Power distribution network weight and overload probability prediction method and device, equipment and storage medium

Abstract

The invention relates to a power distribution network weight and overload probability prediction method, device and equipment, a storage medium and a computer program product. The method comprises the following steps: respectively establishing a power distribution network general load training set and an electric vehicle charging load training set; respectively constructing a general load extreme learning machine prediction model and a charging load extreme learning machine prediction model according to the power distribution network general load training set and the electric vehicle charging load training set, and respectively carrying out model training; based on the trained general load extreme learning machine prediction model and the trained charging load extreme learning machine prediction model, respectively carrying out calculation by combining a quantile regression model, and respectively obtaining a power distribution network general load prediction result and an electric vehicle charging load prediction result at the prediction moment; and according to the power distribution network general load prediction result and the electric vehicle charging load prediction result at the prediction moment, calculating to obtain a power distribution network heavy overload probability prediction result. The method can improve the stability of the whole power distribution network.

Description

Technical Field

[0001] The present application relates to the technical field of power load prediction for distribution networks, and in particular to a method, apparatus, device, storage medium, and computer program product for predicting the probability of heavy overload in distribution networks. Background Art

[0002] With the rapid growth in the number of electric vehicles, the increased load power of electric vehicle charging stations has led to an increase in the total load and line flow of the distribution network, exposing the power of the distribution network's transformers to the risk of severe overload. Specifically, when the load power of electric vehicle charging stations increases, the line flow in the power system will also increase. This is because electric vehicle charging stations require more electricity to meet the charging needs of electric vehicles, which in turn leads to an increase in the current through the transmission lines. The increase in line flow will cause an increase in the voltage drop on the line, which in turn affects the voltage stability of the entire distribution network.

[0003] Currently, existing technologies lack a comprehensive solution for predicting the probability of severe overloads in distribution networks, taking into account both the general load of the distribution network and the load of electric vehicle charging stations. Therefore, how to comprehensively consider the impact of both the general load and the load of electric vehicle charging stations on the distribution network and predict the probability of severe overloads in the distribution network to improve the stability of the entire distribution network is a pressing technical challenge. Summary of the Invention

[0004] Based on this, it is necessary to provide a distribution network heavy overload probability prediction method, device, equipment, storage medium and computer program product to address the above technical problems, which can comprehensively consider the impact of the general load of the distribution network and the load of electric vehicle charging stations on the distribution network, and realize the prediction of the distribution network heavy overload probability to improve the stability of the entire distribution network.

[0005] In a first aspect, the present application provides a method for predicting the probability of severe overload in a distribution network, the method comprising:

[0006] Collect characteristic data related to general load forecasting of distribution network and characteristic data related to electric vehicle charging load forecasting respectively, and establish general load training sets for distribution network and electric vehicle charging load training sets respectively;

[0007] Constructing extreme learning machine prediction models based on the distribution network general load training set and the electric vehicle charging load training set, respectively, and performing model training to obtain a trained general load extreme learning machine prediction model and a trained charging load extreme learning machine prediction model;

[0008] Based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model combined with the quantile regression model, calculations are performed respectively to obtain the general load prediction result of the distribution network and the charging load prediction result of the electric vehicle at the prediction time;

[0009] The distribution network heavy overload probability prediction result is calculated based on the general load prediction result of the distribution network and the electric vehicle charging load prediction result at the prediction time.

[0010] Furthermore, the model training includes:

[0011] Selecting a matching number of hidden layer neuron nodes according to the distribution network general load training set or the electric vehicle charging load training set;

[0012] Randomly generate the input hidden layer weight and input hidden layer bias value corresponding to each hidden layer neuron node;

[0013] According to the input hidden layer weight, the input hidden layer bias value, the hidden layer output weight, the actual input value and the actual output value, a single-layer feedforward neural network model with the number of hidden layer neuron nodes is constructed through an excitation function and trained.

[0014] Furthermore, the method constructs a single-layer feedforward neural network model having the number of hidden layer neuron nodes through an activation function according to the input hidden layer weight, the input hidden layer bias value, the hidden layer output weight, the actual input value, and the actual output value, and performs training, including:

[0015] Calculate the output hidden layer matrix;

[0016] Solving the generalized inverse matrix of the output hidden layer matrix;

[0017] The hidden layer output weight specific solution is calculated based on the generalized inverse matrix and the output matrix.

[0018] Furthermore, the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model are combined with a quantile regression model to respectively perform calculations to obtain the general load prediction results of the distribution network and the electric vehicle charging load prediction results at the prediction time, respectively, including:

[0019] Calculating, by gradient descent method, quantile regression models based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model, respectively, to obtain corresponding general load quantile sequences and charging load quantile sequences;

[0020] Calculate the general load cumulative probability distribution function and the corresponding general load probability density distribution function of the distribution network at the prediction time according to the general load quantile sequence;

[0021] The cumulative probability distribution function of the electric vehicle charging load at the predicted moment and the corresponding charging load probability density distribution function are calculated based on the charging load quantile sequence.

[0022] Furthermore, the distribution network heavy overload probability prediction result is calculated based on the general load prediction result of the distribution network and the electric vehicle charging load prediction result at the prediction time, including:

[0023] According to the general load cumulative probability distribution function and the general load probability density distribution function as well as the charging load cumulative probability distribution function and the charging load probability density distribution function, the total load probability density distribution function of the distribution network at the prediction moment is calculated by a convolution formula.

[0024] Furthermore, the calculation of the distribution network heavy overload probability prediction result based on the general load prediction result of the distribution network and the electric vehicle charging load prediction result at the prediction time further includes:

[0025] Calculate the total load cumulative probability distribution function of the total load power of the distribution network according to the total load probability density distribution function;

[0026] The distribution network heavy overload probability prediction result is calculated based on the total load cumulative probability distribution function.

[0027] In a second aspect, the present application further provides a device for predicting the probability of heavy overload in a distribution network, the device comprising:

[0028] An acquisition module is used to respectively acquire characteristic data related to the general load forecast of the distribution network and the characteristic data related to the charging load forecast of electric vehicles, and to establish a general load training set for the distribution network and a charging load training set for electric vehicles respectively;

[0029] A training module is used to construct a general load limit learning machine prediction model and a charging load limit learning machine prediction model based on the distribution network general load training set and the electric vehicle charging load training set, and perform model training respectively to obtain a trained general load limit learning machine prediction model and a trained charging load limit learning machine prediction model;

[0030] The prediction module is used to perform calculations based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model in combination with a quantile regression model to obtain the general load prediction results of the distribution network and the electric vehicle charging load prediction results at the prediction time; and is also used to calculate the distribution network heavy overload probability prediction result based on the general load prediction results of the distribution network and the electric vehicle charging load prediction results at the prediction time.

[0031] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are performed:

[0032] Collect characteristic data related to general load forecasting of distribution network and characteristic data related to electric vehicle charging load forecasting respectively, and establish general load training sets for distribution network and electric vehicle charging load training sets respectively;

[0033] Constructing extreme learning machine prediction models based on the distribution network general load training set and the electric vehicle charging load training set, respectively, and performing model training to obtain a trained general load extreme learning machine prediction model and a trained charging load extreme learning machine prediction model;

[0034] Based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model combined with the quantile regression model, calculations are performed respectively to obtain the general load prediction result of the distribution network and the charging load prediction result of the electric vehicle at the prediction time;

[0035] The distribution network heavy overload probability prediction result is calculated based on the general load prediction result of the distribution network and the electric vehicle charging load prediction result at the prediction time.

[0036] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the following steps:

[0037] Collect characteristic data related to general load forecasting of distribution network and characteristic data related to electric vehicle charging load forecasting respectively, and establish general load training sets for distribution network and electric vehicle charging load training sets respectively;

[0038] Constructing extreme learning machine prediction models based on the distribution network general load training set and the electric vehicle charging load training set, respectively, and performing model training to obtain a trained general load extreme learning machine prediction model and a trained charging load extreme learning machine prediction model;

[0039] Based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model combined with the quantile regression model, calculations are performed respectively to obtain the general load prediction result of the distribution network and the charging load prediction result of the electric vehicle at the prediction time;

[0040] The distribution network heavy overload probability prediction result is calculated based on the general load prediction result of the distribution network and the electric vehicle charging load prediction result at the prediction time.

[0041] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the following steps:

[0042] Collect characteristic data related to general load forecasting of distribution network and characteristic data related to electric vehicle charging load forecasting respectively, and establish general load training sets for distribution network and electric vehicle charging load training sets respectively;

[0043] Constructing extreme learning machine prediction models based on the distribution network general load training set and the electric vehicle charging load training set, respectively, and performing model training to obtain a trained general load extreme learning machine prediction model and a trained charging load extreme learning machine prediction model;

[0044] Based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model combined with the quantile regression model, calculations are performed respectively to obtain the general load prediction result of the distribution network and the charging load prediction result of the electric vehicle at the prediction time;

[0045] The distribution network heavy overload probability prediction result is calculated based on the general load prediction result of the distribution network and the electric vehicle charging load prediction result at the prediction time.

[0046] The embodiments of the present application have the following beneficial effects:

[0047] The distribution network heavy overload probability prediction method, device, equipment, storage medium and computer program product provided in the embodiments of the present application can avoid the computational bottleneck of the traditional neural network gradient descent method by combining the quantile regression model with the extreme learning machine prediction model, while having smaller training error, better generalization performance and faster training speed, thereby achieving quantitative prediction of the distribution network heavy overload probability and realizing refined risk assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 1 is a flow chart of a method for predicting the probability of severe overload of a distribution network according to an embodiment;

[0049] Figure 21. A schematic diagram of a flow chart of model training in a method for predicting the probability of severe overload in a distribution network according to an embodiment;

[0050] Figure 3 A schematic diagram of a specific process of model training in another embodiment;

[0051] Figure 4 1 is a flow chart of a method for calculating a quantile regression model based on an extreme learning machine prediction model in one embodiment;

[0052] Figure 5 FIG. 4 is a structural block diagram of a device for predicting the probability of heavy overload in a distribution network in one embodiment. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0054] Example 1

[0055] In one embodiment, a method for predicting the probability of heavy overload in a distribution network is provided. This embodiment uses the method applied to a terminal as an example for illustration. It is understandable that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. The terminal can be, but is not limited to, various personal computers, laptops, smart phones, tablets, Internet of Things devices, and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server can be implemented as an independent server or a server cluster consisting of multiple servers.

[0056] In this embodiment, refer to Figure 1 , the method comprises the following steps:

[0057] S1. Collect characteristic data related to general load forecasting of distribution network and characteristic data related to electric vehicle charging load forecasting, and establish general load training sets for distribution network and electric vehicle charging load training sets respectively;

[0058] S2. Construct a general load limit learning machine prediction model and a charging load limit learning machine prediction model based on the distribution network general load training set and the electric vehicle charging load training set, and perform model training on them respectively to obtain a trained general load limit learning machine prediction model and a trained charging load limit learning machine prediction model;

[0059] S3. Calculate based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model in combination with the quantile regression model to obtain the general load prediction results of the distribution network and the electric vehicle charging load prediction results at the prediction time, respectively;

[0060] S4. Calculate the distribution network heavy overload probability prediction result based on the general load prediction result of the distribution network and the electric vehicle charging load prediction result at the prediction time.

[0061] Specifically, the load types of the distribution network can be divided into general distribution network load and electric vehicle charging load. Feature data can include at least one of historical load data for general distribution network load and electric vehicle charging load, historical meteorological data, and electric vehicle ownership data. Meteorological data includes at least one of solar irradiance, temperature, rainfall, wind speed, and humidity. Feature data related to general distribution network load prediction and electric vehicle charging load prediction are collected separately, and training sets for general distribution network load and electric vehicle charging load are established. Furthermore, test and validation sets can be established for subsequent model testing and validation. Then, a general load limit learning machine prediction model and a charging load limit learning machine prediction model are constructed and trained separately to obtain the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model. Due to the different load types, general distribution network load (e.g., residential and industrial electricity load) is typically cyclical and stable, affected by factors such as weather, season, and time of day, while electric vehicle charging load is random and time-varying, affected by dynamic factors such as charging behavior, vehicle ownership, and electricity pricing policies. Separately modeling and training the general load of the distribution network and the electric vehicle charging load allows for different input features to be used for each load, more accurately reflecting the corresponding influencing mechanisms and enabling more precise predictions. Separate modeling also reduces the input dimensions of a single model, avoids feature redundancy, and reduces model complexity through feature decoupling, thereby accelerating training speed. After model training, the trained general load extreme learning machine prediction model and the trained charging load extreme learning machine prediction model are combined with a quantile regression model to calculate the general load forecast for the distribution network and the electric vehicle charging load forecast at the prediction time, respectively. By combining the quantile regression model with the extreme learning machine prediction model, the computational bottlenecks of traditional neural network gradient descent methods are avoided, while achieving smaller training errors, better generalization performance, and faster training speed. Ultimately, this method achieves a quantitative prediction of the probability of severe overload in the distribution network and enables refined risk assessment.

[0062] In one embodiment, referring to Figure 2 , model training includes:

[0063] S21. Selecting a matching number of hidden layer neuron nodes based on a general load training set for a distribution network or a charging load training set for an electric vehicle;

[0064] S22, randomly generating an input hidden layer weight and an input hidden layer bias value corresponding to each hidden layer neuron node;

[0065] S23. According to the input hidden layer weight, the input hidden layer bias value, the hidden layer output weight, the actual input value and the actual output value, a single-layer feedforward neural network model with the number of hidden layer neuron nodes is constructed through an activation function and trained.

[0066] Specifically, the general load of the distribution network and the charging load of electric vehicles are modeled separately and trained, and single-layer feedforward neural network models with different numbers of hidden layer neuron nodes are constructed. Among them, the general load of the distribution network corresponds to the first single-layer feedforward neural network model with the first number of hidden layer neuron nodes; the charging load of electric vehicles corresponds to the second single-layer feedforward neural network model with the second number of hidden layer neuron nodes. The difference between the first single-layer feedforward neural network model and the second single-layer feedforward neural network model is that differentiated input features are used for the general load of the distribution network and the charging load of electric vehicles, respectively, to more accurately distinguish and reflect the influence mechanism of the corresponding load, thereby achieving more accurate prediction. In addition, for a single model, it can reduce the feature input dimension, avoid feature redundancy, reduce noise, and reduce model complexity through feature decoupling, thereby improving training speed.

[0067] For example, assume there are N discrete training samples ,in , the single-layer feedforward neural network model can be expressed as:

[0068]

[0069] Where L is the number of hidden layer neuron nodes, g(m) is the activation function, and m in this embodiment is , Used to describe the above-mentioned input hidden layer weights (i.e., the weight values ​​of the input layer neurons and the jth neuron in the hidden layer), Used to describe the above-mentioned input hidden layer bias value (i.e. the bias value between the input layer neuron and the jth neuron in the hidden layer), is the actual input value, is the actual output value, Used to describe the weight value of the jth neuron in the hidden layer and the output layer.

[0070] For example, taking a single-layer feedforward neural network with L hidden layer neuron nodes as an example, N discrete training samples from the same continuous nonlinear system can be approximated by zero error. , then the above formula can be expressed as:

[0071]

[0072] The matrix form can be expressed as:

[0073]

[0074] The hidden layer output matrix can be expressed as:

[0075]

[0076] The hidden layer output weight matrix can be expressed as:

[0077]

[0078] And the fitted output matrix

[0079]

[0080] By adopting this technical solution, a first single-layer feedforward neural network model with a first hidden layer number of neuron nodes can be established for general distribution network loads, and a second single-layer feedforward neural network model with a second hidden layer number of neuron nodes can be established for electric vehicle charging loads. Differentiated input features can be used for general distribution network loads and electric vehicle charging loads, respectively, to more accurately distinguish and reflect the impact mechanisms of the corresponding loads, thereby more accurately predicting general distribution network loads and electric vehicle charging loads. At the same time, for each model, the feature input dimension can be reduced, feature redundancy can be avoided, noise can be reduced, and model complexity can be reduced through feature decoupling, thereby improving training speed.

[0081] In one embodiment, referring to Figure 3 , S23 includes:

[0082] S231, calculating the output hidden layer matrix;

[0083] S232, solving the generalized inverse matrix of the output hidden layer matrix;

[0084] S233. Obtain the specific solution of the hidden layer output weights based on the generalized inverse matrix and the output matrix.

[0085] Specifically, in the training process of the extreme learning machine, the first step is to give any given input hidden layer weight value and input hidden layer bias value. The input hidden layer weight value and input hidden layer bias value will not change during the model training process. Then, the activation function is used to map them to a high-dimensional space. The whole solution process is similar to solving a linear system. A least squares particular solution .

[0086] It can be expressed as:

[0087]

[0088] According to the theory of generalized inverse, the solution is It can be expressed as:

[0089]

[0090] in, is the Moore-Penrose generalized inverse of the hidden layer output matrix H.

[0091] For example, the performance of the extreme learning machine regression model is largely subject to the activation function and input data. The activation function can be the following Sigmoid kernel function:

[0092]

[0093] For example, the appropriate number of hidden layer neuron nodes can be selected for the general load of the distribution network or the charging load of electric vehicles, and then the input hidden layer weight value and input hidden layer bias value of [-1,1] are randomly generated. The above-mentioned Sigmoid kernel function is selected as the excitation function, the output hidden layer matrix H is calculated, and then the generalized inverse matrix of the output hidden layer matrix H is solved. , and finally calculate the least squares solution By adopting this technical solution, the general load of the distribution network or the electric vehicle charging load can be trained using the extreme learning machine model separately. Through the random initialization of the hidden layer parameters and the linear solution mechanism of the extreme learning machine model, the training time of a single model (the first single-layer feedforward neural network model or the second single-layer feedforward neural network model) is greatly reduced compared to traditional neural networks. At the same time, the separate modeling of the general load of the distribution network or the electric vehicle charging load further avoids high-dimensional data processing, significantly improving the overall training efficiency and preventing the cross-propagation of errors in the comprehensive prediction, thereby improving the overall prediction stability.

[0094] In one embodiment, referring to Figure 4 , S3 includes:

[0095] S31. Calculate, by gradient descent method, quantile regression models based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model, respectively, to obtain corresponding general load quantile sequences and charging load quantile sequences;

[0096] S32. Calculate the general load cumulative probability distribution function and the corresponding general load probability density distribution function of the general load of the distribution network at the prediction time according to the general load quantile sequence;

[0097] S33. Calculate the charging load cumulative probability distribution function and the corresponding charging load probability density distribution function of the electric vehicle charging load at the prediction moment based on the charging load quantile sequence.

[0098] For example, the predicted power (the power of the general load or the electric vehicle charging station load) at the prediction time (for example, time t) is recorded as , the input factors of the prediction model at the corresponding prediction time are recorded as vector .when When the power is to be predicted, there will inevitably be errors in the result of point prediction. There is uncertainty in the valuation of is considered as a random variable, and its probability density distribution function and cumulative distribution function are respectively denoted as and ,but The τth quantile (denoted as ) is defined as:

[0099]

[0100] Wherein, the value range of τ is [0,1].

[0101] According to the definition of the cumulative distribution function, the above formula can also be equivalently expressed as:

[0102]

[0103] Among them, the τth quantile can be obtained by minimizing the following asymmetric weighted loss function:

[0104]

[0105] Where N is the number of samples in the data set used to construct the quantile regression model. It is called the test function, and its expression is:

[0106]

[0107] For conditional quantiles, let the independent variable vector be , the parameters of the regression model are , then the corresponding quantile can be expressed as .

[0108] At this point, the optimization problem for solving quantile regression is:

[0109]

[0110] The penalty term is introduced into the above formula to prevent the network structure from falling into an overfitting state, and the optimization problem for solving quantile regression is updated as follows:

[0111]

[0112] in, is the Frobenius norm of the matrix, which is the square root of the sum of the squares of all the elements of the matrix.

[0113] The following is an equivalent deduction of the optimization problem of quantile regression:

[0114]

[0115] make

[0116]

[0117] Use gradient descent to solve the optimization problem of quantile regression:

[0118]

[0119] The demand is to take its derivative:

[0120]

[0121] For the regression model, it is a prediction model based on extreme learning machine:

[0122]

[0123] Therefore for and The derivatives of are:

[0124]

[0125]

[0126] Therefore, the gradient descent method is used to solve the optimization problem of quantile regression, and the parameters of the extreme learning machine model are The update formula is:

[0127]

[0128] Where, is the number of iterations, is the learning rate of the gradient descent method. The increase, when The change is less than a certain threshold Right now When , the gradient descent method can be considered to have converged. Optimize the parameters of the extreme learning machine model in the quantile regression problem The solution.

[0129] For new model input , the first The quantile can be calculated as follows:

[0130]

[0131] In the probability range [0,1], the Uniformly take values ​​and get the probability sequence ,in , , respectively solve the corresponding quantile regression optimization problem based on the extreme learning machine model and obtain the quantile sequence:

[0132]

[0133] Using polynomial fitting of quantile sequence points, the model input can be obtained The corresponding output results The cumulative probability distribution function of :

[0134]

[0135] Where M is the highest order term, is the coefficient of the m-th order term, 、 Corresponding respectively 、 point.

[0136] remember The power of general load and electric vehicle charging station load at the moment is and , respectively solve Cumulative probability distribution function of general load and electric vehicle charging station load power at each moment and :

[0137]

[0138] The corresponding probability density distribution function and for:

[0139]

[0140] By adopting this technical solution, the corresponding quantile sequence can be calculated based on different extreme learning machine prediction models combined with the quantile regression model, and then the cumulative probability distribution function and probability density function of general load and charging station load can be obtained. The uncertainty of load prediction can be quantified through the quantile regression model. In addition, the extreme learning machine is combined with random weight initialization and generalized inverse solution. Compared with the traditional gradient descent neural network method, it has smaller training error, better generalization performance, and faster training speed, thereby shortening the solution time of probability prediction.

[0141] In one embodiment, S4 includes:

[0142] S41. According to the general load cumulative probability distribution function and the general load probability density distribution function as well as the charging load cumulative probability distribution function and the charging load probability density distribution function, the probability density distribution function of the total load of the distribution network at the prediction time is calculated by a convolution formula.

[0143] For example, The total load power of the distribution network at the moment is , the corresponding probability density distribution function is When the distribution network loss is ignored:

[0144]

[0145] According to the convolution formula, under the assumption and For independent random variables, calculate :

[0146]

[0147] By adopting this technical solution, the general load conditions of the distribution network and the load conditions of electric vehicle charging stations can be comprehensively considered. The general load calculation of the distribution network and the load calculation of the electric vehicle charging station are relatively independent and do not interfere with each other. The probability distribution of the two types of loads is integrated through the convolution formula to obtain the probability density function of the total load of the distribution network. The load conditions are comprehensively considered without affecting the training speed, thereby improving the overall load prediction accuracy of the distribution network.

[0148] In one embodiment, S4 further includes:

[0149] S42. Calculate the total load cumulative probability distribution function of the total load power of the distribution network according to the total load probability density distribution function;

[0150] S43. Calculate the distribution network heavy overload probability prediction result based on the total load cumulative probability distribution function.

[0151] For example, the heavy overload capacity threshold of the distribution network transformer is , then the active power threshold of heavy overload for:

[0152]

[0153] in, is the maximum load rate of the distribution network, is the power factor of the distribution network.

[0154] Since the cumulative probability distribution function of the total load power of the distribution network is for:

[0155]

[0156] Therefore, the cumulative probability distribution function of the total load power of the distribution network can be used to determine whether the total load of the distribution network is greater than or equal to The probability of distribution network overload probability prediction result is:

[0157]

[0158] By adopting such a technical solution, the distribution network heavy overload probability prediction result can be obtained by combining the heavy overload capacity threshold and the cumulative probability distribution of the total load power of the distribution network, and the distribution network heavy overload probability can be quantitatively predicted, providing reliable technical support for the distribution network heavy overload risk assessment.

[0159] In this embodiment, the general load of the distribution network and the electric vehicle charging load can be modeled and trained separately. This allows different loads to use differentiated input features, thereby more accurately reflecting the corresponding influencing mechanisms and achieving more accurate predictions. Separate modeling can also reduce the input dimensions of a single model, avoid feature redundancy, and reduce model complexity through feature decoupling, thereby improving training speed. After model training, the trained general load extreme learning machine prediction model and the trained charging load extreme learning machine prediction model are combined with a quantile regression model to perform calculations, respectively, to obtain the general load prediction results and electric vehicle charging load prediction results at the prediction time. By combining the quantile regression model with the extreme learning machine prediction model, the computational bottlenecks of traditional neural network gradient descent methods can be avoided, while achieving smaller training errors, better generalization performance, and faster training speed. Ultimately, a quantitative prediction of the probability of severe overload in the distribution network can be achieved, enabling refined risk assessment.

[0160] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0161] Example 2

[0162] Based on the same inventive concept, an embodiment of the present application further provides a distribution network heavy overload probability prediction device for implementing the distribution network heavy overload probability prediction method involved above. The implementation solution provided by this device is similar to the implementation solution described in the above method, so the specific limitations in one or more distribution network heavy overload probability prediction device embodiments provided below can be found in the above limitations of the distribution network heavy overload probability prediction method, and will not be repeated here.

[0163] In one embodiment, Figure 5 As shown, a distribution network heavy overload probability prediction device is provided, including: an acquisition module, a training module and a prediction module, wherein:

[0164] An acquisition module is used to respectively acquire characteristic data related to the general load forecast of the distribution network and the characteristic data related to the charging load forecast of electric vehicles, and to establish a general load training set for the distribution network and a charging load training set for electric vehicles respectively;

[0165] A training module is used to construct a general load limit learning machine prediction model and a charging load limit learning machine prediction model based on the distribution network general load training set and the electric vehicle charging load training set, and perform model training respectively to obtain a trained general load limit learning machine prediction model and a trained charging load limit learning machine prediction model;

[0166] The prediction module is used to perform calculations based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model in combination with a quantile regression model to obtain the general load prediction results of the distribution network and the electric vehicle charging load prediction results at the prediction time; and is also used to calculate the distribution network heavy overload probability prediction result based on the general load prediction results of the distribution network and the electric vehicle charging load prediction results at the prediction time.

[0167] An acquisition module is used to respectively acquire characteristic data related to the general load forecast of the distribution network and the characteristic data related to the charging load forecast of electric vehicles, and to establish a general load training set for the distribution network and a charging load training set for electric vehicles respectively;

[0168] Furthermore, the training module is also used to select a matching number of hidden layer neuron nodes based on the general load training set of the distribution network or the electric vehicle charging load training set; and to randomly generate input hidden layer weights and input hidden layer bias values ​​corresponding to each hidden layer neuron node; and is also used to construct a single-layer feedforward neural network model with the said number of hidden layer neuron nodes through an excitation function based on the said input hidden layer weights, the said input hidden layer bias values, the hidden layer output weights, the actual input values ​​and the actual output values ​​and to perform training.

[0169] Furthermore, the training module is also used to calculate the output hidden layer matrix; and to solve the generalized inverse matrix of the output hidden layer matrix; and to calculate the special solution of the hidden layer output weight based on the generalized inverse matrix and the output matrix.

[0170] Furthermore, the prediction module is also used to calculate the quantile regression model based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model through the gradient descent method to obtain the corresponding general load quantile sequence and charging load quantile sequence; and is used to calculate the general load cumulative probability distribution function of the distribution network and the corresponding general load probability density distribution function at the prediction moment based on the general load quantile sequence; and is also used to calculate the cumulative probability distribution function of the electric vehicle charging load and the corresponding charging load probability density distribution function at the prediction moment based on the charging load quantile sequence.

[0171] Furthermore, the prediction module is also used to calculate the probability density distribution function of the total load of the distribution network at the prediction time through a convolution formula based on the general load cumulative probability distribution function and the general load probability density distribution function as well as the charging load cumulative probability distribution function and the charging load probability density distribution function.

[0172] Furthermore, the prediction module is also used to calculate the total load cumulative probability distribution function of the total load power of the distribution network based on the total load probability density distribution function; and to calculate the distribution network heavy overload probability prediction result based on the total load cumulative probability distribution function.

[0173] Each module in the above-mentioned distribution network severe overload probability prediction device can be implemented in whole or in part through software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the corresponding operations of each of the above modules.

[0174] Example 3

[0175] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0176] Collect characteristic data related to general load forecasting of distribution network and characteristic data related to electric vehicle charging load forecasting respectively, and establish general load training sets for distribution network and electric vehicle charging load training sets respectively;

[0177] Constructing extreme learning machine prediction models based on the distribution network general load training set and the electric vehicle charging load training set, respectively, and performing model training to obtain a trained general load extreme learning machine prediction model and a trained charging load extreme learning machine prediction model;

[0178] Based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model combined with the quantile regression model, calculations are performed respectively to obtain the general load prediction result of the distribution network and the charging load prediction result of the electric vehicle at the prediction time;

[0179] The distribution network heavy overload probability prediction result is calculated based on the general load prediction result of the distribution network and the electric vehicle charging load prediction result at the prediction time.

[0180] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0181] Selecting a matching number of hidden layer neuron nodes according to the distribution network general load training set or the electric vehicle charging load training set;

[0182] Randomly generate the input hidden layer weight and input hidden layer bias value corresponding to each hidden layer neuron node;

[0183] According to the input hidden layer weight, the input hidden layer bias value, the hidden layer output weight, the actual input value and the actual output value, a single-layer feedforward neural network model with the number of hidden layer neuron nodes is constructed through an excitation function and trained.

[0184] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0185] Calculate the output hidden layer matrix;

[0186] Solving the generalized inverse matrix of the output hidden layer matrix;

[0187] The hidden layer output weight specific solution is calculated based on the generalized inverse matrix and the output matrix.

[0188] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0189] Calculating, by gradient descent method, quantile regression models based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model, respectively, to obtain corresponding general load quantile sequences and charging load quantile sequences;

[0190] Calculate the general load cumulative probability distribution function and the corresponding general load probability density distribution function of the distribution network at the prediction time according to the general load quantile sequence;

[0191] The cumulative probability distribution function of the electric vehicle charging load at the predicted moment and the corresponding charging load probability density distribution function are calculated based on the charging load quantile sequence.

[0192] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0193] According to the general load cumulative probability distribution function and the general load probability density distribution function as well as the charging load cumulative probability distribution function and the charging load probability density distribution function, the total load probability density distribution function of the distribution network at the prediction moment is calculated by a convolution formula.

[0194] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0195] Calculate the total load cumulative probability distribution function of the total load power of the distribution network according to the total load probability density distribution function;

[0196] The distribution network heavy overload probability prediction result is calculated based on the total load cumulative probability distribution function.

[0197] Example 4

[0198] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0199] Collect characteristic data related to general load forecasting of distribution network and characteristic data related to electric vehicle charging load forecasting respectively, and establish general load training sets for distribution network and electric vehicle charging load training sets respectively;

[0200] Constructing extreme learning machine prediction models based on the distribution network general load training set and the electric vehicle charging load training set, respectively, and performing model training to obtain a trained general load extreme learning machine prediction model and a trained charging load extreme learning machine prediction model;

[0201] Based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model combined with the quantile regression model, calculations are performed respectively to obtain the general load prediction result of the distribution network and the charging load prediction result of the electric vehicle at the prediction time;

[0202] The distribution network heavy overload probability prediction result is calculated based on the general load prediction result of the distribution network and the electric vehicle charging load prediction result at the prediction time.

[0203] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0204] Selecting a matching number of hidden layer neuron nodes according to the distribution network general load training set or the electric vehicle charging load training set;

[0205] Randomly generate the input hidden layer weight and input hidden layer bias value corresponding to each hidden layer neuron node;

[0206] According to the input hidden layer weight, the input hidden layer bias value, the hidden layer output weight, the actual input value and the actual output value, a single-layer feedforward neural network model with the number of hidden layer neuron nodes is constructed through an excitation function and trained.

[0207] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0208] Calculate the output hidden layer matrix;

[0209] Solving the generalized inverse matrix of the output hidden layer matrix;

[0210] The hidden layer output weight specific solution is calculated based on the generalized inverse matrix and the output matrix.

[0211] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0212] Calculating, by gradient descent method, quantile regression models based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model, respectively, to obtain corresponding general load quantile sequences and charging load quantile sequences;

[0213] Calculate the general load cumulative probability distribution function and the corresponding general load probability density distribution function of the distribution network at the prediction time according to the general load quantile sequence;

[0214] The cumulative probability distribution function of the electric vehicle charging load at the predicted moment and the corresponding charging load probability density distribution function are calculated based on the charging load quantile sequence.

[0215] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0216] According to the general load cumulative probability distribution function and the general load probability density distribution function as well as the charging load cumulative probability distribution function and the charging load probability density distribution function, the total load probability density distribution function of the distribution network at the prediction moment is calculated by a convolution formula.

[0217] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0218] Calculate the total load cumulative probability distribution function of the total load power of the distribution network according to the total load probability density distribution function;

[0219] The distribution network heavy overload probability prediction result is calculated based on the total load cumulative probability distribution function.

[0220] Example 5

[0221] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0222] Collect characteristic data related to general load forecasting of distribution network and characteristic data related to electric vehicle charging load forecasting respectively, and establish general load training sets for distribution network and electric vehicle charging load training sets respectively;

[0223] Constructing extreme learning machine prediction models based on the distribution network general load training set and the electric vehicle charging load training set, respectively, and performing model training to obtain a trained general load extreme learning machine prediction model and a trained charging load extreme learning machine prediction model;

[0224] Based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model combined with the quantile regression model, calculations are performed respectively to obtain the general load prediction result of the distribution network and the charging load prediction result of the electric vehicle at the prediction time;

[0225] The distribution network heavy overload probability prediction result is calculated based on the general load prediction result of the distribution network and the electric vehicle charging load prediction result at the prediction time.

[0226] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0227] Selecting a matching number of hidden layer neuron nodes according to the distribution network general load training set or the electric vehicle charging load training set;

[0228] Randomly generate the input hidden layer weight and input hidden layer bias value corresponding to each hidden layer neuron node;

[0229] According to the input hidden layer weight, the input hidden layer bias value, the hidden layer output weight, the actual input value and the actual output value, a single-layer feedforward neural network model with the number of hidden layer neuron nodes is constructed through an excitation function and trained.

[0230] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0231] Calculate the output hidden layer matrix;

[0232] Solving the generalized inverse matrix of the output hidden layer matrix;

[0233] The hidden layer output weight specific solution is calculated based on the generalized inverse matrix and the output matrix.

[0234] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0235] Calculating, by gradient descent method, quantile regression models based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model, respectively, to obtain corresponding general load quantile sequences and charging load quantile sequences;

[0236] Calculate the general load cumulative probability distribution function and the corresponding general load probability density distribution function of the distribution network at the prediction time according to the general load quantile sequence;

[0237] The cumulative probability distribution function of the electric vehicle charging load at the predicted moment and the corresponding charging load probability density distribution function are calculated based on the charging load quantile sequence.

[0238] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0239] According to the general load cumulative probability distribution function and the general load probability density distribution function as well as the charging load cumulative probability distribution function and the charging load probability density distribution function, the total load probability density distribution function of the distribution network at the prediction moment is calculated by a convolution formula.

[0240] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0241] Calculate the total load cumulative probability distribution function of the total load power of the distribution network according to the total load probability density distribution function;

[0242] The distribution network heavy overload probability prediction result is calculated based on the total load cumulative probability distribution function.

[0243] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), data processing logic devices based on quantum computing, and the like.

[0244] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0245] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A method for predicting the probability of heavy overload in a distribution network, characterized in that: The method comprises: Collect characteristic data related to general load forecasting of distribution network and characteristic data related to electric vehicle charging load forecasting respectively, and establish general load training sets for distribution network and electric vehicle charging load training sets respectively; Constructing extreme learning machine prediction models based on the distribution network general load training set and the electric vehicle charging load training set, respectively, and performing model training to obtain a trained general load extreme learning machine prediction model and a trained charging load extreme learning machine prediction model; Based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model combined with the quantile regression model, calculations are performed respectively to obtain the general load prediction result of the distribution network and the charging load prediction result of the electric vehicle at the prediction time; The distribution network heavy overload probability prediction result is calculated based on the general load prediction result of the distribution network and the electric vehicle charging load prediction result at the prediction time.

2. The method according to claim 1, characterized in that The model training includes: Selecting a matching number of hidden layer neuron nodes according to the distribution network general load training set or the electric vehicle charging load training set; Randomly generate the input hidden layer weight and input hidden layer bias value corresponding to each hidden layer neuron node; According to the input hidden layer weight, the input hidden layer bias value, the hidden layer output weight, the actual input value and the actual output value, a single-layer feedforward neural network model with the number of hidden layer neuron nodes is constructed through an excitation function and trained.

3. The method according to claim 2, characterized in that The method comprises constructing a single-layer feedforward neural network model having the number of hidden layer neuron nodes through an activation function according to the input hidden layer weight, the input hidden layer bias value, the hidden layer output weight, the actual input value and the actual output value and performing training, including: Calculate the output hidden layer matrix; Solving the generalized inverse matrix of the output hidden layer matrix; The hidden layer output weight specific solution is calculated based on the generalized inverse matrix and the output matrix.

4. The method according to claim 1, wherein The training general load limit learning machine prediction model and the training charging load limit learning machine prediction model are combined with a quantile regression model to respectively calculate and obtain the general load prediction result of the distribution network and the charging load prediction result of the electric vehicle at the prediction time, respectively, including: Calculating, by gradient descent method, quantile regression models based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model, respectively, to obtain corresponding general load quantile sequences and charging load quantile sequences; Calculate the general load cumulative probability distribution function and the corresponding general load probability density distribution function of the distribution network at the prediction time according to the general load quantile sequence; The cumulative probability distribution function of the electric vehicle charging load at the predicted moment and the corresponding charging load probability density distribution function are calculated based on the charging load quantile sequence.

5. The method according to claim 4, characterized in that The distribution network heavy overload probability prediction result is calculated based on the general load prediction result of the distribution network at the prediction time and the electric vehicle charging load prediction result, including: According to the general load cumulative probability distribution function and the general load probability density distribution function as well as the charging load cumulative probability distribution function and the charging load probability density distribution function, the total load probability density distribution function of the distribution network at the prediction moment is calculated by a convolution formula.

6. The method according to claim 5, characterized in that The calculation of the distribution network heavy overload probability prediction result based on the general load prediction result of the distribution network and the electric vehicle charging load prediction result at the prediction time also includes: Calculate the total load cumulative probability distribution function of the total load power of the distribution network according to the total load probability density distribution function; The distribution network heavy overload probability prediction result is calculated based on the total load cumulative probability distribution function.

7. A distribution network heavy overload probability prediction device, characterized in that: The device comprises: An acquisition module is used to respectively acquire characteristic data related to the general load forecast of the distribution network and the characteristic data related to the charging load forecast of electric vehicles, and to establish a general load training set for the distribution network and a charging load training set for electric vehicles respectively; A training module is used to construct a general load limit learning machine prediction model and a charging load limit learning machine prediction model based on the distribution network general load training set and the electric vehicle charging load training set, and perform model training respectively to obtain a trained general load limit learning machine prediction model and a trained charging load limit learning machine prediction model; The prediction module is used to perform calculations based on the trained general load limit learning machine prediction model and the trained charging load limit learning machine prediction model in combination with a quantile regression model to obtain the general load prediction results of the distribution network and the electric vehicle charging load prediction results at the prediction time; and is also used to calculate the distribution network heavy overload probability prediction result based on the general load prediction results of the distribution network and the electric vehicle charging load prediction results at the prediction time.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images