Micro-grid electrical load prediction method for high-speed service area

Abstract

The invention discloses a microgrid electrical load prediction method for a high-speed service area, and belongs to the technical field of electrical load prediction, and the method comprises the steps: obtaining historical traffic flow data and electrical load data of the service area, and constructing an electrical network topology structure of the service area; analyzing the topological structure of the power utilization network in the service area based on a power flow calculation method to obtain a corresponding relation between the network loss data and the traffic flow; performing parameter estimation on a corresponding relation between the network loss data and the traffic flow based on a multivariate nonlinear regression model to obtain a quantitative mapping relation; training the attention neural network through a quantitative mapping relation among the traffic flow, the load and the network loss, and generating a multivariable time sequence prediction model; and predicting the electrical load of the micro-grid in the high-speed service area based on the multivariable time sequence prediction model. According to the method, accurate dynamic prediction of the power consumption load of the service area is realized, decision support is provided for optimizing an operation strategy, and the power consumption management level and the operation efficiency of the service area can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power load prediction, and in particular relates to a method for predicting power load of a microgrid in a high-speed service area. Background Art

[0002] There is a technical contradiction in the prediction of power load in microgrids in high-speed service areas, that is, how to take into account the impact of power load on traffic flow while considering the eddy current characteristics of traffic flow. On the one hand, the eddy current distribution of traffic flow reflects the gathering and staying rules of vehicles in the service area. By analyzing the eddy current characteristics of traffic flow, the power load in different areas can be predicted, and more refined load management can be achieved.

[0003] How to find a balance between the eddy current characteristics of vehicle flow and the impact of power load, and rationally utilize power grid loss data, is a technical problem that needs to be solved urgently in the power load forecasting of microgrids in highway service areas. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a method for predicting power load of a microgrid in a high-speed service area to solve the problems existing in the above-mentioned prior art.

[0005] To achieve the above object, the present invention provides a method for predicting power load of a microgrid in a high-speed service area, comprising:

[0006] Acquire historical traffic flow data and power load data of the service area, and construct a power network topology structure of the service area based on the historical traffic flow data and power load data of the service area;

[0007] Analyzing the topological structure of the power network in the service area based on the power flow calculation method to obtain the corresponding relationship between network loss data and vehicle flow;

[0008] Based on the multivariate nonlinear regression model, the corresponding relationship between the network loss data and the vehicle flow is estimated to obtain the quantitative mapping relationship between the vehicle flow and the load and network loss;

[0009] An attention neural network is constructed based on the attention mechanism and the long-short time network model, and the attention neural network is trained through the quantitative mapping relationship between the vehicle flow and the load and the network loss to generate a multivariate time series prediction model;

[0010] The power load of the microgrid in the high-speed service area is predicted based on the multivariable time series prediction model.

[0011] Preferably, the process of obtaining historical traffic flow data and power load data of the service area, and constructing the power network topology structure of the service area based on the historical traffic flow data and power load data of the service area includes:

[0012] Acquire historical vehicle flow data and power load data of the service area, and preprocess the acquired historical vehicle flow data and power load data of the service area to obtain preprocessed data;

[0013] Calculating the preprocessed data by Pearson correlation coefficient method to obtain a correlation coefficient matrix;

[0014] The service area power network topology structure is constructed based on the correlation coefficient matrix.

[0015] Preferably, the historical traffic flow data and power load data of the service area are preprocessed, and the process of obtaining the preprocessed data includes: cleaning and normalizing the historical traffic flow data and power load data of the service area, and unifying the data to the same time scale to obtain the preprocessed data.

[0016] Preferably, the process of analyzing the topological structure of the power network in the service area based on the power flow calculation method and obtaining the corresponding relationship between the network loss data and the vehicle flow includes:

[0017] Analyze the topological structure of the power network in the service area based on a graph theory algorithm to obtain key nodes and lines with network losses greater than a preset threshold;

[0018] Based on the power flow calculation method, the key nodes and lines whose network loss is greater than a preset threshold are simulated to obtain network loss distribution data;

[0019] The network loss distribution data is grouped by a clustering algorithm to obtain a corresponding relationship between the network loss data and the vehicle flow.

[0020] Preferably, the process of obtaining the quantitative mapping relationship between vehicle flow, load and network loss includes:

[0021] A multivariate nonlinear regression model is constructed based on the eddy current characteristics of vehicle flow and the corresponding relationship between the network loss data and vehicle flow;

[0022] The parameters of the multivariate nonlinear regression model are estimated based on the least squares method, and the mean square error and the coefficient of determination of the regression model are calculated to determine the goodness of fit and the prediction accuracy of the regression model, and to generate a trained regression model;

[0023] A quantitative mapping relationship between the vehicle flow, load and network loss is generated based on the trained regression model.

[0024] Preferably, the process of constructing an attention neural network based on the attention mechanism and the long-short time network model includes:

[0025] Construct the attention mechanism module and the initial long-short time network model;

[0026] The attention mechanism module is deployed to each time step of the long-short time network model to generate the attention neural network.

[0027] Preferably, the process of training the attention neural network through the quantitative mapping relationship between the vehicle flow and the load and network loss to generate the multivariate time series prediction model includes:

[0028] Constructing a data set based on the quantitative mapping relationship between the vehicle flow and the load and network loss;

[0029] Training the attention neural network using the data set, calculating the prediction value and loss function through forward propagation, and updating the weight parameters of the model using a backpropagation algorithm;

[0030] The performance of the LSTM neural network model is evaluated on the data set, and the model is optimized by adjusting the number of hidden layers, the number of neurons, and the learning rate hyperparameters to generate the multivariate time series prediction model.

[0031] Preferably, the process of optimizing the model by adjusting the number of hidden layers, the number of neurons, and the learning rate hyperparameters includes: training the model using the Adam optimization algorithm, setting the learning rate to 0.01, the batch size to 24, training 50 epochs, evaluating the model performance on the validation set after each epoch, and recording the root mean square error.

[0032] Preferably, the number of hidden layers of the multivariate time series prediction model is 2, and the number of neurons is 128 and 64 respectively.

[0033] Preferably, the process of predicting the power load of the microgrid in the high-speed service area based on the multivariable time series prediction model includes:

[0034] Obtaining output results of a multivariate time series prediction model, performing sensitivity analysis on the output results, and obtaining sensitivity analysis results;

[0035] The sensitivity analysis results are processed using data visualization technology to generate visual prediction results.

[0036] Compared with the prior art, the present invention has the following advantages and technical effects:

[0037] The present invention discloses a method for predicting the power load in a service area based on the eddy current characteristics of vehicle flow. The method constructs a multivariate nonlinear regression model by analyzing the correlation between vehicle flow and power load and network loss, and establishes a quantitative mapping relationship between vehicle flow and load and network loss. A long short-term memory neural network algorithm is adopted, and a multivariate time series prediction model is built with historical data of vehicle flow, load, and network loss as input. An attention mechanism is introduced to dynamically adjust the weight according to the degree of influence of the eddy current characteristics of vehicle flow on the load. An abnormal data detection mechanism is added to improve the robustness of the model. Finally, the prediction results are visualized and explained to reveal the influence mechanism of each factor on the prediction. The present invention realizes accurate dynamic prediction of the power load in the service area, provides decision support for optimizing the operation strategy, and helps to improve the power management level and operation efficiency of the service area. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0039] Figure 1 This is a flow chart of a method for predicting power load of a microgrid in a high-speed service area according to an embodiment of the present invention. DETAILED DESCRIPTION

[0040] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0041] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0042] Embodiment 1

[0043] like Figure 1 As shown, this embodiment provides a method for predicting power load of a microgrid in a high-speed service area, including:

[0044] Firstly, based on the historical traffic flow data and power load data of the service area, the correlation analysis method is used to calculate the correlation coefficient matrix between traffic flow and power load, determine the strength of the correlation between the two, and obtain a quantitative description of the impact of traffic flow on power load.

[0045] Acquire historical vehicle flow data and power load data of the service area, preprocess the vehicle flow data and the power load data to obtain preprocessed vehicle flow data and preprocessed power load data; use the Pearson correlation coefficient method to calculate the correlation coefficient between the preprocessed vehicle flow data and the preprocessed power load data to obtain a correlation coefficient matrix; determine whether the element value in the correlation coefficient matrix is ​​greater than a preset threshold value, and if so, determine that there is a strong correlation between the vehicle flow and the power load;

[0046] To further optimize the solution, we first obtain the traffic flow data and power load data of the past year from the historical database of the service area, clean and normalize these data, eliminate the influence of outliers and missing values, and unify the data to the same time scale, such as the average traffic flow and average power load per hour. Then, we use the Pearson correlation coefficient method to calculate the correlation coefficient matrix between the preprocessed traffic flow data and power load data. If the element value in the matrix is ​​greater than 8, it is considered that there is a strong correlation between traffic flow and power load.

[0047] Secondly, by analyzing the topological structure of the power network in the service area, the key nodes and lines with large network losses are identified. The power flow calculation method is used to simulate the network loss distribution under the eddy current characteristics of different vehicle flows, and the corresponding relationship between network loss data and vehicle flow is obtained.

[0048] Acquire the topological structure data of the power network in the service area, analyze the topological structure data by using a graph theory algorithm, and obtain key nodes and lines whose network loss is greater than a preset threshold; acquire historical vehicle flow data and network loss data of a corresponding period, analyze the historical vehicle flow data and the network loss data by using a data mining algorithm, and obtain association rules between vehicle flow and network loss; according to the association rules between vehicle flow and network loss, use a power flow calculation method to simulate the network loss distribution under different vehicle flow eddy current characteristics, and obtain network loss distribution data;

[0049] To further optimize the scheme, we first obtain the topological structure data of the power network from the service area distribution automation system, including the connection relationship and electrical parameters of components such as substations, distribution lines, and switchgear. Then, we use the Dijkstra shortest path algorithm to analyze the topological structure data, calculate the electrical distance between any two points in the network, and generate a minimum spanning tree with the substation as the root node to identify the key nodes and lines that have a greater impact on network loss, such as branches with an average electrical distance greater than 5. Then, we obtain the traffic flow data and network loss data for the past year from the historical database of the service area, clean and normalize the data, and use the Apriori association rule mining algorithm to analyze the association rules between traffic flow and network loss. For example, when the traffic flow exceeds 500 vehicles / hour, the average network loss rate of the distribution transformer increases by 2%. On this basis, the power flow calculation method is used to simulate the network loss distribution under different traffic flows, and a set of network loss distribution data is obtained. This set of data is clustered using the K-means clustering algorithm, and the network loss distribution pattern is divided into three levels: high, medium, and low.

[0050] Then, according to the correlation between the eddy current characteristics of vehicle flow and power load and network loss, a multivariate nonlinear regression model is constructed, with vehicle flow as the independent variable and power load and network loss as the dependent variables. Through model parameter estimation, the quantitative mapping relationship between vehicle flow and load and network loss is obtained.

[0051] The historical vehicle flow, power load and network loss data are obtained, and the data are preprocessed to remove outliers and missing values ​​to obtain an effective training data set; a multivariate nonlinear regression model is constructed according to the nonlinear relationship between the vehicle flow, the power load and the network loss, and the multivariate nonlinear regression model includes a polynomial regression model and a support vector regression model; the vehicle flow is used as the independent variable of the regression model, and the power load and the network loss are used as dependent variables to construct a multivariate nonlinear regression equation; the least squares method or gradient descent method is used to estimate the parameters of the regression model, and the vehicle flow and the load are obtained. The method comprises the steps of: calculating the mean square error and the coefficient of determination of the regression model, and judging the goodness of fit and the prediction accuracy of the regression model; if the performance of the regression model does not meet the preset requirements, adjusting the structure or parameters of the regression model, and re-executing the step of constructing the multivariate nonlinear regression model; predicting the power load and the network loss value in the future period according to the real-time vehicle flow data and the trained regression model; obtaining new vehicle flow, load and network loss data, adding the new data to the training data set, re-estimating the parameters of the regression model, and updating the regression model.

[0052] To further optimize the scheme, we first obtained the traffic flow, power load and network loss data of the past year from the monitoring system and energy consumption management system of the service area, and cleaned and preprocessed the data. The Tukey algorithm was used to identify and remove outliers in the traffic flow and load data, and the outlier threshold was set to ±5 times the interquartile range of the upper and lower quartiles. The nearest neighbor interpolation method was used to fill the missing values. If there were more than 5 consecutive missing data, the data of the time period was directly removed. After preprocessing, a training data set containing 8760 samples was obtained. By analyzing the scatter plots between traffic flow, power load and network loss, it was found that there was a significant nonlinear relationship between the three. Therefore, a multivariate nonlinear regression model was constructed, including quadratic polynomial regression and radial basis kernel function support vector regression. With traffic flow as the independent variable and power load and network loss as the dependent variables, a regression equation was constructed. The least squares method was used to estimate the parameters of the polynomial regression model, and the optimal polynomial order was determined to be 2 through five-fold cross validation. The grid search and particle swarm optimization algorithms were used to optimize the parameters of the support vector regression model, and the optimal penalty coefficient C = 10 and kernel function parameter g = 1 were obtained. The training data set was input into the regression model for training to obtain the quantitative mapping relationship between vehicle flow, load and network loss. The root mean square error RMSE and determination coefficient R^2 of the regression model on the test set were calculated. The RMSE of the polynomial regression model was 15, R^2 = 92, and the RMSE of the support vector regression model was 12, R^2 = 95, all of which met the preset performance requirements.

[0053] The least square method or the gradient descent method is used to estimate the parameters of the regression model to obtain a quantitative mapping relationship between the vehicle flow and the load and the network loss.

[0054] Obtain a training data set and a test data set including historical vehicle flow, load, and network loss rate, wherein the training data set is used to train a regression model, and the test data set is used to evaluate the performance of the regression model; select at least one regression model from a linear regression model, a polynomial regression model, and a support vector regression model according to preset business requirements; use the least squares method or the gradient descent method to estimate the model parameters of the at least one regression model so as to minimize the fitting error of the at least one regression model on the training data set; calculate the mean square error and the coefficient of determination of each regression model on the test data set, and determine whether each regression model meets the preset business requirements according to the mean square error and the coefficient of determination; if the performance of at least one regression model does not meet the preset business requirements, adjust the model type or hyperparameters of the regression model according to its performance evaluation results, and re-execute parameter estimation and performance evaluation until the performance of the regression model meets the preset business requirements;

[0055] Then, the long short-term memory neural network (LSTM) algorithm is used to build a multivariate time series prediction model with historical time series data of vehicle flow, power load, and power grid loss as input. Through model training and parameter tuning, dynamic prediction of the power load in the service area in the future can be achieved.

[0056] According to the preprocessed data set, an LSTM neural network model including an input layer, an LSTM hidden layer and an output layer is designed, wherein the structure of the input layer is determined by the dimension of the input feature, the number of layers and the number of neurons in the LSTM hidden layer are set according to the complexity of the prediction problem and the sequence length, and the activation function and loss function of the output layer are designed according to the type of prediction target;

[0057] Dividing the preprocessed data set into a training set, a validation set and a test set, using the training set data to train the LSTM neural network model, calculating the predicted value and the loss function through forward propagation, and using the back propagation algorithm to update the weight parameters of the model;

[0058] Evaluate the performance of the LSTM neural network model on the validation set, optimize the model by adjusting the number of hidden layers, the number of neurons, and the learning rate hyperparameters to obtain an optimized LSTM neural network model;

[0059] Applying the optimized LSTM neural network model to the test set data, calculating the error index between the predicted value and the true value, and judging the generalization ability and prediction performance of the model;

[0060] A neural network model with two LSTM hidden layers is designed. The first hidden layer contains 64 neurons and the second hidden layer contains 32 neurons. Considering that the prediction is a continuous variable, a linear activation function is used in the output layer, and the mean square error is used as the loss function. The data set is randomly divided into a training set, a validation set, and a test set in a ratio of 8:1:1. The model is trained using the Adam optimization algorithm, with a learning rate set to 0.01, a batch size of 24, and 50 epochs. The model performance is evaluated on the validation set after each epoch, and the root mean square error (RMSE) is recorded. The number of LSTM layers and the number of neurons are optimized by grid search. Finally, the optimal model is determined to have 2 hidden layers, 128 neurons, and 64 neurons, respectively. The RMSE on the validation set is reduced to 20. The optimal model is applied to the test set, and the RMSE of the test set is 22, which is close to the performance of the validation set, indicating that the model has good generalization ability. Finally, the trained model is deployed to the energy management system of the service area, which automatically obtains the latest traffic flow, load and network loss data every hour to predict the hourly power load in the next 24 hours. The prediction results are used to optimize the power consumption plan and scheduling plan of the service area.

[0061] To further optimize the scheme, the normal threshold range of each data is obtained by statistically analyzing the historical traffic flow, power load and network loss data. The normal threshold range of traffic flow data is 500-2000 vehicles / hour, the normal threshold range of power load data is 1000-5000kWh, and the normal threshold range of network loss data is 5%-15%. After obtaining real-time data, the Dixon test method is used to detect outliers on the data, and the data that exceeds the normal threshold range and the Dixon test value is greater than the critical value is determined as abnormal data. The data smoothing method is used to process the abnormal data. The weighted moving average method is used to set the window size to 5 and the weighting coefficients to 2, 3, and 5 to smooth the abnormal data. The processed data is input into the support vector machine model, the radial basis kernel function is used, the penalty coefficient C is set to 10, the model is trained and the parameters are optimized. The trained model is used to predict the traffic flow, power load and network loss data for the next 24 hours to obtain the prediction result data. The reliability analysis of the prediction results is performed, the residual between the predicted value and the actual value is calculated, and the 95% confidence interval is estimated. If the residual mean is greater than 100 or the confidence interval width is greater than 20%, it is considered that the reliability does not meet the requirements and the data anomaly detection and processing steps are re-executed.

[0062] Through statistical analysis of historical vehicle flow, power load and network loss data, the normal threshold range of each data is determined, and real-time data is obtained for comparison with the threshold to determine whether there is abnormal data. The abnormal data is corrected using methods such as data smoothing and interpolation. The processed data is input into the prediction model for training and optimization to obtain reliable prediction results.

[0063] Obtain historical vehicle flow, power load and network loss data, and determine the normal threshold range of historical data through data analysis; obtain real-time vehicle flow, power load and network loss data, compare the real-time data with the normal threshold range, and determine whether the real-time data is abnormal; if the real-time data is abnormal, use data smoothing and interpolation methods to correct the abnormal data to obtain corrected data; input the corrected data into the prediction model, optimize the parameters of the prediction model through training, and obtain the optimized prediction model; according to the optimized prediction model, predict the vehicle flow, power load and network loss data in a future period of time to obtain prediction results; compare the prediction results with the actual data, and improve the prediction model through error analysis; according to the prediction results, warn of abnormal situations that may occur in the future period of time.

[0064] To further optimize the scheme, by statistically analyzing the historical traffic volume, power load and network loss data of the highway service area, the normal threshold range of traffic volume data is 500-2000 vehicles / hour, the normal threshold range of power load data is 1000-5000kWh, and the normal threshold range of network loss data is 5%-15%. The data acquisition system is used to obtain the traffic volume, power load and network loss data of the service area in real time, and the real-time data is compared with the determined normal threshold range. If the data exceeds the normal range, it is determined to be abnormal data. The data smoothing and interpolation methods are used to correct the abnormal data. The weighted moving average method is used to smooth the abnormal data. The window size is set to 5 and the weighting coefficient is [1,2,4,2,1]. The Lagrange interpolation method is used to interpolate the missing data to obtain the corrected data. The corrected data is input into the support vector machine regression model, and the penalty coefficient C of the model and the parameter g of the radial basis kernel function are optimized by the grid search method. The optimal parameters C = 10, g = 1 are obtained through 5-fold cross validation, and the optimized support vector machine prediction model is obtained by training with the optimal parameters. The prediction model is used to predict the traffic flow, power load and network loss data for the next 24 hours to obtain the prediction result data. The prediction results are compared with the actual data, and the mean absolute percentage error MAPE is calculated. When the MAPE is greater than 10%, the prediction model is adjusted, and time series features such as hours and weeks are introduced to improve the prediction accuracy to less than 5%. According to the comparison between the prediction results and the normal threshold range, it is judged whether there will be a surge in traffic flow, excessive power load or abnormal network loss in the next 24 hours. If an abnormality occurs, an early warning will be issued, and the early warning information will be sent to the service area management personnel to prompt them to deploy resources in advance to deal with abnormal situations and ensure the safe and smooth operation of the service area.

[0065] Real-time vehicle flow, power load and network loss data are obtained, and the real-time data is compared with a preset historical normal threshold range to determine whether the real-time data exceeds the threshold range; if so, it is determined that the real-time data is abnormal, and the abnormal data is processed by a data smoothing algorithm and an interpolation algorithm to obtain corrected real-time data; according to a pre-established prediction model, the vehicle flow, power load and network loss data in a future period are predicted, and the predicted data is compared with the corrected real-time data. If the deviation between the predicted data and the corrected real-time data exceeds a preset abnormal threshold, it is determined that data abnormalities will occur in the future period; for the abnormal situation that occurs, a time series analysis algorithm is used to evaluate the severity of the abnormal situation to obtain abnormal severity data; according to the abnormal severity data, an abnormal information extraction algorithm is used to extract key information from the abnormal data to generate abnormal warning information; the abnormal warning information is sent to a preset management terminal, and the abnormal warning information is displayed through the management terminal.

[0066] Furthermore, combined with the actual needs of service area operation and management, the output results of the prediction model are visualized and explained. Through sensitivity analysis and feature importance ranking, the influencing mechanism of factors such as vehicle flow eddy current characteristics, power load, and power grid loss on the prediction results is revealed, providing decision support for optimizing the service area operation strategy.

[0067] To further optimize the scheme, the time series data of key indicators including traffic flow, power load and power grid loss were obtained by collecting and sorting the historical operation data of the service area in the past three years. A multivariate time series prediction model was established using the long short-term memory (LSTM) neural network algorithm. The input features of the model include historical traffic flow, historical power load, weather conditions, holidays and other factors, and the output is the predicted values ​​of traffic flow and power load in the next 24 hours. During the model training process, the Adam optimizer was used, and the optimal hyperparameter combination was determined through grid search. The prediction results were displayed through a dynamic dashboard, and the changing trends of traffic flow and power load were intuitively presented in the form of line charts and bar charts. At the same time, the SHAP method was used to perform sensitivity analysis on the input features of the model and calculate the contribution of each feature to the prediction results. The analysis found that historical traffic flow and weather conditions were the most important factors affecting the prediction results, and their average SHAP values ​​reached 35 and 28 respectively. Further analysis of the traffic flow data showed that during the morning and evening peak hours of each day, the traffic flow showed obvious periodic fluctuations, indicating the presence of eddy current characteristics. The impact of power load and network loss data on the prediction results is relatively small, with the average SHAP values ​​of 12 and 08 respectively. Based on the above analysis results, the following conclusions can be drawn: First, the traffic flow prediction needs to focus on the impact of historical traffic flow and weather conditions; second, the eddy current characteristics during peak hours in the morning and evening have an important impact on traffic flow prediction; third, the impact of power load and network loss data on the prediction results is relatively small. These conclusions can provide important decision-making support for optimizing the operation strategy of the service area, such as increasing the number of service windows during peak traffic hours and adjusting the operating power of power equipment.

[0068] Acquire historical data and real-time data of vehicle flow, use time series analysis method to determine whether the vehicle flow data has periodic fluctuation characteristics, if there is periodic fluctuation, then determine that there is eddy current characteristics, and determine input characteristics of the vehicle flow prediction model according to the analysis results of the eddy current characteristics;

[0069] Using a sensitivity analysis method, by changing the value of the input feature, determining the degree of influence of each input feature on the output result of the prediction model, and ranking the input features according to importance according to the sensitivity analysis result to obtain a feature importance ranking result;

[0070] Obtaining power load data and network loss data, using the power load data and network loss data as inputs of the prediction model, and determining the degree of influence on the prediction result by changing the values ​​of the power load data and network loss data to obtain a sensitivity analysis result of the power data;

[0071] According to the eddy current characteristic analysis results, the feature importance ranking results, the sensitivity analysis results and the power consumption data sensitivity analysis results, an integrated learning algorithm is used to build a vehicle flow prediction model, and an optimal input feature combination of the prediction model is determined by a feature selection method;

[0072] Obtaining a vehicle flow prediction result, a power load prediction result, and a network loss prediction result, taking the vehicle flow prediction result, the power load prediction result, and the network loss prediction result as input, using a multi-objective optimization algorithm to determine an operation strategy optimization model, and obtaining optimal operation strategy parameters by solving the optimization model;

[0073] To further optimize the scheme, by analyzing the historical traffic data of the service area in the past three years, it was found that the traffic data had obvious periodic fluctuation characteristics, indicating the existence of eddy current phenomenon. The wavelet transform method was used to decompose the traffic data, and the fluctuation characteristics on multiple time scales such as daily cycle, weekly cycle and seasonal cycle were extracted, and used as the input features of the traffic flow prediction model. The Sobol sensitivity analysis method was used to change the values ​​of each input feature, and it was found that historical traffic flow, weather conditions, holidays and other factors had the greatest impact on the prediction results, and the sensitivity indexes were 35, 28 and 2 respectively. At the same time, the power load data and network loss data were used as the input of the prediction model. Through the Morris sensitivity analysis method, it was found that their impact on the prediction results was relatively small, and the sensitivity indexes were 12 and 08 respectively. Based on the above analysis results, the AdaBoost ensemble learning algorithm was used to construct the traffic flow prediction model, and the five optimal input feature combinations were selected by the recursive feature elimination method, including historical traffic flow, weather conditions, holidays, power load and network loss. The genetic algorithm was used to solve the operation strategy optimization model, and the optimal operation strategy parameters such as the number of service windows and the power of power equipment were obtained. The optimized operation strategy was evaluated by system dynamics simulation technology. The results showed that increasing the number of service windows during peak traffic flow can effectively alleviate queue congestion and reduce passenger waiting time by more than 20%; optimizing the power of electrical equipment can reduce electricity costs by about 15% and improve energy efficiency. The optimized operation strategy was applied to actual business scenarios to continuously monitor changes in business indicators such as traffic flow, power load, and network loss. Statistical results show that the optimized operation strategy has achieved good results in actual applications, with the accuracy of traffic flow prediction increased to more than 95%, the error of power load prediction reduced to less than 5%, and the operating cost reduced by about 10%. At the same time, an online learning algorithm is used to dynamically update and adjust the optimization model so that it can adapt to changes in actual business scenarios and ensure the continuity and stability of the optimization effect.

[0074] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for predicting power load of a microgrid in a high-speed service area, characterized in that: The following steps are involved: Acquire historical traffic flow data and power load data of the service area, and construct a power network topology structure of the service area based on the historical traffic flow data and power load data of the service area; Analyzing the topological structure of the power network in the service area based on the power flow calculation method to obtain the corresponding relationship between network loss data and vehicle flow; Based on the multivariate nonlinear regression model, the corresponding relationship between the network loss data and the vehicle flow is estimated to obtain the quantitative mapping relationship between the vehicle flow and the load and network loss; An attention neural network is constructed based on the attention mechanism and the long-short time network model, and the attention neural network is trained through the quantitative mapping relationship between the vehicle flow and the load and the network loss to generate a multivariate time series prediction model; The power load of the microgrid in the high-speed service area is predicted based on the multivariable time series prediction model.

2. The method for predicting power load of microgrid in high-speed service area according to claim 1, characterized in that: The process of obtaining historical traffic flow data and power load data of the service area and constructing the power network topology structure of the service area based on the historical traffic flow data and power load data of the service area includes: Acquire historical vehicle flow data and power load data of the service area, and preprocess the acquired historical vehicle flow data and power load data of the service area to obtain preprocessed data; Calculating the preprocessed data by Pearson correlation coefficient method to obtain a correlation coefficient matrix; The service area power network topology structure is constructed based on the correlation coefficient matrix.

3. The method for predicting power load of microgrid in high-speed service area according to claim 2, characterized in that: The historical traffic flow data and power load data of the service area are preprocessed, and the process of obtaining the preprocessed data includes: cleaning and normalizing the historical traffic flow data and power load data of the service area, and unifying the data to the same time scale to obtain the preprocessed data.

4. The method for predicting power load of microgrid in high-speed service area according to claim 1, characterized in that: The process of analyzing the topological structure of the power network in the service area based on the power flow calculation method and obtaining the corresponding relationship between the network loss data and the vehicle flow includes: Analyze the topological structure of the power network in the service area based on a graph theory algorithm to obtain key nodes and lines with network losses greater than a preset threshold; Based on the power flow calculation method, the key nodes and lines whose network loss is greater than a preset threshold are simulated to obtain network loss distribution data; The network loss distribution data is grouped by a clustering algorithm to obtain a corresponding relationship between the network loss data and the vehicle flow.

5. The method for predicting power load of microgrid in high-speed service area according to claim 1, characterized in that: The process of obtaining the quantitative mapping relationship between vehicle flow, load and network loss includes: A multivariate nonlinear regression model is constructed based on the eddy current characteristics of vehicle flow and the corresponding relationship between the network loss data and vehicle flow; The parameters of the multivariate nonlinear regression model are estimated based on the least squares method, and the mean square error and the coefficient of determination of the regression model are calculated to determine the goodness of fit and the prediction accuracy of the regression model, and to generate a trained regression model; A quantitative mapping relationship between the vehicle flow, load and network loss is generated based on the trained regression model.

6. The method for predicting power load of microgrid in high-speed service area according to claim 1, characterized in that: The process of constructing an attention neural network based on the attention mechanism and the long-short time network model includes: Construct the attention mechanism module and the initial long-short time network model; The attention mechanism module is deployed to each time step of the long-short time network model to generate the attention neural network.

7. The method for predicting power load of microgrid in high-speed service area according to claim 1, characterized in that: The process of training the attention neural network through the quantitative mapping relationship between the vehicle flow, load and network loss to generate a multivariate time series prediction model includes: Constructing a data set based on the quantitative mapping relationship between the vehicle flow and the load and network loss; Training the attention neural network using the data set, calculating prediction values ​​and loss functions through forward propagation, and updating weight parameters of the model using a backpropagation algorithm; The performance of the model is evaluated on the data set, and the model is optimized by adjusting the number of hidden layers, the number of neurons, and the learning rate hyperparameters to generate the multivariate time series prediction model.

8. The method for predicting power load of microgrid in high-speed service area according to claim 7, characterized in that: The process of model optimization by adjusting the number of hidden layers, the number of neurons, and the learning rate hyperparameters includes: using the Adam optimization algorithm to train the model, setting the learning rate to 0.01, the batch size to 24, training for 50 epochs, and evaluating the model performance on the validation set after each epoch, recording the root mean square error.

9. The method for predicting power load of microgrid in high-speed service area according to claim 8, characterized in that: The number of hidden layers of the multivariate time series prediction model is 2, and the number of neurons is 128 and 64 respectively.

10. The method for predicting power load of microgrid in high-speed service area according to claim 1, characterized in that: The process of predicting the microgrid power load in the high-speed service area based on the multivariable time series prediction model includes: Obtaining output results of a multivariate time series prediction model, performing sensitivity analysis on the output results, and obtaining sensitivity analysis results; The sensitivity analysis results are processed using data visualization technology to generate visualized prediction results.

Images