Data-driven super-multi-objective optimized electric vehicle charging space-time drainage method and device

Abstract

The invention discloses a data-driven super-multi-objective optimized electric vehicle charging space-time drainage method and device. The method comprises the following steps: S1, constructing a super-multi-target optimization model which simultaneously considers five targets of minimum energy load, minimum charging or battery lease cost, minimum electricity price subsidy cost, shortest user waiting time and highest charging facility utilization rate; s2, forming a real-time closed-loop feedback system according to the super-multi-objective optimization model and a pre-trained machine learning agent model, and calculating to obtain optimized vehicle charging space-time distribution; and S3, accessing the optimized vehicle charging space-time distribution into a navigation system, generating an electricity price subsidy cost through a linear feedback model, and guiding the vehicle to select an optimal charging strategy in combination with the navigation strategy and the electricity price subsidy cost. According to the method, five targets with conflict relations in electric vehicle charging scheduling can be efficiently optimized at the same time, reasonable distribution of charging resources and effective guidance of user behaviors are realized, and the method has real-time performance and high efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent battery swapping for electric vehicles, and specifically, relates to a method and device for spatio-temporal diversion of electric vehicle charging based on data-driven multi-objective optimization. Background Art

[0002] With the increasing demand for addressing energy shortages and environmental pollution, new energy vehicles (EVs) have been widely used. When a large number of new energy vehicles are connected to the power grid, due to the randomness and uncertainty of charging demand, grid tidal phenomena will occur, which will further lead to grid instability, difficulty in reconfiguring the distribution network, congestion in waiting for charging (swapping) vehicles, and an increase in the deployment cost of charging infrastructure. In response, many scholars have studied various methods in recent years, but all have certain limitations, as follows:

[0003] (1) Most of the existing technologies conduct local research on issues related to electric vehicle charging and rarely consider the needs of multiple parties. For example, some focus on modeling customers' responses to electricity prices but do not fully consider the needs of multiple stakeholders such as battery swapping stations, the government, and charging vehicles.

[0004] (2) The models designed by the existing technologies are relatively fixed and cannot be adjusted according to real-time data, making it difficult to meet the needs of the current time, current city, and current scenario, and cannot support all available machine learning algorithms, with poor generality and portability.

[0005] (3) The existing methods have a slow convergence rate, cannot quickly and real-time adjust optimization strategies for real-time data, and cannot face highly dynamic scenarios. Summary of the Invention

[0006] The first object of the invention is to overcome the disadvantages and deficiencies in the existing technology and provide a method for spatio-temporal diversion of electric vehicle charging based on data-driven multi-objective optimization. The present invention can simultaneously and efficiently optimize five conflicting objectives in electric vehicle charging scheduling, realize the reasonable allocation of charging resources and the effective guidance of user behavior, effectively solve the problems of uneven distribution of electric vehicle charging resources, long user waiting time, and low utilization rate of charging facilities, and has real-time performance and high efficiency.

[0007] The second object of the present invention is to provide a device for spatio-temporal diversion of electric vehicle charging based on data-driven multi-objective optimization.

[0008] The object of the present invention is achieved by the following technical solutions: The method for spatio-temporal diversion of electric vehicle charging based on data-driven multi-objective optimization includes the following steps:

[0009] S1. Construct a multi-objective optimization model that simultaneously takes into account five objectives: minimum energy load, lowest charging or battery rental cost, least electricity price subsidy cost, shortest user waiting time, and highest utilization rate of charging facilities;

[0010] S2. A real-time closed-loop feedback system is composed of a multi-objective optimization model and a pre-trained machine learning agent model to calculate the optimized spatio-temporal distribution of vehicle charging.

[0011] The real-time closed-loop feedback system specifically includes: using the pre-trained machine learning agent model to dynamically adjust the prediction result according to the output result of the multi-objective optimization model to obtain the predicted spatio-temporal distribution of vehicle charging; based on the adjusted prediction result, using the multi-objective optimization algorithm to perform real-time collaborative optimization on five objectives and output the optimized spatio-temporal distribution of vehicle charging.

[0012] S3. Connect the optimized spatio-temporal distribution of vehicle charging to the navigation system, generate the electricity price subsidy cost through a linear feedback model, and combine the navigation strategy and the electricity price subsidy cost to guide the vehicle to select the optimal charging strategy.

[0013] Preferably, in step S1, the objective functions of the multi-objective optimization model are respectively defined as follows:

[0014] (1) Minimization of energy load:

[0015]

[0016] where f1 is the minimum energy load, P grid (t) is the total load power at time t, T is the total scheduling time period;

[0017] (2) Minimization of charging or battery rental cost:

[0018]

[0019] where f2 is the lowest charging or battery rental cost, i is the vehicle number sequence number, j is the charging facility number sequence number, t is the time, n is the total number of charging vehicles, m is the total number of charging facilities, C ij (t) is the charging or battery rental unit cost, Q ij (t) is the charging amount;

[0020] (3) Minimization of electricity price subsidy cost:

[0021]

[0022] where f3 is the lowest electricity price subsidy cost, S ij (t) is the electricity price subsidy unit cost;

[0023] (4) Minimization of user waiting time:

[0024]

[0025] Among them, f4 is the minimum waiting time value, Wi is the waiting time of the i-th vehicle, ai is the arrival time, and Si is the start charging time;

[0026] (5) Maximize the utilization rate of charging facilities:

[0027]

[0028] Among them, f5 is the maximum utilization rate value of the charging facilities, Pj(t) is the actual charging power of the j-th charging facility at time t, and P j rated is the rated power;

[0029] Preferably, in step S2, the input of the machine learning agent model is time-related data and space-related data, and the output is the prediction result of the time and space distribution of the electric vehicle charging demand. The construction method includes the following steps:

[0030] Step a: Collect time-related data and space-related data and perform data preprocessing;

[0031] Step b: Model selection and construction: Build an LSTM-CNN hybrid model for optimization processing to obtain an optimized machine learning agent model;

[0032] Step c: Model training: Use historical charging demand spatio-temporal distribution data for training to obtain an initial machine learning agent model; then use real-time charging demand spatio-temporal distribution data to continuously optimize and update the initial machine learning agent model, so that the machine learning agent model can dynamically output the latest prediction results according to the decision variables output by the multi-objective optimization model.

[0033] Preferably, in step a, the time-related data includes charging records, external time factors and their corresponding time series matrix A and external time factor vector

[0034] The space-related data includes the geographical location information of the charging facilities, vehicle driving trajectory data, geospatial data and their corresponding geographical location matrix G, vehicle driving trajectory matrix H and regional geospatial matrix S.

[0035] Preferably, in step a, the data preprocessing specifically includes:

[0036] Convert the time data in the time series matrix A into a unified format, and extract features according to the converted time series matrix A and the external time factor vector to obtain a new feature vector

[0037] Convert the coordinate data of the geographical location matrix G to the same coordinate system, and perform quantization processing on the vehicle driving trajectory matrix H and the regional geographical space matrix S.

[0038] Preferably, step b specifically includes the following steps:

[0039] Construct the LSTM network part, including an input layer, a hidden layer, and an output layer. Among them, the preprocessed time-related data is input into the LSTM network, the basic unit formula of the LSTM network is used to calculate the hidden state, and the output of the LSTM network is used as the preliminary prediction result of the time distribution.

[0040] Construct the CNN network part, including an input layer, a convolutional layer, a pooling layer, and an output layer. Among them, the preprocessed space-related data is constructed into a spatial data matrix, convolution operations are performed using convolutional kernels to extract spatial features, pooling operations are performed on the output of the convolutional layer to reduce the data dimension, and the output of the CNN network is used as the preliminary prediction result of the spatial distribution.

[0041] Model optimization includes replacing LSTM with the gated recurrent unit GRU and temporal separable convolution, and replacing CNN with depthwise separable convolution and channel attention mechanism; and adopting a distributed computing strategy, that is, the central node continuously operates, and the edge nodes use lightweight offline models for calculation and regularly update the model from the central node.

[0042] Construct the fusion part, including fusing the output of the LSTM and the output of the CNN. The fused result undergoes a fully connected layer operation to generate the final prediction results of the time and space distribution of the electric vehicle charging demand.

[0043] Preferably, in step S2, the real-time collaborative optimization of the five objectives by using the multi-objective optimization algorithm specifically refers to: solving by using the multi-objective optimization algorithm based on the adaptive reference line hybrid projection distance, including the following steps:

[0044] S21. Initialization: Randomly generate the initial population P t , initialize the reference vector set R, and set the maximum number of evaluations t max as the termination condition;

[0045] S22. Evolution loop: Generate the offspring population through crossover and mutation operations, and merge the parent population P t and the offspring population Q t to form a new population; establish an evenly distributed initialization reference line, and associate the population individuals to the nearest reference line; calculate the fitness and select the population individuals with the best fitness; adaptively adjust the reference line according to the association relationship.

[0046] S23. Termination and output: When the maximum number of evaluations t is reachedmax , terminate the evolutionary cycle and return the final population, i.e., obtain the optimized solution set.

[0047] Preferably, step S22 specifically includes:

[0048] (1) Fitness calculation based on the hybrid projection distance:

[0049]

[0050] where φ(p i ) is the fitness function, the first term represents the projection distance of the solution p i on the reference line ri; the second term represents the global distance from the solution p i to the minimum coordinate point zmin of all solutions, multiplied by a weight t is the iteration time, t max is the upper limit of the evaluation times, M is the number of objectives; the third term represents the perpendicular distance from the solution p i to the reference line ri;

[0051] (2) Adaptive reference line adjustment:

[0052] Sparse reference line splitting: If the number of solutions associated with the reference line r i is less than the average value, i.e., generate a new reference line r new around it, and the formula is:

[0053]

[0054] where r is a random vector and M is the number of objectives;

[0055] Dense reference line merging: If the number of solutions associated with the reference line r i is greater than the average value, i.e., then select the reference line r j with the fewest associated solutions among its adjacent reference lines for merging, and the formula is:

[0056]

[0057] Preferably, in step S3, the linear feedback model is:

[0058]

[0059] where s0 is the benchmark subsidy number and s0 ≥ 0, N ij (t) is the actual charging demand, which is predicted by the machine learning proxy model.

[0060] Preferably, step S3 specifically includes:

[0061] Integrate the spatial distribution in the optimized spatio-temporal distribution of vehicle charging into the navigation system to real-time guide the vehicle to select the optimal charging route to reach the nearest and least occupied charging facility for charging; and generate the electricity price subsidy cost through a linear feedback model, and guide the vehicle to select the optimal charging time according to the electricity price subsidy cost.

[0062] A data-driven spatio-temporal diversion device for electric vehicle charging with multi-objective optimization, comprising:

[0063] A first construction module for constructing a multi-objective optimization model with the minimum energy load, the lowest charging or battery rental cost, the least electricity price subsidy cost, the shortest user waiting time, and the highest utilization rate of charging facilities;

[0064] A second construction module for forming a real-time closed-loop feedback system according to the multi-objective optimization model and a pre-trained machine learning agent model, and calculating the optimized spatio-temporal distribution of vehicle charging,

[0065] The real-time closed-loop feedback system specifically includes: using the pre-trained machine learning agent model to dynamically adjust the prediction result according to the output result of the multi-objective optimization model to obtain the predicted spatio-temporal distribution of vehicle charging; based on the adjusted prediction result, using the multi-objective optimization algorithm to perform real-time collaborative optimization on the five objectives, and output the optimized spatio-temporal distribution of vehicle charging;

[0066] A guiding module for integrating the optimized spatio-temporal distribution of vehicle charging into the navigation system, and generating the electricity price subsidy cost through a linear feedback model, and guiding the vehicle to select the optimal charging strategy according to the electricity price subsidy cost.

[0067] The present invention has the following advantages and effects compared with the prior art:

[0068] (1) The present invention provides a data-driven spatio-temporal diversion method for electric vehicle charging with multi-objective optimization. First, a multi-objective optimization model (five-objective optimization model) is constructed, taking into account five conflicting objectives including energy load, charging or battery rental cost, electricity price subsidy cost, user waiting time, and charging facility utilization rate. It has the advantages of being more detailed and comprehensive, and being able to adapt to more complex scenarios (including comprehensively considering the balance and interest requirements of multiple conflicting objectives of charging facility operators, the government, users, etc.), overcoming the drawback of the prior art lacking consideration of multiple-party needs. By combining the multi-objective optimization model and a machine learning surrogate model to form a real-time closed-loop feedback system, it can effectively optimize the five objectives in electric vehicle charging scheduling, achieve reasonable allocation of charging resources and effective guidance of user behavior, and effectively solve problems such as uneven distribution of electric vehicle charging resources, long user waiting time, and low charging facility utilization rate, providing technical support for the popularization of electric vehicles and the intelligent management of charging networks.

[0069] (2) The present invention constructs a machine learning surrogate model using a machine learning method based on big data (LSTM-CNN hybrid model). By optimizing on the basis of the LSTM plus CNN architecture, including replacing LSTM with a gated recurrent unit GRU plus temporal separable convolution, which can reduce the number of LSTM parameters, and replacing CNN with depthwise separable convolution plus channel attention mechanism, which can reduce the computational amount, enabling this machine learning surrogate model to be embedded in the multi-objective evolutionary algorithm, providing a basis for the continuous and rapid optimization of the multi-objective evolutionary algorithm. At the same time, a distributed computing strategy is also adopted, that is, the method of central node (cloud) computing and using the distilled light offline model on the edge side, so that each edge side (charging facility) can achieve continuous optimization only relying on a low-computing power module. This machine learning surrogate model can real-time predict the spatio-temporal distribution requirements of charging for all vehicles accessing the system according to the latest input data, and has the characteristics of strong practicability and wide application range.

[0070] (3) The present invention uses a multi-objective evolutionary optimization algorithm based on adaptive reference line hybrid projection distance, which can quickly converge in a complex high-dimensional objective space, thus realizing real-time and continuous optimization iteration of the multi-objective optimization model, being able to adapt to high-dynamic scenarios (providing the best charging time, charging route, and charging facilities for vehicle owners continuously for 24 hours, and providing dynamic and currently most suitable charging strategies and electricity price subsidy plans for charging facility operators or the government), ensuring real-time performance and high efficiency, and overcoming the drawbacks of slow convergence speed and difficulty in dealing with high-dynamic scenarios in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 It is a schematic flow chart of the data-driven spatio-temporal diversion method for electric vehicle charging with multi-objective optimization of the present invention.

[0072] Figure 2 This is a logical schematic diagram of the data-driven spatio-temporal diversion method for electric vehicle charging with multi-objective optimization of the present invention.

[0073] Figure 3 This is a schematic diagram of the spatio-temporal distribution of charging demand based on machine learning and the real-time optimization framework of "cloud-edge-end" integration of the present invention. Detailed implementation manners

[0074] The present invention will be further described in detail below in conjunction with embodiments and the accompanying drawings, but the implementation manners of the present invention are not limited thereto.

[0075] Embodiment 1

[0076] As Figure 1 shown is a flow schematic diagram of the data-driven spatio-temporal diversion method for electric vehicle charging with multi-objective optimization, including the following steps:

[0077] S1. Construct a multi-objective optimization model that simultaneously takes into account five objectives: minimum energy load, lowest charging or battery rental cost, least electricity price subsidy cost, shortest user waiting time, and highest charging facility utilization rate;

[0078] S2. According to the multi-objective optimization model and the pre-trained machine learning agent model, form a real-time closed-loop feedback system, and calculate the optimized spatio-temporal distribution of vehicle charging;

[0079] The real-time closed-loop feedback system specifically includes: using the pre-trained machine learning agent model to dynamically adjust the prediction result according to the output result of the multi-objective optimization model to obtain the predicted spatio-temporal distribution of vehicle charging; based on the adjusted prediction result, using the multi-objective optimization algorithm to perform real-time collaborative optimization on the five objectives, and output the optimized spatio-temporal distribution of vehicle charging;

[0080] S3. Connect the optimized spatio-temporal distribution of vehicle charging to the navigation system, generate the electricity price subsidy cost through a linear feedback model, and combine the navigation strategy and the electricity price subsidy cost to guide the vehicle to select the optimal charging strategy.

[0081] Specifically, the present invention provides a data-driven spatio-temporal diversion method for electric vehicle charging with multi-objective optimization. First, a multi-objective optimization model (five-objective optimization model) is constructed, which simultaneously takes into account five conflicting objectives: energy load, charging or battery rental cost, electricity price subsidy cost, user waiting time, and charging facility utilization rate. It has the advantages of being more detailed and comprehensive and being able to adapt to more complex scenarios (including comprehensively considering the balance and interest needs of multiple conflicting objectives of charging facility operators, the government, users, etc.), and overcomes the shortcomings of the prior art lacking consideration of multiple-party needs.

[0082] Specifically, the present invention incorporates the spatial distribution in the optimized spatio-temporal distribution of vehicle charging into the navigation system, which can guide vehicles in real time to select the optimal charging route to reach the nearest and least occupied charging facility or charging station for charging, so as to avoid congestion caused by a large number of vehicles flocking into the same charging station. Additionally, the present invention generates the electricity price subsidy cost through a linear feedback model and guides vehicles to select the optimal charging time according to the electricity price subsidy cost, thereby preventing a large number of vehicles from charging at the same time, which may lead to congestion at the charging station and overloading of the power grid.

[0083] This multi-objective optimization model can guide the navigation system to direct vehicle owners to charge at the optimal charging facility at the most suitable time. Meanwhile, it also guides vehicle owners to select the optimal charging facility at the most appropriate time through the charging subsidy policies of each charging station. These two spatio-temporal guidance methods not only consider the maximization of vehicle owners' interests but also the minimization of costs for the charging facilities (private) or the government (state-owned) that provide subsidies.

[0084] In addition, as Figure 2 shown, by combining the multi-objective optimization model and the machine learning agent model to form a real-time closed-loop feedback system, the present invention can predict and dynamically adjust the spatio-temporal distribution data in real time and continuously, effectively optimizing the five conflicting objectives in electric vehicle charging scheduling, realizing the reasonable allocation of charging resources and the effective guidance of user behavior, and effectively solving problems such as uneven distribution of electric vehicle charging resources, long user waiting times, and low utilization rates of charging facilities, providing technical support for the popularization of new energy vehicles and the intelligent management of the charging network.

[0085] In step S1, the objective functions of the multi-objective optimization model are defined as follows:

[0086] (1) Minimization of energy load:

[0087]

[0088] Among them, f1 is the minimum energy load, and P grid (t) is the total load power at time t, and T is the total scheduling time period;

[0089] (2) Minimization of charging or battery rental cost:

[0090]

[0091] Among them, f2 is the lowest charging or battery rental cost, i is the serial number of the vehicle, j is the serial number of the charging facility, t is the time, n is the total number of charging vehicles, m is the total number of charging facilities, C ij (t) is the unit cost of charging or battery rental, and Q ij (t) is the charging amount;

[0092] (3) Minimization of electricity price subsidy cost:

[0093]

[0094] Among them, f3 is the minimum electricity price subsidy cost, and S ij (t) is the unit cost of electricity price subsidy; S ij (t) is calculated according to the linear feedback model, and the expression of this linear feedback model is:

[0095]

[0096] Among them, s0 is the benchmark subsidy number and s0≥0, and N ij (t) is the actual charging demand, which is predicted by the machine learning proxy model.

[0097] (4) Minimization of user waiting time:

[0098]

[0099] Among them, f4 is the shortest waiting time value, and W i is the waiting time of the i-th vehicle, and a i is the arrival time, and S i is the start charging time;

[0100] (5) Maximization of charging facility utilization rate:

[0101]

[0102] Among them, f5 is the maximum value of charging facility utilization rate, and P j (t) is the actual charging power of the j-th charging facility at time t, and P j rated is the rated power,

[0103] In step S2, the input of the machine learning proxy model is time-related data and space-related data, and the output is the prediction result of the time and space distribution of electric vehicle charging demand. The construction method includes the following steps:

[0104] Step a: Collect time-related data and space-related data and perform data preprocessing;

[0105] Step a specifically includes the following process:

[0106] (1) Modeling data collection:

[0107] Time-related data collection: The time-related data includes charging records and external time factors, and corresponding time series matrices A and external time factor vectors are constructed respectively with the charging records and external time factors

[0108] Specifically, in this embodiment, by obtaining charging records from the database of the charging operator and organizing them into a time series matrix A ∈ R n×2 , where n is the number of charging records, t si represents the start time of the i-th charging, and t ei represents the end time of the i-th charging.

[0109] Obtain seasonal information from the meteorological department, determine weekdays / weekends from calendar data, and integrate these external time factors into a vector For example s represents the season (numerical codes 1 - 4 represent spring, summer, autumn, and winter), w represents weekdays / weekends (0 for weekends, 1 for weekdays), and h represents the time period of the day (0 - 23 represents hours).

[0110] Spatial - related data collection: The spatial - related data includes the geographical location information of charging facilities, vehicle driving trajectory data, and geospatial data. Corresponding geographical location matrix G, vehicle driving trajectory matrix H, and regional geospatial matrix S are constructed respectively with the geographical location information, vehicle driving trajectory data, and geospatial data.

[0111] Specifically, in this embodiment, obtain the charging location and route data of the vehicle owner from an electronic map provider; obtain the geographical location information of the charging facilities from the charging facility operation platform, and construct the geographical location matrix G, that is, the coordinate matrix G ∈ R k×2 , where k is the number of charging facilities, x i and y i are the longitude and latitude coordinates of the i - th charging facility respectively.

[0112] Obtain vehicle driving trajectory data through in - vehicle devices or relevant traffic data platforms and organize it into a vehicle driving trajectory matrix, that is, a three - dimensional matrix H ∈ R l×2×p , where l is the number of vehicle driving trajectory points, 2 represents the coordinate dimensions (longitude and latitude), p is the number of vehicles, and H i,j,k represents the j - th coordinate (j = 1 represents longitude, j = 2 represents latitude) of the i - th driving trajectory point of the k - th vehicle.

[0113] Obtain geospatial data such as population density and land use type of the region from the government geographical information department and construct the regional geospatial matrix S ∈ R q×r , where q is the number of regions, r is the number of types of geospatial data, such as d i1 represents the population density of the i - th region, d i2Indicates the land use type of the i-th region (numerical codes 1-3 represent commercial areas, residential areas, and industrial areas).

[0114] (2) Data preprocessing

[0115] Time-related data preprocessing: Convert the time data in the time series matrix A into a unified timestamp format or an appropriate relative time scale; perform feature engineering on the time data, for example, extract more features from and the transformed time series matrix A, such as calculating the duration t of each charge ei -t si , and combine it with external time factors to form a new feature vector to better represent the time pattern.

[0116] Spatial-related data preprocessing: Perform necessary coordinate conversions on the coordinate data in the geographical location matrix G (if the coordinate systems are inconsistent) to ensure that all coordinates are in the same coordinate system; perform quantization processing on the regional geographical space matrix S and the vehicle travel trajectory matrix H. For example, divide the geographical area into grids, and count information such as the number of charging facilities in each grid (according to G), the number of vehicle trajectory points (according to H), etc., and reconstruct a more suitable spatial data representation form for model input.

[0117] Step b, Model selection and construction: Build an LSTM-CNN hybrid model and perform optimization processing to obtain an optimized machine learning proxy model;

[0118] Step b specifically includes the following processes:

[0119] (1) Build the LSTM network part, including the input layer, hidden layer, and output layer:

[0120] Build the input layer, combine the preprocessed time-related data (including features extracted from charging records and external time factors) into an input vector and input it into the LSTM network;

[0121] The LSTM network contains multiple LSTM units, and the calculation of its hidden layer is based on the basic unit formula of LSTM. For example, the forget gate input gate cell state update output gate h t =o t ⊙tanh(C t ), where W f 、b f 、W i 、b i 、W C 、b C, W o , b o are the corresponding weight matrix and bias vector respectively, and σ is the sigmoid function h t-1 is the hidden state at the previous moment, is the input time feature vector at time t;

[0122] After calculations over multiple time steps, the output y of the LSTM LSTM can be the hidden state h at the last moment T (assuming the length of the time series is T) or the result after passing through a fully connected layer. This output will be used as the preliminary prediction result of the time distribution.

[0123] (2) Construct the CNN network part, including the input layer, convolutional layer, pooling layer, and output layer:

[0124] Construct the input layer, and construct the spatial data matrix X ∈ R H×W×D from the preprocessed spatially related data (such as information on the density of charging facilities and population density after gridification), where H and W represent the height and width of the spatial data matrix (e.g., the number of rows and columns of the gridified geographical area), and D represents the number of channels (e.g., different spatial features);

[0125] The CNN part includes a convolutional layer, a pooling layer, etc. For the convolutional layer calculation: Let the convolutional kernel K ∈ R k×k×D×M , where k is the size of the convolutional kernel and M is the number of convolutional kernels (number of output channels). The convolution operation where Y ∈ R H′×W′×M is the output of the convolutional layer, H′ = H - k + 1, W′ = W - k + 1, and b m is the bias of the m-th output channel. After operations such as passing through a pooling layer (e.g., max pooling), the final CNN output y CNN is obtained. This output will be used as the preliminary prediction result of the spatial distribution.

[0126] (3) Optimization of the LSTM-CNN model:

[0127] The LSTM-CNN model may have a large computational amount or slow speed. Therefore, replace the LSTM with a gated recurrent unit GRU and temporal separable convolution to reduce the number of parameters; replace the CNN with depthwise separable convolution and channel attention mechanism to reduce the computational amount;

[0128] Meanwhile, a distributed computing strategy is also adopted, i.e., a time-division strategy where the central node continuously operates and the edge nodes use lightweight offline models. The edge side (such as the computing units of each charging facility) adopts lightweight offline models and regularly updates the distilled (or pruned) lightweight offline models from the central node to ensure the currency and consistency of the models. According to this distributed computing strategy, a "cloud-edge-end" integrated real-time optimization framework is constructed, as Figure 3 shown. Among them, the cloud is responsible for all model calculations and iterations, and regularly distributes the distilled offline models to the corresponding units (charging facilities). The edge side is responsible for regularly uploading the latest data and feeding back to the cloud central model, and regularly downloading the offline models to achieve fast inference and optimization with low cost at the terminal. The terminal is responsible for regularly uploading updated spatio-temporal data and downloading the optimal charging strategy to the navigation system or charging APP, etc. This data-driven real-time optimization "cloud-edge-end" hierarchical deployment framework enables each edge side to continuously optimize only relying on low-computing-power modules.

[0129] (4) Construction of the fusion part:

[0130] Fuse the output y LSTM of LSTM and the output y CNN of CNN. For example, use a simple concatenation method. The fused output y = [y LSTM , y CNN . This fused result will further go through operations such as a fully connected layer to obtain the final prediction result Y on the temporal and spatial distribution of the charging demand of electric vehicles.

[0131] Step c, model training: Use historical spatio-temporal distribution data of charging demand for training to obtain an initial machine learning proxy model; then use real-time spatio-temporal distribution data of charging demand to continuously optimize and update the initial machine learning proxy model, so that the machine learning proxy model can dynamically output the latest prediction results according to the decision variables output by the multi-objective optimization model (such as the electricity price subsidy cost of all charging stations at time t, the charging station locations recommended for each vehicle at time t).

[0132] Specifically, in this embodiment, the training of the core large model is deployed in the cloud to achieve global model iteration at the minute level. The edge side performs local updates of the model according to regional charging demand and traffic conditions to achieve low energy consumption and a model update speed at the second level. The terminal side, i.e., the terminal (vehicle, charging station), updates the latest offline model in real time on demand through the high-speed network 5G to provide terminal results. The "cloud-edge-end" integrated deployment enables the model and method to adapt to the real-time changes in vehicle positions and charging demands in high-dynamic scenarios with lower energy consumption and computing requirements.

[0133] Specifically, in this embodiment, the optimized model architecture has a time feature extraction part (GRU plus time-domain separable convolution reduces the computational complexity of time feature extraction), a spatial feature extraction part (depthwise separable convolution plus channel attention mechanism reduces the computational complexity of spatial feature extraction), a feature fusion part, and a distributed deployment strategy (the distributed computing strategy combined with model distillation ensures the real-time inference ability of edge nodes). Therefore, the above optimization method can significantly reduce the computational complexity and resource consumption of the LSTM-CNN model, while improving the real-time performance and practicality of the model.

[0134] Furthermore, after model training, the final model, i.e., the pre-trained machine learning agent model, can predict the spatio-temporal distribution demand of the charging of all vehicles accessing the system based on the latest time-related data and space-related data. This demand is real-time and spatio-temporal, so it can adapt to different scales of cities, special scenarios such as the Spring Festival travel rush holidays, and changes in traffic and charging demands caused by seasons, and has broad application prospects.

[0135] In step S2, the real-time collaborative optimization of the five objectives by using the multi-objective optimization algorithm specifically refers to: using the multi-objective evolutionary optimization algorithm based on the adaptive reference line hybrid projection distance for optimization, including the following steps:

[0136] S21. Initialization: Randomly generate an initial population P t (the population size is S), initialize the reference vector set R = {r1, r2,..., r S} and set the maximum number of evaluations t max as the termination condition;

[0137] S22. Evolution loop:

[0138] 1) Generate an offspring population through crossover and mutation operations, and merge the parent population P t and the offspring population Q t to form a new population;

[0139] 2) Establish an initialized reference line with a uniform distribution and associate population individuals to the nearest reference line:

[0140] For each population individual, calculate its distance from each reference quantity, associate each population individual to the nearest reference line, and record the association relationship;

[0141] 3) Fitness calculation and selection:

[0142] That is, for each reference line, calculate the hybrid projection distance fitness of the population individuals associated with it. The fitness calculation based on the hybrid projection distance is:

[0143]

[0144] Among them, φ(p i ) is the fitness function. The first term represents the projection distance of the solution p i on the reference line r i ; The second term represents the global distance from the solution p i to the minimum coordinate point z min of all solutions, multiplied by a weight t is the iteration time, t max is the upper limit of the evaluation times, and M is the number of objectives; The third term represents the vertical distance from the solution p i to the reference line r i .

[0145] Select the population individuals with the optimal fitness to join the next-generation population;

[0146] 4) Adaptive reference line adjustment: That is, adaptively adjust the reference line according to the correlation between the current population and the reference line:

[0147] Sparse reference line splitting: If the number of solutions associated with the reference line r i is less than the average value, that is generate a new reference line r new around it. The formula is:

[0148]

[0149] where r is a random vector and M is the number of objectives;

[0150] Dense reference line merging: If the number of solutions associated with the reference line r i is greater than the average value, that is then select the reference line r j with the fewest associated solutions among its adjacent reference lines for merging. The formula is:

[0151]

[0152] S23. Termination and output: When the maximum evaluation times t max is reached, terminate the evolutionary loop and return the final population, that is, obtain the optimized solution set.

[0153] Specifically, the present invention uses a multi-objective evolutionary optimization algorithm based on the mixed projection distance of adaptive reference lines, which can quickly converge in a complex high-dimensional objective space, thereby enabling the real-time and continuous optimization iteration of the multi-objective optimization model, and can adapt to high-dynamic scenarios (providing the best charging time, charging route, and charging facilities for vehicle owners continuously for 24 hours, and providing dynamic and currently most suitable charging strategies and electricity price subsidy schemes for charging facility operators or the government), ensuring real-time performance and high efficiency, and overcoming the disadvantages of slow convergence speed and difficulty in coping with high-dynamic scenarios in the prior art.

[0154] Embodiment 2

[0155] A data-driven spatio-temporal diversion device for electric vehicle charging with multi-objective optimization, comprising:

[0156] A first construction module for constructing a multi-objective optimization model that simultaneously takes into account five objectives: minimum energy load, lowest charging or battery rental cost, least electricity price subsidy cost, shortest user waiting time, and highest charging facility utilization rate;

[0157] A second construction module for forming a real-time closed-loop feedback system according to the multi-objective optimization model and a pre-trained machine learning agent model, and calculating the optimized spatio-temporal distribution of vehicle charging;

[0158] The real-time closed-loop feedback system specifically includes: using the pre-trained machine learning agent model to dynamically adjust the prediction result according to the output result of the multi-objective optimization model to obtain the predicted spatio-temporal distribution of vehicle charging; based on the adjusted prediction result, using the multi-objective optimization algorithm to perform real-time collaborative optimization on the five objectives, and outputting the optimized spatio-temporal distribution of vehicle charging;

[0159] A guiding module for accessing the optimized spatio-temporal distribution of vehicle charging into the navigation system, generating the electricity price subsidy cost through a linear feedback model, and guiding the vehicle to select the optimal charging strategy in combination with the navigation strategy and the electricity price subsidy cost.

[0160] The above embodiments are preferred embodiments of the present invention and cannot limit the present invention. Any other changes or other equivalent replacement methods that do not deviate from the technical solution of the present invention are included in the protection scope of the present invention.

Claims

1. A method for spatio-temporal diversion of electric vehicle charging driven by data for multi-objective optimization, characterized in that, It includes the following steps: S1. Construct a multi-objective optimization model that takes into account five objectives simultaneously, namely, minimizing the energy load, minimizing the charging or battery rental cost, minimizing the electricity price subsidy cost, minimizing the user waiting time, and maximizing the utilization rate of charging facilities; S2. According to the multi-objective optimization model and the pre-trained machine learning agent model, form a real-time closed-loop feedback system to calculate the optimized spatio-temporal distribution of vehicle charging; The real-time closed-loop feedback system specifically includes: using the pre-trained machine learning agent model to dynamically adjust the prediction result according to the output result of the multi-objective optimization model to obtain the predicted spatio-temporal distribution of vehicle charging; Based on the adjusted prediction result, use the multi-objective optimization algorithm to perform real-time collaborative optimization on the five objectives and output the optimized spatio-temporal distribution of vehicle charging; S3. Connect the optimized spatio-temporal distribution of vehicle charging to the navigation system, generate the electricity price subsidy cost through a linear feedback model, and combine the navigation strategy and the electricity price subsidy cost to guide the vehicle to select the optimal charging strategy.

2. The data-driven spatio-temporal diversion method for electric vehicle charging with ultra-many objectives optimization according to claim 1, wherein, In step S1, the objective functions of the multi-objective optimization model are respectively defined as follows: (1) Minimizing the energy load: Among them, f1 is the minimum energy load, and P grid (t) is the total load power at time t, T is the total scheduling time period; (2) Minimizing the charging or battery rental cost: Among them, f2 is the lowest charging or battery rental cost, i is the vehicle number sequence number, j is the charging facility number sequence number, t is the time, n is the total number of charging vehicles, m is the total number of charging facilities, C ij (t) is the charging or battery rental unit cost, Q ij (t) is the charging amount; (3) Minimizing the electricity price subsidy cost: Among them, f3 is the lowest electricity price subsidy cost, and S ij (t) is the unit cost of electricity price subsidy; (4) Minimizing the user waiting time: Among them, f4 is the shortest waiting time value, and W i is the waiting time of the i-th vehicle, and a i is the arrival time, and S i is the start charging time; (5) Maximizing the utilization rate of charging facilities: Among them, f5 is the maximum value of the charging facility utilization rate, and P j (t) is the actual charging power of the j-th charging facility at time t, is the rated power, 3. The data-driven spatio-temporal diversion method for optimizing electric vehicle charging with multiple objectives according to claim 1, wherein In step S2, the input of the machine learning agent model is time-related data and space-related data, and the output is the prediction result of the spatio-temporal distribution of electric vehicle charging demand. The construction method includes the following steps: Step a. Collect time-related data and space-related data and perform data preprocessing; Step b. Model selection and construction: Build an LSTM-CNN hybrid model for optimization processing to obtain the optimized machine learning agent model; Step c. Model training: Use the historical spatio-temporal distribution data of charging demand for training to obtain the initial machine learning agent model; then use the real-time spatio-temporal distribution data of charging demand to continuously optimize and update the initial machine learning agent model, so that the machine learning agent model can dynamically output the latest prediction result according to the decision variables output by the multi-objective optimization model.

4. The method for spatio-temporal diversion of electric vehicle charging based on data-driven multi-objective optimization according to claim 3, characterized in that In step a, the time-related data includes charging records, external time factors, and their corresponding time series matrix A and external time factor vectors The space-related data includes the geographical location information of charging facilities, vehicle driving trajectory data, geospatial data, and their corresponding geographical location matrix G, vehicle driving trajectory matrix H, and regional geospatial matrix S.

5. The method for spatio-temporal diversion of electric vehicle charging in data-driven multi-objective optimization according to claim 4, characterized in that In step a, the data preprocessing specifically includes: Convert the time data in the time series matrix A into a unified format, and extract features based on the converted time series matrix A and the external time factor vector to obtain a new feature vector Converting the coordinate data of the geographical location matrix G to the same coordinate system, and performing quantization processing on the vehicle driving trajectory matrix H and the regional geospatial matrix S.

6. The method for spatio-temporal diversion of electric vehicle charging based on data-driven multi-objective optimization according to claim 4, characterized in that Step b specifically includes the following steps: Construct the LSTM network part, including an input layer, a hidden layer, and an output layer. Among them, the preprocessed time-related data is input into the LSTM network, and the basic unit formula of the LSTM network is used to calculate the hidden state, and the output of the LSTM network is used as the preliminary prediction result of the time distribution; The part of constructing the CNN network includes an input layer, a convolutional layer, a pooling layer, and an output layer. Among them, the preprocessed spatially related data is constructed into a spatial data matrix, convolution operations are performed using convolution kernels to extract spatial features, pooling operations are performed on the output of the convolutional layer to reduce the data dimension, and the output of the CNN network is used as the preliminary prediction result of the spatial distribution; Model optimization includes replacing LSTM with a gated recurrent unit GRU and temporal separable convolution, and replacing CNN with depthwise separable convolution and channel attention mechanism; and adopting a distributed computing strategy, that is, the central node continuously operates, and the edge nodes use lightweight offline models for calculation and update the model from the central node regularly; The part of constructing the fusion includes fusing the output of LSTM and the output of CNN, and the fused result is operated through a fully connected layer to generate the final prediction result of the temporal and spatial distribution of the electric vehicle charging demand.

7. The method for spatio-temporal diversion of electric vehicle charging based on data-driven multi-objective optimization according to claim 1, characterized in that In step S2, the specific meaning of using the multi-objective optimization algorithm to perform real-time collaborative optimization on five objectives is: solving using the multi-objective optimization algorithm based on the adaptive reference line hybrid projection distance, including the following steps: S21. Initialization: Randomly generate the initial population P t , initialize the reference vector set R, and set the maximum number of evaluations t max as the termination condition; S22. Evolution cycle: It includes generating an offspring population through crossover and mutation operations, merging the parent population P t and the offspring population Q t to form a new population; establishing an initialization reference line with a uniform distribution and associating population individuals to the nearest reference line; calculating fitness and selecting the population individuals with the optimal fitness; adaptively adjusting the reference line according to the association relationship; S23. Termination and Output: When the maximum number of evaluations t is reached max , the evolutionary loop is terminated, and the final population is returned, i.e., the optimized solution set is obtained.

8. The data-driven spatio-temporal drainage method for ultra-multi-objective optimization of electric vehicle charging according to claim 7, wherein Step S22 specifically includes: (1) Fitness calculation based on the hybrid projection distance: Among them, φ(p i ) is the fitness function. The first term represents the projection distance of the solution p i on the reference line r i . The second term represents the global distance from the solution p i to the minimum coordinate point z min of all solutions, multiplied by a weight t is the iteration time, t max is the upper limit of the evaluation times, and M is the number of objectives. The third term represents the vertical distance from the solution p i to the reference line r i . (2) Adaptive reference line adjustment: Sparse reference line splitting: If the number of solutions associated with the reference line r i is less than the average, that is new reference lines r are generated around it new , and the formula is: Among them, r is a random vector, and M is the number of objectives; Dense reference line merging: If the number of solutions associated with reference line r i is greater than the average, that is then select the reference line r with the fewest associated solutions among its adjacent reference lines j for merging. The formula is:

9. The method for spatio-temporal diversion of electric vehicle charging based on data-driven multi-objective optimization according to claim 2, characterized in that, In step S3, the linear feedback model is: Among them, s0 is the benchmark subsidy number and s0 ≥ 0, N ij (t) is the actual charging demand, which is predicted by the machine learning proxy model.

10. The data-driven spatio-temporal diversion method for electric vehicle charging with ultra-many objective optimization according to claim 1, characterized in that, Step S3 specifically includes: Connect the spatial distribution in the optimized vehicle charging spatio-temporal distribution to the navigation system to guide the vehicle to select the optimal charging route in real time to reach the nearest and least occupied charging facility for charging; and generate the electricity price subsidy cost through the linear feedback model, and guide the vehicle to select the optimal charging time according to the electricity price subsidy cost.

11. An electric vehicle charging spatio-temporal diversion device for data-driven multi-objective optimization, characterized in that, It includes: The first construction module is used to construct a multi-objective optimization model with the minimum energy load, the lowest charging or battery rental cost, the least electricity price subsidy cost, the shortest user waiting time, and the highest charging facility utilization rate; The second construction module is used to form a real-time closed-loop feedback system according to the multi-objective optimization model and the pre-trained machine learning agent model, and calculate the optimized vehicle charging spatio-temporal distribution, The real-time closed-loop feedback system specifically includes: using the pre-trained machine learning agent model to dynamically adjust the prediction result according to the output result of the multi-objective optimization model to obtain the predicted vehicle charging spatio-temporal distribution; based on the adjusted prediction result, using the multi-objective optimization algorithm to perform real-time collaborative optimization on five objectives, and output the optimized vehicle charging spatio-temporal distribution; The guiding module is used to connect the optimized vehicle charging spatio-temporal distribution to the navigation system, and generate the electricity price subsidy cost through the linear feedback model, and guide the vehicle to select the optimal charging strategy according to the electricity price subsidy cost.

Images