Power grid load prediction method based on deep learning
Through the combination of deep learning models DBN and RBM/BP neural network, the problems of low accuracy and complexity of existing load prediction methods are solved, high-precision prediction of grid load is achieved, and the grid stability and regulation capabilities are improved.
Patent Information
- Application Number
- CN202411121627.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-15
- Publication Date
- 2024-11-15
AI Technical Summary
Among the existing load prediction methods, the statistical method has low accuracy and the traditional machine learning method is highly complex, which cannot meet the accuracy requirements of smart grid load prediction.
Using a deep learning-based grid load prediction method, using a deep confidence network (DBN) model, combined with RBM and BP neural networks, outlier samples are eliminated through data cleaning, transformation and screening, and initial weights are trained using the improved sparrow algorithm, and the initial weights are trained through inverse Fine-tune the propagation to achieve accurate prediction of grid load.
It improves the accuracy of grid load prediction and model training efficiency, and enhances the stability and reliability control capabilities of grid power systems.
Smart Images

Figure CN118970938A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of power grid load prediction. Background Art
[0002] The grid load refers to the total amount of electric power that the electric energy users' electrical equipment draws from the power system at a certain moment. The regional grid load is calculated based on the total load within the dispatching range of the regional grid power dispatching and control center. With economic growth, production efficiency continues to improve, and the number of electrical equipment increases, the load on the grid continues to increase. Therefore, in order to effectively improve the efficiency of power companies in controlling and analyzing grid operations in different regions, it is necessary to predict the load change trend of the grid and detect abnormalities in the grid load to ensure stable operation of the grid and balance energy supply and demand.
[0003] Load forecasting is to use historical data of power load to scientifically predict future power load by referring to economic and social development, meteorology and other factors. At present, the main load forecasting methods are statistical methods and traditional machine learning methods. Due to the low precision of statistical methods and the high complexity of traditional machine learning methods, they cannot meet the accuracy requirements of smart grid load forecasting. Summary of the invention
[0004] The present invention aims to solve the problems of low accuracy of statistical methods and high complexity of traditional machine learning methods in existing load forecasting, and now provides a power grid load forecasting method based on deep learning.
[0005] A power grid load forecasting method based on deep learning, comprising:
[0006] Input the current grid load parameters into the deep learning model to obtain the grid load forecast results;
[0007] The deep learning model is a DBN model, the input of the DBN model is a power grid load parameter, the power grid load parameter includes a time-series climate of the power grid region and power grid operating equipment parameters, and the output of the DBN model is the power grid load.
[0008] Furthermore, the training process of the above deep learning model includes:
[0009] S1: Collect the power grid load parameters under various historical working conditions to form a data set, and randomly divide the parameters in the data set into a test set and a training set in a ratio of 3:7;
[0010] S2: Use all the data in the training set to build a DBN model and train it to obtain a preliminary training model;
[0011] S3: Input all the data in the test set into the preliminary training model, and determine whether the accuracy of the result obtained by the preliminary training model meets the standard. If yes, complete the training to obtain the trained DBN model, otherwise execute S4;
[0012] S4: Randomly divide the parameters in the data set into a test set and a training set in a ratio of 2:8, and then return to S2.
[0013] Furthermore, before dividing the parameters in the above data set, the data in the data set is first cleaned, transformed and outlier sample data is screened out.
[0014] Furthermore, the above data cleaning includes: removing irrelevant data, redundant data and information affected by noise in the data sample;
[0015] The data transformation includes: normalizing the data samples;
[0016] The method of screening out outlier sample data includes: using a K-means-based clustering algorithm, first calculating the centroid of each cluster of data, then calculating the shortest centroid length and relative length of each data element, and comparing the critical value with the relative length. If the relative length corresponding to the data item is greater than the critical value, the data item is regarded as an outlier sample and screened out.
[0017] Furthermore, the DBN model includes multiple RBM networks and BP neural networks, wherein the RBM network is used for unsupervised pre-training, and the BP neural network is used for fine-tuning the RBM network by back propagation.
[0018] Furthermore, the improved sparrow algorithm ISSA is used to train the initial weights of the DBN model, and the RBM network is used for unsupervised pre-training to optimize the weight mapping of each layer;
[0019] The BP neural network is used to back-propagate the error information back to each RBM network for fine-tuning until the accuracy requirements are met.
[0020] The power grid load forecasting method based on deep learning described in the present invention realizes the prediction of power grid load by utilizing the DBN model, thereby effectively enhancing the stability and reliability control capability of the power grid power system. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 It is the DBN network structure diagram;
[0022] Figure 2 This is a flowchart of a power grid load forecasting method based on deep learning as described in a specific implementation method 1. DETAILED DESCRIPTION
[0023] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict.
[0024] Deep learning (DL) is an artificial intelligence method that mimics the working mode of the human brain when processing data and making decisions. A deep learning network is a subset of artificial intelligence learning that has the ability to learn unsupervised from unstructured or unlabeled data. The deep learning method aims to obtain semantic information features with higher-level structures in higher-level structures through stacked operations. Automatic learning at multiple levels of abstraction directly maps data from input to output without relying entirely on manually created functions. In deep learning, different types of operations are called "layers" in the deep learning model, and the process of obtaining the target loss through multiple layers of operations is called "feedforward operation". By calculating the error or loss between the predicted value and the true value, the error or loss is fed back from the last layer layer by layer by means of the back propagation algorithm to update the parameters of each layer, which is called "back propagation". Through repeated feedforward operations and back propagation, the deep learning model obtains the model parameters that best suit the target task. This process is called the "model training process", and the model with the optimal parameters is called the "converged model".
[0025] Deep Belief Network (DBN) is a neural network model in deep learning methods. This model is based on the restricted Boltzmann machine and uses multiple RBMs to achieve probability generation. Its probability generation method is mainly through constructing a joint distribution function, which is between the input data and the sample label. Figure 1 As shown, the hidden layer is connected to the next hidden layer, and the last hidden layer is connected to the visible layer, but the units in the same layer are not connected. Unlike RBM, DBN has multiple hidden layers, which can perform higher-level correlation operations on the data. This also makes the hidden layers before and after training have more order. After the previous hidden layer is fully trained, the next hidden layer can be trained again with the training data obtained from the previous hidden layer. Repeat this process and finally obtain the trained data. These characteristics make the model quite extensible and flexible. Therefore, the deep learning model selected in this embodiment is the DBN model.
[0026] Specific implementation method 1: refer to Figure 2Specifically describing this embodiment, a power grid load prediction method based on deep learning described in this embodiment includes:
[0027] S1: Collecting historical grid load parameters under various working conditions to form a data set, wherein the grid load parameters include the climate of the grid region and grid operation equipment parameters with a time series.
[0028] S2: Clean, transform and filter out outlier sample data in the data set.
[0029] The data cleaning includes: clearing irrelevant data, redundant data and information affected by noise in the data sample;
[0030] The data transformation includes: normalizing the data samples;
[0031] The method of screening out outlier sample data includes: using a K-means-based clustering algorithm, first calculating the centroid of each cluster of data, then calculating the shortest centroid length and relative length of each data element, and comparing the critical value with the relative length. If the relative length corresponding to the data item is greater than the critical value, the data item is regarded as an outlier sample and screened out.
[0032] S3: The parameters in the dataset are randomly divided into a test set and a training set in a ratio of 3:7.
[0033] S4: Use all the data in the training set to build a DBN model and train it to obtain a preliminary training model.
[0034] The DBN model includes multiple RBM networks and BP neural networks, the RBM networks are used for unsupervised pre-training, and the BP neural network is used for back propagation to fine-tune the RBM networks.
[0035] S5: Input all the data in the test set into the preliminary training model, and determine whether the accuracy of the results obtained by the preliminary training model meets the standard. If yes, complete the training to obtain the trained DBN model and then execute S7; otherwise, execute S6.
[0036] S6: Randomly divide the parameters in the data set into a test set and a training set in a ratio of 2:8, and then return to S3.
[0037] S7: Input the current grid load parameters into the deep learning model to obtain the grid load prediction results.
[0038] Specific implementation method 2: This implementation method is similar to the power grid load prediction method based on deep learning described in the implementation method 1, except that:
[0039] The improved sparrow algorithm ISSA is used to train the initial weights of the DBN model, and the RBM network is used for unsupervised pre-training to optimize the weight mapping of each layer.
[0040] The BP neural network is used to back-propagate the error information back to each RBM network for fine-tuning until the accuracy requirements are met.
[0041] Although the present invention is described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. It should therefore be understood that many modifications may be made to the exemplary embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the various dependent claims and features described herein may be combined in a manner different from that described in the original claims. It should also be understood that the features described in conjunction with a single embodiment may be used in other described embodiments.
Claims
1. A power grid load forecasting method based on deep learning, characterized in that: include: Input the current grid load parameters into the deep learning model to obtain the grid load forecast results; The deep learning model is a DBN model, the input of the DBN model is a power grid load parameter, the power grid load parameter includes a time-series climate of the power grid region and power grid operating equipment parameters, and the output of the DBN model is the power grid load.
2. The power grid load forecasting method based on deep learning according to claim 1, characterized in that: The training process of the deep learning model includes: S1: Collect the power grid load parameters under various historical working conditions to form a data set, and randomly divide the parameters in the data set into a test set and a training set in a ratio of 3:7; S2: Use all the data in the training set to build a DBN model and train it to obtain a preliminary training model; S3: Input all the data in the test set into the preliminary training model, and determine whether the accuracy of the result obtained by the preliminary training model meets the standard. If yes, complete the training to obtain the trained DBN model, otherwise execute S4; S4: Randomly divide the parameters in the data set into a test set and a training set in a ratio of 2:8, and then return to S2.
3. The power grid load forecasting method based on deep learning according to claim 2, characterized in that: Before dividing the parameters in the data set, the data in the data set is first cleaned, transformed, and outlier sample data is screened out.
4. The method for power grid load forecasting based on deep learning according to claim 3, characterized in that: The data cleaning includes: clearing irrelevant data, redundant data and information affected by noise in the data sample; The data transformation includes: normalizing the data samples; The method of screening out outlier sample data includes: using a K-means-based clustering algorithm, first calculating the centroid of each cluster of data, then calculating the shortest centroid length and relative length of each data element, and comparing the critical value with the relative length. If the relative length corresponding to the data item is greater than the critical value, the data item is regarded as an outlier sample and screened out.
5. A method for forecasting power grid load based on deep learning according to claim 1, 2, 3 or 4, characterized in that: The DBN model includes multiple RBM networks and a BP neural network. The RBM network is used for unsupervised pre-training, and the BP neural network is used for back propagation to fine-tune the RBM network.
6. The method for power grid load forecasting based on deep learning according to claim 1, characterized in that: The improved sparrow algorithm ISSA is used to train the initial weights of the DBN model, and the RBM network is used for unsupervised pre-training to optimize the weight mapping of each layer; The BP neural network is used to back-propagate the error information back to each RBM network for fine-tuning until the accuracy requirements are met.
