Vehicle-pile-network multi-space-time regulation potential quantification method based on space-time diagram

Abstract

The invention specifically discloses a vehicle-pile-network multi-space-time regulation potential quantification method based on a space-time diagram, and relates to the technical field of smart power grids. The method comprises the following steps: S1, based on a Monte Carlo simulation method, establishing a city-level vehicle-pile-network interactive operation base line under constraint conditions; s2, potential index parameters are calculated and adjusted, key factors influencing potential parameters are screened by means of Pearson's correlation coefficients, and data complexity is reduced; and S3, constructing a multi-space-time adjustment potential quantitative evaluation model based on the space-time diagram neural network STGNN, and realizing adjustment potential prediction in combination with an online learning framework. The method can sense the adjustment potential of the vehicle-pile-network system in all directions, improves the operation digitization level and scheduling flexibility of the power grid, enhances the operation stability of the power grid, and provides technical support for the construction of a novel power system.

Description

Technical Field

[0001] This invention relates to the field of smart grid technology, and in particular to a method for quantifying the multi-temporal and spatial regulation potential of vehicle-pile-grid based on spatiotemporal maps. Background Technology

[0002] With the advancement of global energy transition goals, the number of electric vehicles (EVs) has experienced explosive growth, and their large-scale charging behavior has become a key factor affecting the safe and stable operation of urban power distribution networks. On the one hand, EV charging is characterized by uneven spatial and temporal distribution and large fluctuations in single power output. During peak hours (such as evenings after commuting), concentrated charging demand significantly overlaps with the grid's basic load, leading to problems such as transformer overload and excessive voltage deviation, directly threatening the safety of grid operation and the reliability of power supply. On the other hand, EVs are essentially "mobile energy storage units" that can feed power back into the grid through vehicle-to-grid (V2G) technology, possessing the potential to participate in ancillary services such as peak shaving, frequency regulation, and backup, making them an important distributed and flexible resource in new power systems.

[0003] However, the quantitative assessment of the regulation potential of the current vehicle-charging-network system still faces significant technical bottlenecks: First, existing assessment systems mostly focus on single scenarios (such as peak shaving or frequency regulation), lacking multi-dimensional quantitative indicators covering the entire "vehicle-charging pile-distribution network" link, making it difficult to comprehensively reflect the system's regulation capacity at different spatiotemporal scales (such as hourly, daily, and regional levels); Second, traditional assessment methods mostly rely on static models, failing to fully consider the coupled effects of dynamic variables such as environmental factors (such as temperature and humidity), user behavior (such as charging start-up SOC and time demand), and electricity price signals, resulting in significant deviations between assessment results and actual operating conditions; Third, for large-scale urban vehicle-charging-network systems, existing methods struggle to effectively handle the spatiotemporal correlation and uncertainty of massive data, failing to achieve real-time and accurate prediction of regulation potential, thus hindering the efficient utilization of flexible resources for electric vehicles. Summary of the Invention

[0004] The purpose of this invention is to propose a multi-temporal and spatial regulation potential quantification method for vehicle-pile-network based on spatiotemporal graphs. By constructing a multi-dimensional potential parameter quantification index system under various application scenarios, a precise city-level multi-temporal and spatial regulation potential quantification model for vehicle-pile-network is established, enabling comprehensive and high-precision perception and prediction of the regulation potential of electric vehicles.

[0005] To achieve the above objectives, this invention proposes a method for quantifying the multi-temporal and spatial adjustment potential of vehicle-pile-network based on spatiotemporal diagrams, the specific steps of which are as follows: Step S1: Based on the Monte Carlo simulation method, establish a city-level vehicle-pile-network interaction operation baseline under constraints. The steps are as follows: Step S11: Collect environmental data, user vehicle behavior data, charging pile data, and power grid operation data, and preprocess the collected data; Step S12: Establish constraints based on environmental factors, vehicle status, and electricity price signals; Step S13: Based on the constraints, use Monte Carlo simulation to generate the vehicle-pile-network interaction operation baseline; Step S2: Calculate the indicator parameters of the adjustment potential and use the Pearson correlation coefficient to screen the factors affecting the potential parameters; Step S3: Construct a multi-spatiotemporal regulation potential quantitative assessment model based on the spatiotemporal graph neural network STGNN, and combine it with an online learning framework to realize regulation potential prediction.

[0006] Preferably, in step S11, the preprocessing operations include missing and outlier removal, data format conversion, and data normalization. Missing values ​​are imputed using the mean imputation method, as shown in the following formula: ; in, This represents the mean of the data variable. For missing values ​​in the data, i It is an integer; Outliers are identified using the Z-score method, as shown in the formula below: ; in, Z-score value For the first in the dataset i Data points, The mean of the dataset. The standard deviation of the dataset; Normalization is calculated using the Min-Max method, as shown in the following formula: ; in, X' The result after data normalization. The original data before normalization. The maximum value of the dataset. This represents the minimum value in the dataset.

[0007] Preferably, in step S12, the mathematical constraints include: The vehicle-side charging time constraint is calculated using the following formula: ; ; in, The set charging start time and desired end time, For charging time, The environmental coefficient is the ambient temperature and humidity. For the battery's rated capacity, and These refer to the expected state of charge and the initial state of charge, respectively. This refers to the actual charging power. The battery safety power constraint is given by the following formula: ; in, Minimum charging power for battery charging. The maximum charging power allowed by the battery. This refers to the actual charging power of the battery. The total power allocation constraint for charging piles is defined by the following formula: ; in, for t Time of the first i The charging power of each charging station N This represents the total number of charging stations currently in a charging state. The total rated power for charging stations, The power consumption of auxiliary equipment in charging stations; The grid-side capacity constraint is given by the following formula: ; in, for t Time of the first j The load of each charging station j It is an integer. M This represents the total number of charging stations. This refers to the rated capacity of the transformer. The maximum load rate that can be operated; V2G reverse power supply constraint, the formula is as follows: ; in, The power required to charge electric vehicles to the grid. The maximum power that can be transmitted. The minimum charge state that allows for power transmission. Let be the state of charge of the vehicle battery at time t; The economic optimization constraint is given by the following formula: ; in, This can include time-of-use electricity pricing and charging service fees. for t Charging power at any time The set charging time period.

[0008] Preferably, in step S13, a vehicle-pile-network interactive operation baseline is generated using Monte Carlo simulation. The specific steps are as follows: Step S131: Establish a probability model based on the preprocessed data, including: a time probability model of vehicle arrival at charging pile based on Gaussian mixture model (GMM), an initial SOC probability model based on conditional Beta distribution, a climate environment model that varies over time, and a deterministic power grid price model. Step S132: Determine the baseline logic for interactive operation, including: Electric vehicle charging logic: Under the constraints of physics and basic rules, the charging strategy of the charging station is defined as instant charging, and charging starts with the maximum charging power; Grid interaction logic: Within a charging station, the total load of the charging pile group is the sum of the power of all vehicles currently charging, as shown in the following formula: ; in, The total load of the charging pile group The number of vehicles currently charging. for t Time of the first k The charging power of each charging station k It is an integer; Step S133: Generate a baseline using Monte Carlo simulation, including: Load baseline, calculated at each time point t The average total load is obtained using the following formula: ; in, As the load baseline, For the first d During the next simulation t The total load of the charging pile group at any given time. To simulate the total number of times, d It is an integer; The economic baseline is obtained by calculating the average charging cost for users across all time periods, using the following formula: ; in, As an economic baseline, Let k be the start time of the charging event. Let k be the end time of the charging event. for t The electricity price at any given time; The power grid risk baseline is obtained by statistically analyzing transformer overload probabilities, using the following formula: ; in, As a baseline for power grid risk, for > Duration, The maximum power of the power grid. To simulate the total duration.

[0009] Preferably, in step S131, the time probability model for the vehicle's arrival at the charging station based on the Gaussian Mixture Model (GMM) is as follows: ; in, The probability of reaching a charging station in time. For different normal distribution weights, It is a normal distribution function. The mean, Standard deviation; The initial SOC probability model based on the conditional Beta distribution is as follows: ; in, For Beta functions, This refers to the battery's state of charge. , Let be the shape parameter of the beta distribution.

[0010] Preferably, in step S2, the specific steps are as follows: Step S21: Define and calculate the index parameters of the adjustment potential, including adjustment capacity, adjustment depth, adjustment time and charging pile availability; the adjustment capacity includes power increase, power decrease, V2G discharge power, V2G discharge amount and adjustable power; the adjustment time includes response time, adjustable time period and V2G sustainable duration. Step S22: Use the Pearson correlation coefficient to screen factors affecting the moderating potential parameter. The formula for calculating the Pearson correlation coefficient is as follows: ; in, r The Pearson correlation coefficient is used. X It is an influencing factor. Y It is an adjustment potential indicator parameter. and These are the mathematical expectations of the two variables. and It is the variance of the two sets of variables.

[0011] The preferred formula for calculating the power increase is as follows: ; in, To increase power, For the maximum allowable power, For the current power, The actual charging power of the vehicle. The maximum permissible power for charging electric vehicles. The rated maximum output power of the charging pile, The maximum allowable additional load on the power grid to which the vehicle is connected; The formula for calculating the reduced power is as follows: ; in, To reduce power; The formula for calculating the discharge power of V2G is as follows: ; in, for, for, To support the maximum power of V2G for charging stations, The rated capacity of the electric vehicle battery, The maximum allowable discharge rate for the vehicle. This refers to the ambient temperature coefficient. The formula for calculating the discharge capacity of V2G is as follows: ; in, This refers to the discharge capacity of V2G. For the first i The state of charge of a vehicle while it is charging. The minimum charge state allowed for discharge. For battery discharge efficiency; The formula for calculating adjustable power is as follows: ; in, Adjustable power consumption; The formula for adjusting the depth is as follows: ; in, To adjust the depth, The charging power after receiving the adjustment command. This is the maximum charging power; The formula for calculating the adjustment time is as follows: ; in, For response time, For adjustable time periods, For V2G sustainability duration, This refers to the communication duration from the start of command issuance to the receipt by the charging station. The time consumed by the charging pile controller to parse commands. The response time of the vehicle's battery management system (BMS) to commands. The time spent away from the charging station. The time for connecting to the charging station. The capacity for V2G reverse power transmission to vehicles. The power of continuous discharge; The formula for calculating the availability of charging piles is as follows: ; in, For the availability of charging stations, For the first j Rated power of each charging station E This represents the total number of charging stations.

[0012] Preferably, in step S3, a multi-spatiotemporal regulation potential quantitative evaluation model is constructed based on the spatiotemporal graph neural network STGNN. The specific steps are as follows: Step S311: Define the spatiotemporal graph and node features. The spatiotemporal graph structure is as follows: G= ( V,E,A ); in, V This is a collection of nodes for each charging station. E Let be the set of edges in the graph. A for N × N The adjacency matrix; The model input is a three-dimensional tensor X input =( B,F,Z ), B For the number of nodes, F For characteristic number, Z The time window length for historical data; the model output is Y output =( B,C,T C represents the number of adjustment potential indicators; Step S312: Construct a spatial convolution module, using a three-dimensional tensor as input, to build an input matrix of historical data in the same time dimension. X t =( X 1, X 2,..., X N The graph convolution operation is performed on the input matrix, and the spectral graph method is used to apply the graph convolution operation to the spatiotemporal graph. G In the definition of the Laplace matrix ,A It is an adjacency matrix. The normalized Laplacian matrix is ​​a diagonal matrix composed of node degrees, and the formula is as follows: ; in, It is the identity matrix; The graph convolution operation is performed on the input matrix, as shown in the following formula: ; in, For convolution kernel, U It is the Laplace matrix L Fourier bases after eigenvalue decomposition Λ It is an eigenvalue diagonal matrix. For Laplace matrix, This is a transpose operation; Regarding time t Corresponding graphic signal x t Using Chebyshev polynomials The formula for solving the graph convolutional network is as follows: ; in, Let be the vector of the coefficients in the polynomial. Laplace matrix L The largest eigenvalue, This is the transformed Laplace matrix; Finally, the convolution result... Integrate into a new input matrix as H t =( H 1, H 2,..., H t ); Step S313: Construct a spatiotemporal convolution module and use a multilayer long short-term memory artificial neural network LSTM to perform convolution calculations on the time dimension.

[0013] Preferably, in step S313, convolution calculation is performed on the time dimension, and the steps are as follows: The initial importance value is calculated using the following formula: ; in, This is the initial importance value. , W d , U d For training parameters, d t-1 ands t-1 These are the output value and hidden state of the previous layer, respectively. tanh For activation functions; The initial importance scores are processed using Softmax to obtain the time importance scores, as shown in the following formula: ; in, Importance based on time; Time importance and H t By integrating the data, we obtain the key parameters for all time periods, as shown in the following formula: ; in, For context vectors; context vector Combined with the output of the previous LSTM layer, the formula is as follows: ; in, and For training parameters, This is the output vector at time t-1. Let be the input vector at time t-1. This is the context vector at time t-1; Will As input to a single LSTM unit, temporal convolutions are calculated layer by layer, as shown in the following formula: ; in, For Hadama accumulation, , , , For weights; , , , This is the bias value; For activation function, Output the value for the forget gate. Input values ​​for the input gate. The output value of the output gate. The state at time t, The state at time t-1 The output value at time t-1 The output value at time t; Output value By connecting to a fully connected layer, we obtain a quantitative prediction of the modulation potential. Based on the hourly, daily, and weekly quantitative forecasts of the adjustment potential, the quantitative assessment results of the multi-temporal and spatial adjustment potential are calculated, using the following formula: ; in, The results are a quantitative assessment of the modulation potential after the fusion of three time scales. W h , W d , W w The weights are for hourly, daily, and weekly time scales, respectively. , and These are quantitative forecasts of adjustment potential at three time scales: hourly, daily, and weekly.

[0014] Preferably, in step S3, the modulation potential prediction is achieved by combining an online learning framework, and the steps are as follows: Step S321: Quantitatively evaluate the output modulation potential of historical data based on the STGNN model. Conduct training and assessment; Step S322: Evaluate the new real-time data using incremental fine-tuning, as shown in the following formula: ; in, These are the updated top-level network parameters in the STGNN model. These are the top-level network parameters in the model before the update. For learning rate, The gradient of the loss function. This is the loss function.

[0015] Therefore, this invention proposes a method for quantifying the multi-temporal and spatial adjustment potential of vehicle-pile-network based on spatiotemporal diagrams, with the following beneficial effects: Compared with existing technologies, this invention, under complex constraints, obtains a city-level vehicle-pile-grid interactive operation baseline based on Monte Carlo simulation, thus resolving the uncertainty of system state changes during operation. By combining Pearson correlation coefficients to screen out the main factors affecting regulation potential indicators, the complexity and scale of the data are reduced, and computational efficiency is improved. An STGNN model is constructed based on artificial intelligence algorithms and trained on historical data. Then, online learning is used to predict its regulation potential, improving the digitalization level of power grid operation, increasing the flexibility and stability of power grid dispatch, and promoting the construction of new power systems.

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method for quantifying the multi-temporal and spatial adjustment potential of vehicle-pile-network based on spatiotemporal diagrams according to the present invention; Figure 2 This is a flowchart illustrating the construction of a runtime baseline based on Monte Carlo simulation in an embodiment of the present invention; Figure 3 This is a schematic diagram of the STGNN model construction in this invention. Detailed Implementation

[0018] To make the technical solutions, advantages, and objectives of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below. The described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the protection scope of the present invention.

[0019] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0020] Example like Figure 1 As shown, this invention provides a method for quantifying the multi-temporal and spatial adjustment potential of vehicle-pile-network based on spatiotemporal diagrams. The specific steps are as follows: S1. Based on the Monte Carlo simulation method, establish a city-level vehicle-pile-network interaction baseline under constraints. The steps are as follows: S11. Collect environmental data, user vehicle behavior data, charging pile data, and power grid operation data, and preprocess the collected data; (1) Environmental data: Data can be collected from meteorological bureaus or weather service agencies to obtain hourly temperature and humidity data, as well as weather conditions and seasonal differences; (2) User vehicle behavior data, user's initial SOC and expected SOC for charging, start time and end time of charging, battery capacity of each vehicle, etc. (3) Charging pile data, including allowable charging and discharging power, response time, etc.; (4) Power grid operation data: The historical load curve of the urban area reflects the load level of the power grid. Based on the local electricity price, the data of electricity price in different time periods are considered.

[0021] To ensure data accuracy, data preprocessing is required, including: 1) Clean up missing values ​​and handle outliers Missing data can be filled in or deleted. Filling in data can be done by using the mean of the data as an alternative method. ; in, This represents the mean of the data variable. For missing values ​​in the data, i It is an integer; At the same time, it is also necessary to pay attention to outliers that may exist in the data, and delete or replace outliers. Outliers in the data can be identified using the Z-score method: ; in, Z-score value For the first in the dataset i Data points, The mean of the dataset. The standard deviation of the dataset; 2) Convert data in different formats During the data collection process, the sources of the same data type may be different, so it is necessary to convert them into the same unit to make the integration simpler and more convenient.

[0022] 3) Normalize data with different dimensions. For different data types, data can be normalized to [0,1] to enable horizontal comparison. The specific Min-Max normalization method is as follows: ; in, X' The result after data normalization. The original data before normalization. The maximum value of the dataset. This represents the minimum value in the dataset.

[0023] S12. Establish constraints based on environmental factors, vehicle status, and electricity price signals; Taking into account environmental factors such as temperature, humidity, and season, vehicle status such as initial SOC and user time demand, and price signals such as charging price and grid time-of-use pricing, the system systematically defines the constraints for interactive operation from four dimensions: vehicle side, charging pile side, grid side, and collaborative optimization. These constraints include: The vehicle-side charging time constraint is calculated using the following formula: ; ; in, The set charging start time and desired end time, For charging time, The environmental coefficient is the ambient temperature and humidity. For the battery's rated capacity, and These refer to the expected state of charge and the initial state of charge, respectively. This refers to the actual charging power. The battery safety power constraint is given by the following formula: ; in, Minimum charging power for battery charging. The maximum charging power allowed by the battery. This refers to the actual charging power of the battery. The total power allocation constraint for charging piles is defined by the following formula: ; in, for t Time of the first i Charging power (kW) of each charging station. N This represents the total number of charging stations currently in a charging state. The total rated power for charging stations, The power consumption of auxiliary equipment in charging stations; The grid-side capacity constraint is given by the following formula: ; in, for t Time of the first j The load of each charging station j It is an integer. M This represents the total number of charging stations. This refers to the rated capacity of the transformer. The maximum load rate that can be operated; V2G reverse power supply constraint, the formula is as follows: ; in, The power required to charge electric vehicles to the grid. The maximum power that can be transmitted. The minimum charge state that allows for power transmission. Let be the state of charge of the vehicle battery at time t; The economic optimization constraint is given by the following formula: ; in, This can include time-of-use electricity pricing and charging service fees. for t Charging power at any time The set charging time period.

[0024] S13. Based on the constraints, use Monte Carlo simulation to generate the vehicle-pile-network interaction operation baseline; The city-level vehicle-charging-network interaction baseline refers to the overall natural operating state of the system without other external influences. Its construction is essentially a large-scale system state problem, with the core objective being to meet users' charging needs as much as possible while satisfying complex constraints. The specific steps are as follows: S131. Establish a probabilistic model based on the preprocessed data, including: 1) Vehicle arrival time at charging stations: This typically exhibits a bimodal distribution, which can be fitted using a Gaussian mixture model (GMM) for different time periods within the same day. i The probability of the vehicle arriving; ; in, The probability of reaching a charging station in time. For different normal distribution weights, It is a normal distribution function. The mean, Standard deviation; 2) Initial SOC: This probabilistic model should be based on the charging arrival time interval, establishing different conditional Beta distributions. These are continuous probability distributions defined in [0,1], and consist of two parameters. a and β control; ; in, For Beta functions, This refers to the battery's state of charge. , The shape parameter of the beta distribution; 3) Climate environment: Select a season as the simulation scenario (such as a summer weekday), and the daily temperature and humidity can be recorded as a curve that changes over time; 4) Grid price: For a specific region, a deterministic model for that region is used to determine the time-of-use electricity price function based on different times.

[0025] S132. Determine the baseline logic for interactive operation, including: Electric vehicle charging logic: Under the constraints of physics and basic rules, the charging strategy of the charging station is defined as instant charging, and charging starts with the maximum charging power; Grid interaction logic: Within a charging station, the total load of the charging pile group is the sum of the power of all vehicles currently charging, as shown in the following formula: ; in, The total load of the charging pile group The number of vehicles currently charging. for t Time of the first k The charging power of each charging station k It is an integer; S133. Generate a baseline using Monte Carlo simulation. First, initialize the parameters, setting the total simulation duration to 24 hours and the simulation time step. ∆t and total number of simulations D .

[0026] Before each loop begins, vehicle operation information for the day is randomly generated based on the probability distributions of the input data. t Starting from =0, according to the time step ∆t The simulation involves newly arriving vehicles being charged according to the charging logic. Charging power is allocated considering constraints, and the State of Charge (SOC) is updated. The power of each charging station, the SOC of each vehicle, the total charging load, and the charging cost are recorded and stored in a matrix. A single loop continues until the end of the total simulation time, totaling [number of simulations]. D This is repeated several times. Each iteration generates a set of data matrices, resulting in... D Massive amounts of data, including charging load curves and charging costs, are used to generate an operational baseline. The specific process is as follows: Figure 2 As shown.

[0027] Load baseline, calculated at each time point t The average total load is obtained using the following formula: ; in, As the load baseline, For the first d During the next simulation t The total load of the charging pile group at any given time. To simulate the total number of times, d It is an integer; The economic baseline is obtained by calculating the average charging cost for users across all time periods, using the following formula: ; in, As an economic baseline, Let k be the start time of the charging event. Let k be the end time of the charging event. for t The grid price at any given time (RMB / kWh); The power grid risk baseline is obtained by statistically analyzing transformer overload probabilities, using the following formula: ; in, As a baseline for power grid risk, for > Duration, The maximum power of the power grid. To simulate the total duration.

[0028] S2. Calculate the indicator parameters of the adjustment potential, and use the Pearson correlation coefficient to screen the factors affecting the potential parameters. The specific steps are as follows: S21. Define and calculate the indicator parameters of adjustment potential; The definition and calculation of all potential parameters are based on a certain time period. t Within the specified space, all eligible vehicles are counted. Assume there are a total of... N One charging station, charging station i When in a charging state, the following can be obtained: t At any given time, the relevant parameters and indicator calculation models for the vehicle-pile-network adjustment potential include: (1) Adjustable capacity: The power and energy that the vehicle-pile interaction system can adjust during charging and discharging.

[0029] 1) Increase power P up (kW) refers to the maximum power that a vehicle can increase from its current charging power, and its calculation formula is: ; in, To increase power, For the maximum allowable power, For the current power, The actual charging power of the vehicle. The maximum permissible power for charging electric vehicles. The rated maximum output power of the charging pile, The maximum allowable additional load on the power grid to which the vehicle is connected; 2) Reduce power P down (kW) refers to the maximum power reduction during charging. The calculation formula is: ; in, To reduce power; 3) V2G discharge power (kW): The power of an electric vehicle fed into the grid through a charging station is affected by ambient temperature. The calculation formula is as follows: ; in, for, for, To support the maximum power of V2G for charging stations, For the rated capacity of the electric vehicle battery, The maximum allowable discharge rate for the vehicle. This refers to the ambient temperature coefficient. 4) V2G discharge capacity (kWh) is the amount of electricity a battery can discharge into the grid without affecting user travel. The calculation formula is as follows: ; in, This refers to the discharge capacity of V2G. For the first i The state of charge of a vehicle while it is charging. The minimum charge state allowed for discharge. For battery discharge efficiency; 5) Adjustable power (kWh) is the amount of electricity that can be flexibly allocated during the charging process, and its calculation formula is: ; in, Adjustable power consumption; (2) Adjusting the depth : No. i The formula for calculating the proportion of the vehicle's actual adjusted power to its maximum adjustable range is as follows: ; in, To adjust the depth, The charging power after receiving the adjustment command. This is the maximum charging power; (3) Adjustment time: The first i Some time-related characteristics of electric vehicles participating in potential regulation, including response time. Adjustable time period V2G sustainability duration .

[0030] ; in, For response time, For adjustable time periods, For V2G sustainability duration, This refers to the communication duration from the start of command issuance to the receipt by the charging station. The time consumed by the charging pile controller to parse commands. The response time of the vehicle's battery management system (BMS) to commands. The time spent away from the charging station. The time for connecting to the charging station. The capacity for V2G reverse power transmission to vehicles. The power of continuous discharge; (4) Charging pile availability: The ratio of the power of charging piles in charging status to the sum of the rated power of all charging piles. The formula is as follows: ; in, For the availability of charging piles, For the first j Rated power of each charging station E This represents the total number of charging stations.

[0031] S22. By analyzing the behavior of all electric vehicles during charging, the Pearson correlation coefficient is used to measure the strength of the correlation between an influencing factor and the moderating potential parameter, thus identifying the main influencing factors affecting the vehicle-charging station-network interaction system. The formula is as follows: ; in, r The Pearson correlation coefficient ranges from [-1, 1], and its | r The closer the correlation is to 1, the stronger the correlation between the two variables. X It is an influencing factor. Y It is an adjustment potential indicator parameter. and These are the mathematical expectations of the two variables. and It is the variance of the two sets of variables.

[0032] S3. Construct a multi-temporal and spatial regulation potential quantitative assessment model based on the spatiotemporal graph neural network STGNN, and combine it with an online learning framework to realize regulation potential prediction; S31, such as Figure 3 As shown, a multi-spatiotemporal regulation potential quantitative evaluation model is constructed based on the spatiotemporal graph neural network STGNN. The specific steps are as follows: S311. Define the spatiotemporal graph and node features. The structure of the spatiotemporal graph is as follows: G= ( V,E,A ); in, V This is a collection of nodes for each charging station. E Let be the set of edges in the graph. A for N × N The adjacency matrix; The model input is a three-dimensional tensor X input =( B,F,Z ), B For the number of nodes, F For characteristic number, ZThe time window length for historical data; the model output is Y output =( B,C,T C represents the number of adjustment potential indicators; S312. Construct a spatial convolution module that uses a three-dimensional tensor as input to build an input matrix of historical data in the same time dimension. X t =( X 1, X 2,..., X N The graph convolution operation is performed on the input matrix, and the spectral graph method is used to apply the graph convolution operation to the spatiotemporal graph. G In the definition of the Laplace matrix , A It is an adjacency matrix. The normalized Laplacian matrix is ​​a diagonal matrix composed of node degrees, and the formula is as follows: ; in, It is the identity matrix; The graph convolution operation is performed on the input matrix, as shown in the following formula: ; in, For convolution kernel, U It is the Laplace matrix L Fourier bases after eigenvalue decomposition Λ It is an eigenvalue diagonal matrix. For Laplace matrix, This is a transpose operation; the formula implements the convolution operation on the graph through spectral domain transformation, converting the graph signal to the spectral domain for processing and then mapping it back to the original domain.

[0033] Regarding time t Corresponding graphic signal x t Using Chebyshev polynomials The formula for solving the graph convolutional network is as follows: ; in, Let be the vector of the coefficients in the polynomial. Laplace matrix L The largest eigenvalue, This is the transformed Laplace matrix; Finally, the convolution result... Integrate into a new input matrix as H t =( H 1, H 2,...,H t ); S313. Construct a spatiotemporal convolution module, using a multi-layer long short-term memory artificial neural network (LSTM) to perform convolution calculations on the time dimension. The steps are as follows: The initial importance value is calculated using the following formula: ; in, This is the initial importance value. , W d , U d For training parameters, d t-1 and s t-1 These are the output value and hidden state of the previous layer, respectively. tanh For activation functions; The initial importance scores are processed using Softmax to obtain the time importance scores, as shown in the following formula: ; in, Importance based on time; Time importance and H t By integrating the data, we obtain the key parameters for all time periods, as shown in the following formula: ; in, For context vectors; context vector Combined with the output of the previous LSTM layer, the formula is as follows: ; in, and For training parameters, This is the output vector at time t-1. Let be the input vector at time t-1. This is the context vector at time t-1; Will As input to a single LSTM unit, temporal convolutions are calculated layer by layer, as shown in the following formula: ; in, For Hadama accumulation, , , , For weights; , , , This is the bias value; For activation function, Output the value for the forget gate. Input values ​​for the input gate. The output value of the output gate. The state at time t, The state at time t-1 The output value at time t-1 The output value at time t; Output value By connecting to a fully connected layer, we obtain a quantitative prediction of the modulation potential. Based on the hourly, daily, and weekly quantitative forecasts of the adjustment potential, the quantitative assessment results of the multi-temporal and spatial adjustment potential are calculated, using the following formula: ; in, The results are a quantitative assessment of the modulation potential after the fusion of three time scales. W h , W d , W w The weights are for hourly, daily, and weekly time scales, respectively. , and These are quantitative forecasts of adjustment potential at three time scales: hourly, daily, and weekly.

[0034] S32. After the machine learning model is built, traditional offline training methods cannot reflect the continuous dynamic changes of the vehicle-pile-network system. An online learning framework can be used to fine-tune and update the model with new data to achieve accurate prediction of regulation potential. The steps are as follows: S321. Update the data based on the data collected in step S1, and collect relevant feature data of all nodes, such as the number of charging piles, rated power, segmented electricity price, number of vehicles connected to charging, charging state of charge, charging time, and ambient temperature and humidity. Quantitatively evaluate the output adjustment potential of historical data based on the STGNN model. Training and evaluation are performed; a large dataset is divided into training, validation, and test sets according to time. The training set is used to train the model's capabilities, the validation set is used to tune some parameters, and the test set is used to finally evaluate the model's accuracy and quantify its performance.

[0035] S322. Evaluate new real-time data through incremental fine-tuning; By collecting new data in real time, online models can be fine-tuned to adapt to recent weather, electricity prices, and user behavior. This minimizes the impact of new data on the already trained model, using a very small learning rate. The formula is as follows: ; in, These are the updated top-level network parameters in the STGNN model. These are the top-level network parameters in the model before the update. For learning rate, The gradient of the loss function. The loss function; The loss function calculates the difference between the predicted and actual values. The mean squared error (MSE) function is commonly used, as shown in the following formula: ; in, n For the sample size, For the first i The true value of each sample For the first i The predicted value for each sample.

[0036] It is worth noting that all contents not described in detail in this invention are existing technologies and are well known to those skilled in the art.

[0037] Therefore, this invention provides a method for quantifying the multi-temporal and spatial regulation potential of vehicle-pile-network based on spatiotemporal diagrams. By combining Monte Carlo simulation with multi-dimensional constraints to construct an accurate city-level vehicle-pile-network operation baseline, the uncertainty of system state changes is resolved. At the same time, Pearson correlation coefficients are used to screen key influencing factors and eliminate redundant data to improve computational efficiency. Furthermore, relying on the STGNN model and online learning, accurate prediction of regulation potential is achieved and adaptation to dynamic changes in the system is realized. Ultimately, the digitalization level of the power grid is improved, the overload duration of the distribution network is reduced, and the scheduling flexibility and operational stability are enhanced.

[0038] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for quantifying the multi-temporal and spatial adjustment potential of vehicle-pile-network based on spatiotemporal maps, characterized in that, The specific steps are as follows: Step S1: Based on the Monte Carlo simulation method, establish a city-level vehicle-pile-network interaction operation baseline under constraints. The steps are as follows: Step S11: Collect environmental data, user vehicle behavior data, charging pile data, and power grid operation data, and preprocess the collected data; Step S12: Establish constraints based on environmental factors, vehicle status, and electricity price signals; Step S13: Based on the constraints, use Monte Carlo simulation to generate the vehicle-pile-network interaction operation baseline; Step S2: Calculate the indicator parameters of the adjustment potential and use the Pearson correlation coefficient to screen the factors affecting the potential parameters; Step S3: Construct a multi-spatiotemporal regulation potential quantitative assessment model based on the spatiotemporal graph neural network STGNN, and combine it with an online learning framework to realize regulation potential prediction.

2. The method for quantifying the multi-temporal and spatial adjustment potential of vehicle-pile-network based on spatiotemporal diagrams according to claim 1, characterized in that, In step S11, the preprocessing operations include missing and outlier removal, data format conversion, and data normalization. Missing values ​​are imputed using the mean imputation method, as shown in the following formula: ； in, This represents the mean of the data variable. For missing values ​​in the data, i It is an integer; Outliers are identified using the Z-score method, as shown in the formula below: ； in, Z-score value For the first in the dataset i Data points, The mean of the dataset. The standard deviation of the dataset; Normalization is calculated using the Min-Max method, as shown in the following formula: ； in, X' The result after data normalization. The original data before normalization. The maximum value of the dataset. This represents the minimum value in the dataset.

3. The method for quantifying the multi-temporal and spatial adjustment potential of vehicle-pile-network based on spatiotemporal diagrams according to claim 1, characterized in that, In step S12, the mathematical constraints include: The vehicle-side charging time constraint is calculated using the following formula: ； ； in, The set charging start time and desired end time, For charging time, An environmental coefficient representing ambient temperature and humidity. For the battery's rated capacity, and These refer to the expected state of charge and the initial state of charge, respectively. This refers to the actual charging power. The battery safety power constraint is given by the following formula: ； in, Minimum charging power for battery charging. The maximum charging power allowed by the battery. This refers to the actual charging power of the battery. The total power allocation constraint for charging piles is defined by the following formula: ； in, for t Time of the first i The charging power of each charging station N This represents the total number of charging stations currently in a charging state. The total rated power for charging stations, The power consumption of auxiliary equipment in charging stations; The grid-side capacity constraint is given by the following formula: ； in, for t Time of the first j The load of each charging station j It is an integer. M This represents the total number of charging stations. This refers to the rated capacity of the transformer. The maximum load rate that can be operated; V2G reverse power supply constraint, the formula is as follows: ； in, The power required to charge electric vehicles to the grid. The maximum power that can be transmitted. The minimum charge state that allows for power transmission. Let be the state of charge of the vehicle battery at time t; The economic optimization constraint is given by the following formula: ； in, It can include time-of-use electricity pricing and charging service fees. for t Charging power at any time The set charging time period.

4. The method for quantifying the multi-temporal and spatial adjustment potential of vehicle-pile-network based on spatiotemporal diagrams according to claim 1, characterized in that, In step S13, the vehicle-pile-network interactive operation baseline is generated using Monte Carlo simulation. The specific steps are as follows: Step S131: Establish a probability model based on the preprocessed data, including: a time probability model of vehicle arrival at charging pile based on Gaussian mixture model (GMM), an initial SOC probability model based on conditional Beta distribution, a climate environment model that varies over time, and a deterministic power grid price model. Step S132: Determine the baseline logic for interactive operation, including: Electric vehicle charging logic: Under the constraints of physics and basic rules, the charging strategy of the charging station is defined as instant charging, and charging starts with the maximum charging power; Grid interaction logic: Within a charging station, the total load of the charging pile group is the sum of the power of all vehicles currently charging, as shown in the following formula: ； in, The total load of the charging pile group The number of vehicles currently charging. for t Time of the first k The charging power of each charging station k It is an integer; Step S133: Generate a baseline using Monte Carlo simulation, including: Load baseline, calculated at each time point t The average total load is obtained using the following formula: ； in, As the load baseline, For the first d During the next simulation t The total load of the charging pile group at any given time. To simulate the total number of times, d It is an integer; The economic baseline is obtained by calculating the average charging cost for users across all time periods, using the following formula: ； in, As an economic baseline, Let k be the start time of the charging event. Let k be the end time of the charging event. for t The electricity price at any given time; The power grid risk baseline is obtained by statistically analyzing transformer overload probabilities, using the following formula: ； in, As a baseline for power grid risk, for > Duration, The maximum power of the power grid. To simulate the total duration.

5. The method for quantifying the multi-temporal and spatial adjustment potential of vehicle-pile-network based on spatiotemporal diagrams according to claim 4, characterized in that, In step S131, the time probability model for the vehicle's arrival at the charging station based on the Gaussian mixture model (GMM) is as follows: ； in, The probability of reaching a charging station in time. For different normal distribution weights, It is a normal distribution function. The mean, Standard deviation; The initial SOC probability model based on the conditional Beta distribution is as follows: ； in, For Beta functions, This refers to the battery's state of charge. , Let be the shape parameter of the beta distribution.

6. The method for quantifying the multi-temporal and spatial adjustment potential of vehicle-pile-network based on spatiotemporal diagrams according to claim 1, characterized in that, In step S2, the specific steps are as follows: Step S21: Define and calculate the index parameters of the adjustment potential, including adjustment capacity, adjustment depth, adjustment time and charging pile availability; the adjustment capacity includes power increase, power decrease, V2G discharge power, V2G discharge amount and adjustable power; the adjustment time includes response time, adjustable time period and V2G sustainable duration. Step S22: Use the Pearson correlation coefficient to screen factors affecting the moderating potential parameter. The formula for calculating the Pearson correlation coefficient is as follows: ； in, r The Pearson correlation coefficient is used. X It is an influencing factor. Y It is an adjustment potential indicator parameter. and These are the mathematical expectations of the two variables. and It is the variance of the two sets of variables.

7. The method for quantifying the multi-temporal and spatial adjustment potential of vehicle-pile-network based on spatiotemporal diagrams according to claim 6, characterized in that, The formula for calculating the power increase is as follows: ； in, To increase power, For the maximum allowable power, For the current power, The actual charging power of the vehicle. The maximum permissible power for charging electric vehicles. The rated maximum output power of the charging pile, The maximum allowable additional load on the power grid to which the vehicle is connected; The formula for calculating the reduced power is as follows: ； in, To reduce power; The formula for calculating the discharge power of V2G is as follows: ； in, for, for, To support the maximum power of V2G for charging stations, For the rated capacity of the electric vehicle battery, The maximum allowable discharge rate for the vehicle. This refers to the ambient temperature coefficient. The formula for calculating the discharge capacity of V2G is as follows: ； in, This refers to the discharge capacity of V2G. For the first i The state of charge of a vehicle while it is charging. The minimum charge state allowed for discharge. For battery discharge efficiency; The formula for calculating adjustable power is as follows: ； in, Adjustable power consumption; The formula for adjusting the depth is as follows: ； in, To adjust the depth, The charging power after receiving the adjustment command. This is the maximum charging power; The formula for calculating the adjustment time is as follows: ； in, For response time, For adjustable time periods, For V2G sustainability duration, This refers to the communication duration from the start of command issuance to the receipt by the charging station. The time consumed by the charging pile controller to parse commands. The response time of the vehicle's battery management system (BMS) to commands. The time spent away from the charging station. The time for connecting to the charging station. Capacity for V2G reverse power supply to vehicles. The power of continuous discharge; The formula for calculating the availability of charging piles is as follows: ； in, For the availability of charging piles, For the first j Rated power of each charging station E This represents the total number of charging stations.

8. The method for quantifying the multi-temporal and spatial adjustment potential of vehicle-pile-network based on spatiotemporal diagrams according to claim 1, characterized in that, In step S3, a multi-spatiotemporal regulation potential quantitative evaluation model is constructed based on the spatiotemporal graph neural network STGNN. The specific steps are as follows: Step S311: Define the spatiotemporal graph and node features. The spatiotemporal graph structure is as follows: G= ( V,E,A )； in, V This is a collection of nodes for each charging station. E Let be the set of edges in the graph. A for N × N The adjacency matrix; The model input is a three-dimensional tensor X input =( B,F,Z ), B For the number of nodes, F For characteristic number, Z The time window length for historical data; the model output is Y output =( B,C,T C represents the number of adjustment potential indicators; Step S312: Construct a spatial convolution module, using a three-dimensional tensor as input, to build an input matrix of historical data in the same time dimension. X t =( X 1, X 2,..., X N The graph convolution operation is performed on the input matrix, and the spectral graph method is used to apply the graph convolution operation to the spatiotemporal graph. G In the definition of the Laplace matrix , A It is an adjacency matrix. The normalized Laplacian matrix is ​​a diagonal matrix composed of node degrees, and the formula is as follows: ； in, It is the identity matrix; The graph convolution operation is performed on the input matrix, as shown in the following formula: ； in, For convolution kernel, U It is the Laplace matrix L Fourier bases after eigenvalue decomposition Λ It is an eigenvalue diagonal matrix. For Laplace matrix, This is a transpose operation; Regarding time t Corresponding graphic signal x t Using Chebyshev polynomials The formula for solving the graph convolutional network is as follows: ； in, Let be the vector of the coefficients in the polynomial. Laplace matrix L The largest eigenvalue, This is the transformed Laplace matrix; Finally, the convolution result... Integrate into a new input matrix as H t =( H 1, H 2,..., H t ); Step S313: Construct a spatiotemporal convolution module and use a multilayer long short-term memory artificial neural network LSTM to perform convolution calculations on the time dimension.

9. A method for quantifying the multi-temporal and spatial adjustment potential of vehicle-pile-network based on spatiotemporal diagrams according to claim 8, characterized in that, In step S313, convolution calculation is performed on the time dimension, and the steps are as follows: The initial importance value is calculated using the following formula: ； in, This is the initial importance value. , W d , U d For training parameters, d t-1 and s t-1 These are the output value and hidden state of the previous layer, respectively. tanh For activation functions; The initial importance scores are processed using Softmax to obtain the time importance scores, as shown in the following formula: ； in, Importance based on time; Time importance and H t By integrating the data, we obtain the key parameters for all time periods, as shown in the following formula: ； in, For context vectors; context vector Combined with the output of the previous LSTM layer, the formula is as follows: ； in, and For training parameters, This is the output vector at time t-1. Let be the input vector at time t-1. This is the context vector at time t-1; Will As input to a single LSTM unit, temporal convolutions are calculated layer by layer, as shown in the following formula: ； in, For Hadama accumulation, , , , For weights; , , , This is the bias value; For activation function, Output the value for the forget gate. Input values ​​for the input gate. The output value of the output gate. The state at time t, The state at time t-1 The output value at time t-1 This represents the output value at time t. Output value By connecting to a fully connected layer, we obtain a quantitative prediction of the modulation potential. Based on the hourly, daily, and weekly quantitative forecasts of the adjustment potential, the quantitative assessment results of the multi-temporal and spatial adjustment potential are calculated, using the following formula: ； in, The results are a quantitative assessment of the modulation potential after the fusion of three time scales. W h , W d , W w The weights are for hourly, daily, and weekly time scales, respectively. , and These are quantitative forecasts of adjustment potential at three time scales: hourly, daily, and weekly.

10. A method for quantifying the multi-temporal and spatial adjustment potential of vehicle-pile-network based on spatiotemporal diagrams according to claim 8, characterized in that, In step S3, the modulation potential prediction is implemented using an online learning framework, and the steps are as follows: Step S321: Quantitatively evaluate the output modulation potential of historical data based on the STGNN model. Conduct training and assessment; Step S322: Evaluate the new real-time data using incremental fine-tuning, as shown in the following formula: ； in, These are the updated top-level network parameters in the STGNN model. These are the top-level network parameters in the model before the update. For learning rate, The gradient of the loss function. This is the loss function.

Images