of all time steps are connected in series to form a complete curve.

[0179] For example, assuming that the optimization time domain is one hour in the future and the time step is one minute, there are 60 time steps in total. The speed curve can be expressed as [v1, v2, ..., v60], where vi represents the speed value of the i-th time step. The traction force curve and the charge and discharge power curve are also expressed in a similar way. If there are four optimization variables (speed, traction force, energy storage charging power, energy storage discharging power) in each time step, the total number of optimization variables is 60×4=240.

[0180] The power transfer model of the tram traction system is used to calculate the tram electrical load power demand corresponding to the tram speed curve and traction force curve at each time step. The model takes into account the basic resistance, additional resistance, acceleration resistance, etc. of the tram, and converts mechanical power into electrical power demand based on the efficiency characteristics of the electric transmission system.

[0181] The wind and solar power generation prediction model is used to predict the wind and solar power generation power at each time step. The prediction model is based on historical data, combined with weather forecast information, geographic location parameters, etc., and uses statistical learning or physical modeling methods to obtain the predicted value of wind and solar power generation power in the future.

[0182] For example, assuming that at a certain time step, the tram speed is 80km / h, the traction force is 25kN, and the tram load demand power is 1.2MW according to the power transfer model. At the same time, the wind and solar power generation prediction model gives a wind and solar power generation power prediction value of 0.9MW at this time step.

[0183] Construct a matching difference objective function, expressed as the average absolute deviation between the electric load demand power of the electric vehicle and the wind and solar power generation power in the optimization time domain. For each time step, calculate the absolute value of the difference between the electric load demand power of the electric vehicle and the wind and solar power generation power, and then average the absolute deviations of all time steps to obtain the matching difference. The smaller the matching difference, the higher the matching degree between the electric load of the electric vehicle and the wind and solar power generation.

[0184] Construct the tram energy consumption objective function, which is expressed as the integral of the tram power demand in the optimization time domain. The tram power demand at all time steps is accumulated and summed, and then multiplied by the time step to obtain the total electric energy consumed by the tram in the optimization time domain, that is, the tram energy consumption. The smaller the tram energy consumption, the higher the energy utilization efficiency during the tram operation.

[0185] For example, continuing the example in the previous step, assuming that in three consecutive time steps, the power demand of the electric load of the tram is 1.2MW, 1.0MW, and 0.8MW, respectively, and the corresponding wind and solar power generation power forecast values ​​are 0.9MW, 1.1MW, and 0.7MW, respectively. Then the absolute values ​​of the power deviations in these three time steps are 0.3MW, 0.1MW, and 0.1MW, respectively. After averaging, the matching difference objective function value is (0.3+0.1+0.1) / 3=0.167MW. The objective function value of the tram energy consumption is (1.2+1.0+0.8)×1min=3.0MWmin=180kWh.

[0186] Set the tram operation constraints, including speed constraints, acceleration constraints and traction constraints. The speed constraint limits the upper and lower limits of the tram speed at each time step, the acceleration constraint limits the maximum amplitude of the tram speed change, and the traction constraint limits the output force range of the tram traction motor. These constraints ensure the feasibility and safety of the optimized tram speed curve and traction curve in actual operation.

[0187] Set energy storage system constraints, including state of charge (SOC) constraints and charge and discharge power constraints. The SOC constraint defines the upper and lower limits of the state of charge of the energy storage device to avoid overcharging or over-discharging that affects the battery life. The SOC of the energy storage device can be iteratively calculated by the initial SOC and the charge and discharge power at each time step. The charge and discharge power constraint defines the upper limit of the charge and discharge power of the energy storage device to protect the energy storage device from high power shocks.

[0188] For example, assume that the tram speed constraint is [40,120]km / h, that is, the tram speed must always be maintained between 40 and 120km / h. The acceleration constraint is [-1,1]m / s2, that is, the absolute value of the tram acceleration cannot exceed 1m / s2. The traction constraint is [0,350]kN, that is, the tram traction must vary between 0 and 350kN. The SOC constraint of the energy storage device is [0.2,0.9], that is, the SOC must be maintained between 20% and 90%. The charging and discharging power constraints of the energy storage device are [0,2]MW and [-2,0]MW, respectively, that is, the charging power does not exceed 2MW, and the discharging power does not exceed 2MW.

[0189] The matching difference objective function, the tram energy consumption objective function, the tram operation constraints and the energy storage system constraints are combined to form a complete multi-objective optimization model. The decision variables of this model are the tram speed curve, the traction curve and the energy storage system charging and discharging power curve. The goal is to reduce the matching difference and tram energy consumption as much as possible while meeting various constraints.

[0190] Since matching difference and tram energy consumption are usually conflicting objectives, a trade-off needs to be made between the two. The multi-objective optimization problem can be transformed into a single-objective problem, such as using the weighted summation method to merge the two objectives into a comprehensive objective, or using the ε-constraint method to transform one of the objectives into a constraint condition. It is also possible to retain the multi-objective characteristics and use a multi-objective optimization algorithm to directly solve the Pareto optimal solution set.

[0191] At this point, the multi-objective optimization model for the coordinated operation of the tram-wind-solar system has been completed. The model comprehensively considers the coupling relationship between tram operation, wind-solar power generation, and energy storage system, takes the tram speed curve, traction curve, and energy storage charging and discharging power curve as optimization variables, takes matching difference and tram energy consumption as optimization goals, and takes tram operation constraints and energy storage constraints as feasible domain boundaries, laying the foundation for the formulation of the tram-wind-solar coordinated control strategy.

[0192] In an optional embodiment,

[0193] In the collaborative control process, the power generation of the wind and solar power generation system and the actual power load of the tram are monitored in real time, the matching difference is calculated, and the power generation of the wind and solar power generation system is dynamically adjusted according to the changing trend of the matching difference. At the same time, the actual power load data of the tram is fed back to the user's driving behavior model for online learning, the driving behavior model is updated, and the prediction accuracy is improved, thereby achieving accurate collaborative control of wind and solar power generation and tram load. The steps include:

[0194] According to the changing trend of the matching difference between the electric load demand curve of the tram and the wind-solar power generation curve, the power generation power of the wind-solar power generation system is dynamically adjusted based on the proportional-integral-differential controller, and the wind-solar power generation power adjustment output by the proportional-integral-differential controller is superimposed on the original wind-solar power generation power, and the adjusted wind-solar power generation power is achieved by controlling the power converter of the wind-solar power generation system;

[0195] The actual tram power load data is fed back into the driving behavior model for online learning. The prediction error is calculated by comparing the actual measured tram power load with the tram power load predicted by the driving behavior model. The weight and bias of the driving behavior model are updated through the back propagation algorithm to minimize the prediction error and improve the tram power load prediction accuracy.

[0196] Based on the idea of ​​predictive control of the driving behavior model, the collaborative control strategy is dynamically optimized. At the current moment, the predicted value of wind and solar power generation and the predicted value of tram power load in the future period are obtained. Taking the current state as the initial condition, the matching difference and tram energy consumption in the future prediction time domain as the optimization goals, and the tram operation constraints and energy storage system constraints as constraints, a multi-objective optimization model is established to solve the optimal control sequence in the future prediction time domain. The first element of the optimal control sequence is used as the control quantity at the current moment and implemented in the tram control system and energy storage management system. The above steps are repeated at the next moment to achieve rolling optimization.

[0197] For example, during the actual operation of the tram-wind-solar system, the control strategy needs to be dynamically adjusted according to real-time monitoring data to adapt to the fluctuations of wind-solar power generation and tram load. The core goal of collaborative control is to minimize the matching difference between wind-solar power generation and tram power load, while taking into account tram energy consumption and operating performance. The implementation steps of the collaborative control strategy are as follows:

[0198] According to the changing trend of the matching difference between the electric load demand curve of the tram and the wind and solar power generation curve, the power generation of the wind and solar power generation system is dynamically adjusted. The adjustment process is implemented using a proportional-integral-differential (PID) controller.

[0199] First, the power deviation between the actual measured electric load of the tram and the wind and solar power generation in each time step is calculated as the input of the PID controller. The PID controller calculates the adjustment of the wind and solar power generation according to the size, change speed and accumulation of the deviation. The adjustment includes three components: the proportional component is proportional to the deviation value, the integral component is proportional to the cumulative value of the deviation, and the differential component is proportional to the change speed of the deviation. The three components are weighted and summed to obtain the total wind and solar power generation adjustment.

[0200] Then, the wind-solar power adjustment output by the PID controller is added to the original wind-solar power to obtain the adjusted wind-solar power command. The adjusted wind-solar power is achieved by controlling the power converter of the wind-solar power generation system. The power converter adjusts the pitch angle of the wind turbine and the excitation current of the generator according to the power generation command, adjusts the working voltage and maximum power point tracking of the photovoltaic array, etc., so that the actual output power of the wind-solar power generation system tracks the command value.

[0201] For example, assuming that at a certain time step, the electric load of the tram is 1.5MW and the wind and solar power generation is 1.2MW, the power deviation is 0.3MW. The PID controller calculates the wind and solar power generation power adjustment to be 0.2MW, and after superposition, the new wind and solar power generation power instruction is 1.4MW. The power converter of the wind and solar power generation system adjusts the operating parameters of the wind power and photovoltaic power generation units according to the instruction value of 1.4MW, so that the actual power generation can track the instruction value as quickly as possible and reduce the matching difference with the tram power load.

[0202] The actual electric load data of trams is fed back to the driving behavior model for online learning to continuously improve the prediction accuracy of tram electric load. The driving behavior model predicts the electric load of trams from driving behavior parameters such as vehicle speed, acceleration, and steering angle based on historical data.

[0203] At each time step, the actual measured tram load is compared with the tram load predicted by the driving behavior model to calculate the prediction error. The prediction error is used as the loss function of the back-propagation algorithm, and the weights and biases of the driving behavior model are updated by the gradient descent method to minimize the prediction error. Repeated iterations enable the model to continue to learn and improve when new data arrives, and the prediction accuracy continues to improve.

[0204] Specifically, a long short-term memory (LSTM) neural network is used to construct a driving behavior model. The input is a sequence of driving behavior parameters in a historical time window, and the output is a sequence of tram power load in a future time window. The network consists of an input layer, an LSTM hidden layer, and a fully connected output layer. The mean square error between the actual measured tram power load and the predicted value output by the model is calculated, and the error gradient is fed back to each layer of the network through back propagation. The Adam optimizer is used to update the weights to continuously reduce the prediction loss.

[0205] For example, suppose the LSTM model predicts at a certain moment that the tram power load in the next minute is [1.3, 1.4, 1.5, 1.6] MW, while the actual measured value is [1.2, 1.5, 1.4, 1.7] MW. The prediction error is the mean of the mean square error (1.3-1.2)^2+(1.4-1.5)^2+(1.5-1.4)^2+(1.6-1.7)^2, which is about 0.065MW^2. Using this error as the loss function, the Adam optimizer is used to update the weights of each layer of the LSTM network so that the error decreases in the next iteration. Repeat the training for multiple epochs until the prediction error converges to below the expected value.

[0206] The collaborative control strategy is dynamically optimized based on the idea of ​​driving behavior model predictive control (MPC). MPC uses historical measurement data and future prediction data to solve the optimization control problem in the rolling time domain, and implements the first element of the optimization result as the current control quantity, continuously rolling forward.

[0207] At each control moment, the wind and solar power generation power forecast and the electric vehicle load forecast for a period of time in the future (called the forecast time domain) are first obtained. The wind and solar power generation power forecast is based on factors such as weather forecasts and historical power generation data, and is generated using the forecast model described in Section 4.3.3. The electric vehicle load forecast is given by the driving behavior model updated in step (2).

[0208] Then, with the current tram state (such as speed, position, power, etc.) as the initial condition, the matching difference and tram energy consumption in the future prediction time domain as the optimization target, and the tram operation constraints (such as speed, acceleration, traction limit) and energy storage system constraints (such as SOC, charging and discharging power limit) as constraints, a multi-objective optimization model is established, as described in Section 4.3.4.2. The multi-objective optimization algorithm is used to solve the model and obtain the optimal control sequence of tram speed, traction, and energy storage power in the future prediction time domain.

[0209] Finally, the first element of the optimal control sequence is used as the control instruction at the current control moment and sent to the tram control system and energy storage management system. The tram control system adjusts the torque and power of the motor and the braking force according to the speed and traction instructions to achieve speed tracking. The energy storage management system controls the working state of the energy storage device according to the charge and discharge power instructions to achieve energy storage and release.

[0210] At the next control moment, the above process is repeated to update the current state and future prediction data, generate a new optimal control sequence, and implement new control instructions. Through continuous rolling optimization, the tram operation and energy storage strategies can be continuously and dynamically adjusted to adapt to the real-time changing wind and solar power generation and load requirements, and ultimately achieve precise coordinated control of wind and solar power generation and tram loads.

[0211] For example, assume that at the current control moment, the tram speed is 90km / h, the location is 2km away from the next station, and the power is 60%. According to weather forecasts and historical data, the wind and solar power generation sequence is predicted to be [1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.7, 1.6, 1.5, 1.4] MW in the next 10 minutes. According to the updated driving behavior model, the tram power load sequence is predicted to be [1.4, 1.3, 1.5, 1.6, 1.4, 1.3, 1.5, 1.6, 1.4, 1.3] MW in the next 10 minutes. With the goal of matching the difference and minimizing the energy consumption of the tram, a multi-objective optimization model for the next 10 minutes is established. The constraints are the tram speed range [85,100] km / h, the acceleration range [-0.5,0.5] m / s2, the energy storage SOC range [0.2,0.9], and the energy storage charging and discharging power range [-1,1] MW. Solving the optimization problem, the optimal tram speed sequence for the next 10 minutes is [91,90,92,93,91,90,92,93,91,90] km / h, and the optimal energy storage power sequence is [0.2,0.3,0.1,0.0,0.2,0.3,0.1,0.0,0.2,0.3] MW. The speed of 91 km / h and the energy storage power of 0.2 MW in the first time step are used as the current control instructions to control the tram to accelerate to 91 km / h, and the energy storage device charges 0.2 MW at the same time. At the next control moment, the MPC process is repeated to update the control instructions based on the new state and prediction data.

[0212] In summary, the coordinated control strategy of wind and solar power generation and tram load is realized through real-time data monitoring, PID feedback control, online learning of driving behavior, model prediction dynamic optimization and other steps. This strategy can make full use of the fluctuation characteristics of wind and solar power generation, flexibly adjust the tram operation and energy storage mode, and minimize the matching difference between wind and solar power generation and tram power load while meeting the normal operation of the tram, improve the energy utilization efficiency and economy of the tram system, and provide strong support for the green and sustainable development of rail transportation.

[0213] Figure 2 FIG. 1 is a schematic diagram of a wind-solar power generation and electric vehicle load coordinated control system based on user behavior according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0214] The first unit is used to obtain the historical driving data and the real-time driving data of the tram, train the historical driving data based on the machine learning algorithm to obtain the user's driving behavior model, and use the driving behavior model to analyze the real-time driving data to obtain the tram power load demand curve in the future period; obtain the historical power generation data and real-time power generation data of the wind and solar power generation system, the wind and solar power generation system includes a wind power generation system and a photovoltaic power generation system, the historical power generation data includes historical meteorological parameters and generator set status parameters, and the real-time power generation data includes meteorological parameters and generator set status parameters measured in real time; use a deep learning algorithm to train the historical power generation data to obtain a wind and solar power generation prediction model, the deep learning algorithm is a combination model of a convolutional neural network and a long short-term memory neural network, the spatial characteristics of the meteorological parameters are extracted by the convolutional neural network, and the time series characteristics of the generator set status are extracted by the long short-term memory neural network; use the wind and solar power generation prediction model to analyze the real-time power generation data, and combine it with numerical weather forecast data to predict the power generation curve of the wind and solar power generation system in the future period;

[0215] The second unit is used to match and analyze the electric load demand curve of the tram with the wind and solar power generation curve to obtain the matching difference between the two. When the matching difference is greater than the preset threshold, the collaborative control process is triggered; in the collaborative control process, according to the distribution characteristics of the electric load demand of the tram, the electric load demand of the tram in the forecast time period is divided into peak demand, average demand and valley demand. When the electric load of the tram is at the peak demand, the wind and solar power generation system is controlled to increase the power generation, and the grid power is started to supplement when necessary; when the electric load of the tram is at the average demand, the wind and solar power generation system is controlled to output power steadily according to the predicted power generation; when the electric load of the tram is at the valley demand, the wind and solar power generation system is controlled to reduce the power generation and store the surplus power in the tram battery pack or the grid;

[0216] The third unit is used to monitor the power generation of the wind and solar power generation system and the actual power load of the tram in real time during the coordinated control process, calculate the matching difference, and dynamically adjust the power generation of the wind and solar power generation system according to the changing trend of the matching difference. At the same time, the actual power load data of the tram is fed back to the user's driving behavior model for online learning, the driving behavior model is updated, and the prediction accuracy is improved, thereby realizing precise coordinated control of wind and solar power generation and tram load.

[0217] According to a third aspect of the embodiments of the present invention,

[0218] An electronic device is provided, comprising:

[0219] processor;

[0220] a memory for storing processor-executable instructions;

[0221] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0222] According to a fourth aspect of the embodiments of the present invention,

[0223] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0224] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0225] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0226] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0227] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for coordinated control of wind and solar power generation and electric vehicle load based on user behavior, characterized in that: The following steps are involved: Based on the historical and real-time driving data of trams, the demand curve of tram electricity load in the future is predicted; Based on the historical power generation data and real-time power generation data of the wind and solar power generation system, the wind and solar power generation curve in the future period is predicted; According to the matching difference between the electric vehicle electricity load demand curve and the wind and solar power generation curve, and according to the distribution characteristics of the electric vehicle electricity load demand, the electric vehicle electricity load demand within the forecast time period is classified, and according to the classification result, the wind and solar power generation and the electric vehicle load are coordinated and controlled.

2. The method for coordinated control of wind and solar power generation and electric vehicle load based on user behavior according to claim 1 is characterized in that: In the process of acquiring historical driving data and real-time driving data, the historical driving data includes the GPS trajectory data, accelerator pedal opening data, brake pedal opening data and battery pack charge state data of the tram, and the real-time driving data includes the tram’s current GPS position, speed, accelerator pedal opening, brake pedal opening and battery pack charge state parameter data.

3. The method for coordinated control of wind and solar power generation and electric vehicle load based on user behavior according to claim 2 is characterized in that: In the process of obtaining the electric load demand curve of the electric vehicle, the historical driving data and the real-time driving data are preprocessed, the preprocessed historical driving data are input into the long short-term memory neural network in the form of time series, the input features are mapped into vehicle speed, acceleration and battery pack power as the output of the network through the vehicle dynamics model and the energy consumption model, an encoder-decoder network architecture is adopted, the network weights are trained through the cross entropy loss function and the adaptive moment estimation optimization algorithm, and the driving behavior model is obtained; It is used to predict the electric load demand curve of trams in the future; the preprocessing process includes: removing outliers and invalid values ​​through data cleaning, aligning the data collected by different sensors on the time axis through data alignment, and scaling the data of different physical dimensions to the same numerical range through data normalization.

4. The method for coordinated control of wind and solar power generation and electric vehicle load based on user behavior according to claim 3 is characterized by: In the process of obtaining the electric load demand curve of the tram, based on the driving behavior model and according to the real-time collected tram driving data, the speed, acceleration and energy consumption change trends of the tram in the future period are predicted, and the predicted speed and acceleration changes are converted into power demand changes of the battery pack through the energy consumption model and the battery model, so as to obtain the electric load demand curve of the tram in the predicted time period; when predicting the electric load, the weather conditions and road conditions information at the current location of the tram are introduced as auxiliary inputs, and the weather conditions including temperature, humidity and wind speed are obtained, and the road conditions information including road type, traffic flow and speed limit are obtained, and the weather conditions and road conditions information are encoded and input into the driving behavior model, so that the driving behavior model can adapt to different external conditions and dynamically adjust the electric load demand curve of the tram.

5. The method for coordinated control of wind and solar power generation and electric vehicle load based on user behavior according to claim 4 is characterized in that: In the process of obtaining the wind and solar power generation curve, based on the combined model of convolutional neural network and long short-term memory neural network, the model is trained through historical power generation data, the convolutional neural network is used to extract the spatial characteristics of meteorological parameters, and the long short-term memory neural network is used to extract the time series characteristics of the generator set status, and a wind and solar power generation prediction model is constructed to predict the wind and solar power generation curve in the future.

6. The method for coordinated control of wind and solar power generation and electric vehicle load based on user behavior according to claim 5 is characterized by: In the process of building a wind and solar power generation prediction model, the meteorological feature vector extracted by the convolutional neural network and the generator set state feature vector extracted by the long short-term memory neural network are spliced ​​into a long vector, and then the spliced ​​long vector is mapped to the predicted value of wind and solar power generation through one or more fully connected layers, and the loss function is used to measure the difference between the predicted value of wind and solar power generation and the actual power generation; the wind and solar power generation prediction model is optimized in an end-to-end manner, the back propagation algorithm and the gradient descent optimization algorithm are used to update the parameters of the wind and solar power generation prediction model, and the wind and solar power generation prediction model is optimized through data enhancement, regularization and early stopping strategies.

7. The method for coordinated control of wind and solar power generation and electric vehicle load based on user behavior according to claim 6 is characterized by: In the process of obtaining the matching difference, the matching difference is obtained according to the average absolute deviation of the two curves within a given time range.

8. The method for coordinated control of wind and solar power generation and electric vehicle load based on user behavior according to claim 7 is characterized in that: Before the collaborative control process is performed, the collaborative control process is triggered according to the matching difference, including the following steps: Considering the tram operating conditions, wind and solar power generation forecast results and energy storage system status factors, a multi-objective optimization model is established with minimizing the matching difference and tram energy consumption as the optimization objectives, and tram operating constraints and energy storage system constraints as constraints. Solve the multi-objective optimization model to obtain a set of Pareto optimal solutions, from which the optimal solution of balanced matching difference and tram energy consumption is selected, and its corresponding tram operation parameters and energy management strategy are used as the collaborative control strategy; The formulated coordinated control strategy is sent to the tram control system and energy storage management system. By coordinating and controlling the tram's traction system, braking system and power converter of the energy storage system, the tram's operating parameters and energy management strategy are optimized and adjusted, reducing the imbalance between the tram's power load demand and wind and solar power generation. According to the real-time changes in the tram power load demand and wind and solar power generation as well as the feedback information on the collaborative control effect, the collaborative control strategy is updated and optimized in real time to ensure the operation of the tram system.

9. The method for coordinated control of wind and solar power generation and electric vehicle load based on user behavior according to claim 8 is characterized in that: In the process of constructing the multi-objective optimization model, the speed curve, traction curve and charging and discharging power curve of the tram are selected as the optimization variables of the multi-objective optimization model. The power transfer model of the tram traction system is used to calculate the tram load demand power, and the wind and solar power generation prediction model is used to predict the wind and solar power generation power. Construct a matching difference objective function, expressed as the average absolute deviation between the electric load demand power of the tram and the wind and solar power generation power within a given time range, and construct a tram energy consumption objective function, expressed as the integral of the electric load demand power of the tram within a given time range; set tram operation constraints, including speed constraints, acceleration constraints, and traction constraints; set energy storage system constraints, including state of charge constraints and charge and discharge power constraints, where the state of charge is calculated by the initial state of charge and the charge and discharge power of the energy storage system; The multi-objective optimization model is formed by combining the matching difference objective function, the tram energy consumption objective function, the tram operation constraint conditions and the energy storage system constraint conditions.

10. The method for coordinated control of wind and solar power generation and electric vehicle load based on user behavior according to claim 9 is characterized in that: In the process of coordinated control, according to the changing trend of the matching difference between the electric load demand curve of the tram and the wind and solar power generation curve, the power generation power of the wind and solar power generation system is dynamically adjusted based on the proportional-integral-differential controller, and the wind and solar power generation power adjustment output by the proportional-integral-differential controller is superimposed on the original wind and solar power generation power, and the adjusted wind and solar power generation power is realized by controlling the power converter of the wind and solar power generation system; The actual tram power load data is fed back into the driving behavior model for online learning. The prediction error is calculated by comparing the actual measured tram power load with the tram power load predicted by the driving behavior model, and the weights and biases of the driving behavior model are updated through the back-propagation algorithm. Based on the driving behavior model predictive control, the collaborative control strategy is dynamically optimized. At the current moment, the predicted value of wind and solar power generation and the predicted value of tram power load in the future period are obtained. Taking the current state as the initial condition, the matching difference degree and tram energy consumption in the future prediction time domain as the optimization target, and the tram operation constraints and energy storage system constraints as the constraints, the optimal control sequence in the future prediction time domain is obtained. The first element of the optimal control sequence is used as the control quantity at the current moment and implemented in the tram control system and energy storage management system. The above steps are repeated at the next moment to achieve rolling optimization.

Images