Power load prediction method
The 1D-CNN-Adaboost power load prediction model addresses the limitations of existing methods by integrating multiple neural networks with an Adaboost iterative model, achieving improved accuracy and efficiency in power load forecasting.
Patent Information
- Application Number
- CN202411249003.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-06
- Publication Date
- 2024-12-13
AI Technical Summary
Existing power load forecasting methods, including statistical, machine learning, and deep learning approaches, face challenges in accurately predicting power loads due to complex nonlinear relationships and high-dimensional data, often requiring significant computational resources and failing to achieve optimal prediction results.
A 1D-CNN-Adaboost power load prediction model is constructed by integrating multiple 1D-CNN neural network models with an Adaboost iterative model, incorporating normalization and outlier removal, and a weighted combination process to enhance prediction accuracy.
The 1D-CNN-Adaboost model improves prediction precision and accuracy for large-scale power load forecasting by leveraging weighted combinations of neural networks, reducing computational inefficiencies and enhancing model performance.
Smart Images
Figure CN119129828A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power load prediction, and more particularly to a method for predicting power load. Background Art
[0002] Power load forecasting is a key link in power system planning and operation management, and its accuracy is directly related to the economy and reliability of the power system. With the continuous development of the power market and the diversification of user needs, the accuracy and efficiency of power load forecasting are becoming increasingly higher. Existing power load forecasting methods mainly include statistical methods, machine learning methods, and deep learning methods, each of which has its own unique advantages and limitations.
[0003] Statistical methods: These methods usually establish prediction models based on the statistical relationship between historical load data and related factors (such as temperature, humidity, time, etc.). Common statistical methods include linear regression, time series analysis, etc. The advantages of these methods are that the models are simple, easy to understand and implement, but they often have difficulty capturing complex nonlinear relationships in the data, so when dealing with large-scale, high-dimensional data, the prediction accuracy may be insufficient.
[0004] Machine learning methods: Machine learning methods predict future loads by learning patterns from historical data. Commonly used machine learning algorithms include support vector machines (SVM), random forests, gradient boosting machines (GBM), etc. These methods perform well in dealing with nonlinear relationships and high-dimensional data and can provide high prediction accuracy. However, machine learning models usually require a lot of parameter tuning and training data, which may lead to inefficient computation, especially when the amount of data is huge.
[0005] Deep learning methods: Deep learning methods, such as convolutional neural networks (CNN), recurrent neural networks (RNN), long short-term memory networks (LSTM), and Transformer, automatically extract data features and learn complex nonlinear relationships by building multi-layer neural networks. These methods have significant advantages in processing large-scale data and capturing the dynamic characteristics of time series data. However, a single deep learning model often does not achieve good prediction results when predicting power load. Summary of the invention
[0006] In view of this, the present invention provides a method for predicting power load, so as to improve the accuracy of power load prediction.
[0007] In order to achieve the above object, the present invention adopts the following technical solution:
[0008] The present invention first discloses a method for predicting power load, comprising the following steps:
[0009] Obtain historical power load time series data for the target area;
[0010] Preprocessing the historical power load time series data;
[0011] Constructing a 1D-CNN-Adaboost power load prediction model, wherein the 1D-CNN-Adaboost power load prediction model includes multiple 1D-CNN neural network models and an Adaboost iterative model, wherein the Adaboost iterative model is used to weightedly combine the multiple 1D-CNN neural network models to generate a strong learner for predicting the power load of the target area;
[0012] The 1D-CNN-Adaboost power load forecasting model is used to forecast the preprocessed historical power load time series data.
[0013] Furthermore, the historical power load time series data is preprocessed, specifically comprising the following steps:
[0014] Normalization processing: normalize the historical power load time series data to remove the impact of the dimension;
[0015] Data outlier processing: Use the Hampel filtering algorithm to remove outliers in historical power load time series data.
[0016] Furthermore, the 1D-CNN-Adaboost power load forecasting model is trained by the following steps:
[0017] The preprocessed historical power load time series data is used as a data set and divided into a training set and a test set;
[0018] Assign the same initial weight to each sample of the power load time series data in the training set;
[0019] The samples assigned with initial weights are input into each 1D-CNN neural network model in the 1D-CNN-Adaboost power load forecasting model for training;
[0020] The Adaboost iterative model is used to calculate the error rate of each 1D-CNN neural network model on the training set;
[0021] According to the error rate of each 1D-CNN neural network model on the training set, the model weight corresponding to each 1D-CNN neural network model is calculated;
[0022] According to the error rate of each 1D-CNN neural network model on the training set, the initial weight of each sample is iteratively updated;
[0023] The samples after the initial weight update are input into each 1D-CNN neural network model again for training, and the model weight corresponding to the 1D-CNN neural network model is calculated according to the error rate of the trained 1D-CNN neural network model on the training set, until the number of iterations meets the set threshold;
[0024] When the number of iterations meets the set threshold requirement, multiple 1D-CNN neural network models are weightedly combined according to the model weights and used as strong learners for predicting the power load in the target area.
[0025] Furthermore, according to the error rate of each 1D-CNN neural network model on the training set, the model weight corresponding to each 1D-CNN neural network model is calculated using the following calculation formula:
[0026]
[0027] Among them, αi represents the model weight of the i-th 1D-CNN neural network model; εi represents the error rate of the i-th 1D-CNN neural network model.
[0028] Furthermore, according to the error rate of each 1D-CNN neural network model on the training set, the initial weight of each sample is iteratively updated, specifically by the following formula:
[0029]
[0030] in, represents the sample weight of the jth sample in the training set after update, represents the sample weight of the jth sample in the training set before updating, represents the model weight of the j-th 1D-CNN neural network model; yj is the real sample before training, Represents the predicted data after training, and t represents the number of update iterations.
[0031] Secondly, the present invention also discloses a power load prediction system, comprising:
[0032] A data acquisition module is used to obtain the historical power load time series data of the target area;
[0033] A data preprocessing module, used for preprocessing the historical power load time series data;
[0034] The data prediction module is used to use the 1D-CNN-Adaboost power load prediction model to predict the preprocessed historical power load time series data to obtain the power load prediction result. The 1D-CNN-Adaboost power load prediction model includes multiple 1D-CNN neural network models and an Adaboost iterative model, wherein the Adaboost iterative model is used to weightedly combine multiple 1D-CNN neural network models to generate a strong learner for predicting the power load in the target area.
[0035] Preferably, the above-mentioned prediction system also includes a model training module for training the 1D-CNN-Adaboost power load prediction model.
[0036] Preferably, the model training module is trained by the following steps:
[0037] The preprocessed historical power load time series data is used as a data set and divided into a training set and a test set;
[0038] Assign the same initial weight to each sample of the power load time series data in the training set;
[0039] The samples assigned with initial weights are input into each 1D-CNN neural network model in the 1D-CNN-Adaboost power load forecasting model for training;
[0040] The Adaboost iterative model is used to calculate the error rate of each 1D-CNN neural network model on the training set;
[0041] According to the error rate of each 1D-CNN neural network model on the training set, the model weight corresponding to each 1D-CNN neural network model is calculated;
[0042] According to the error rate of each 1D-CNN neural network model on the training set, the initial weight of each sample is iteratively updated;
[0043] The samples after the initial weight update are input into each 1D-CNN neural network model again for training, and the model weight corresponding to the 1D-CNN neural network model is calculated according to the error rate of the trained 1D-CNN neural network model on the training set, until the number of iterations meets the set threshold;
[0044] When the number of iterations meets the set threshold requirement, multiple 1D-CNN neural network models are weightedly combined according to the model weights and used as strong learners for predicting the power load in the target area.
[0045] Preferably, the above-mentioned prediction system also includes a visualization display module for displaying the power load prediction results.
[0046] Finally, the present invention also discloses a storage medium, in which a computer program is stored. The computer program is executed by a processor to implement any one of the power load prediction methods of the present invention.
[0047] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method for predicting power load and related systems and media, which have the following beneficial effects:
[0048] The present invention uses the 1D-CNN-Adaboost power load forecasting model to predict power load data. By integrating and optimizing multiple 1D-CNN neural network models and performing weighted combination, the precision and accuracy of power load forecasting results are effectively improved, and the model is suitable for load forecasting of large-scale power data. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0050] Figure 1 This is a schematic diagram of the overall flow of the power load forecasting method provided by the present invention.
[0051] Figure 2 This is a schematic diagram of the 1D-CNN convolutional neural network structure provided by the present invention.
[0052] Figure 3 This is a schematic diagram of the prediction process of the 1D-CNN-Adaboost power load prediction model provided by the present invention.
[0053] Figure 4 This is a schematic diagram of the training process of the 1D-CNN-Adaboost power load prediction model provided by the present invention. DETAILED DESCRIPTION
[0054] The following will be combined with the 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, not 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 are within the scope of protection of the present invention.
[0055] The embodiment of the present invention discloses a method for predicting power load, such as Figure 1 As shown, the following steps are included:
[0056] Obtain historical power load time series data for the target area;
[0057] Preprocess the historical power load time series data;
[0058] Constructing a 1D-CNN-Adaboost power load prediction model, wherein the 1D-CNN-Adaboost power load prediction model includes multiple 1D-CNN neural network models and an Adaboost iterative model, wherein the Adaboost iterative model is used to weightedly combine the multiple 1D-CNN neural network models to generate a strong learner for predicting the power load of the target area;
[0059] The 1D-CNN-Adaboost power load forecasting model is used to predict the preprocessed historical power load time series data.
[0060] In the embodiment of the present invention, it is first necessary to obtain historical power load time series data and pre-process the historical power load time series data.
[0061] In a specific embodiment, a computer device may be used to obtain power load demand information of a target area every 15 minutes, and then the obtained power load demand information is statistically processed to obtain historical power load time series data.
[0062] The specific preprocessing of historical power load time series data mainly includes the following steps:
[0063] Normalization processing: normalize the historical power load time series data to remove the impact of the dimension;
[0064] Data outlier processing: Use the Hampel filtering algorithm to remove outliers in historical power load time series data.
[0065] The power load time series data is time series data. Therefore, in the present invention, the 1D-CNN convolutional neural network is first used to predict the power load time series data. The structural principle diagram of the 1D-CNN convolutional neural network is as follows: Figure 2As shown, first input the power load time series data into the input layer, then perform convolution operation in the convolution layer to generate n feature data sequences, then enter the ReLU layer to use the ReLU function for activation, and the activated power load time series sequence enters the pooling layer for pooling, performs dimensionality reduction on the data, and finally outputs the power load prediction value at the next moment through the fully connected layer. The input layer of 1D-CNN contains a one-dimensional vector and a one-dimensional convolution kernel, outputs a one-dimensional vector, and the length of the convolution kernel is much smaller than the length of the input one-dimensional vector, and the length is usually an odd number. The function of the convolution layer is to extract data features. This layer contains multiple convolution kernels, and the parameters of the convolution kernel are continuously corrected during CNN training. The ReLU layer does not change the size of the feature map, but activates each feature of the power load time series data using the ReLU function. The function of the pooling layer is to downsample the data and build the number of data features. In this application, for power load data, the average pooling method is used to pool the activated features.
[0066] In view of the low accuracy of the prediction of power load time series data by a single 1D-CNN convolutional neural network, a 1D-CNN-Adaboost power load prediction model is constructed in the present invention to predict the power load data, such as Figure 3 As shown, the 1D-CNN-Adaboost power load prediction model includes multiple 1D-CNN neural network models CNN and an Adaboost iterative model, wherein the Adaboost iterative model is used to weightedly combine multiple 1D-CNN neural network models to generate a strong learner for predicting the power load in the target area.
[0067] For the 1D-CNN-Adaboost power load forecasting model composed of a weighted combination of multiple 1D-CNN neural network models, the number of 1D-CNN neural network models also plays an important role in the forecasting efficiency and forecasting accuracy. After multiple verifications, when the number of 1D-CNN neural network models is between 5 and 8, it can not only ensure the efficiency of forecasting time, but also meet the requirements of forecasting accuracy.
[0068] The 1D-CNN-Adaboost power load forecasting model disclosed in the present invention is trained using a training set, such as Figure 4 As shown, the specific training steps are as follows:
[0069] The preprocessed historical power load time series data is used as a data set and divided into a training set and a test set. The ratio of the training set to the test set can be set to 7:3.
[0070] The same initial weight is assigned to each sample of the power load time series data in the training set. The samples assigned with the initial weight are input into each 1D-CNN neural network model in the 1D-CNN-Adaboost power load forecasting model for training.
[0071] The Adaboost iterative model is used to calculate the error rate of each 1D-CNN neural network model on the training set.
[0072] According to the error rate of each 1D-CNN neural network model on the training set, the model weight corresponding to each 1D-CNN neural network model is calculated.
[0073] According to the error rate of each 1D-CNN neural network model on the training set, the initial weight of each sample is iteratively updated.
[0074] The samples after the initial weight update are input into each 1D-CNN neural network model again for training, and the model weight corresponding to the 1D-CNN neural network model is calculated according to the error rate of the trained 1D-CNN neural network model on the training set until the number of iterations meets the set threshold.
[0075] When the number of iterations meets the set threshold requirement, multiple 1D-CNN neural network models are weightedly combined according to the model weights and used as strong learners for predicting the power load in the target area.
[0076] In a specific embodiment, the model weight corresponding to each 1D-CNN neural network model can be obtained using the following calculation formula:
[0077]
[0078] Among them, αi represents the model weight of the i-th 1D-CNN neural network model; εi represents the error rate of the i-th 1D-CNN neural network model. In actual application, the sum of the model weights of each 1D-CNN neural network model in the 1D-CNN-Adaboost power load forecasting model should be 1.
[0079] The purpose of updating sample weights is to increase the weights of samples whose error rates are greater than the set threshold, and to reduce the weights of samples whose error rates are less than the set threshold. Specifically, according to the error rate of each 1D-CNN neural network model on the training set, the initial weight of each sample is iteratively updated and updated using the following formula:
[0080]
[0081] in, represents the sample weight of the jth sample in the training set after update, represents the sample weight of the jth sample in the training set before updating, represents the model weight of the j-th 1D-CNN neural network model; yj is the real sample before training, Represents the predicted data after training, and t represents the number of update iterations.
[0082] In order to realize the power load prediction method disclosed in the present invention, an embodiment of the present invention further discloses a power load prediction system, including: a data acquisition module, used to obtain historical power load time series data of a target area; a data preprocessing module, used to preprocess the historical power load time series data; a data prediction module, used to predict the preprocessed historical power load time series data using a 1D-CNN-Adaboost power load prediction model to obtain a power load prediction result, the 1D-CNN-Adaboost power load prediction model includes multiple 1D-CNN neural network models and an Adaboost iterative model, wherein the Adaboost iterative model is used to weightedly combine multiple 1D-CNN neural network models to generate a strong learner for predicting the power load in the target area.
[0083] In addition, the above prediction system also includes a model training module for training the 1D-CNN-Adaboost power load prediction model.
[0084] In addition, in order to view the forecast data, the forecast system also includes a visual display module for displaying the power load forecast results.
[0085] Finally, an embodiment of the present invention further discloses a storage medium, in which a computer program is stored. The computer program is executed by a processor to implement any one of the power load prediction methods described in the present invention.
[0086] The technical solution of the present invention first adopts a 1D-CNN convolutional neural network model to predict the power load time series data, and at the same time constructs a strong learner with better performance by adjusting the sample weights and the weights of a single 1D-CNN convolutional neural network model.
[0087] In order to verify the superiority of the 1D-CNN-Adaboost power load prediction model of the present invention in power load prediction, this embodiment analyzes and compares the prediction performance of the single 1D-CNN and 1D-CNN-Adaboost prediction models under two different power load time series data sets. The prediction results are shown in Table 1:
[0088] Table 1 Comparison of prediction performance results of 1D-CNN alone and 1D-CNN-Adaboost
[0089]
[0090] As can be seen from Table 1, the comparison results of the indicators show that the mean absolute error MAE, mean absolute percentage error MAPE, and root mean square error RMSE values of the 1D-CNN-Adaboost prediction model in different training sets are better than those of the single 1D-CNN. This verifies that the 1D-CNN-Adaboost prediction model can effectively improve the accuracy and stability of power load prediction.
[0091] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0092] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for predicting power load, characterized in that: The following steps are involved: Obtain historical power load time series data for the target area; Preprocessing the historical power load time series data; Constructing a 1D-CNN-Adaboost power load prediction model, wherein the 1D-CNN-Adaboost power load prediction model includes multiple 1D-CNN neural network models and an Adaboost iterative model, wherein the Adaboost iterative model is used to weightedly combine the multiple 1D-CNN neural network models to generate a strong learner for predicting the power load of the target area; The 1D-CNN-Adaboost power load forecasting model is used to forecast the preprocessed historical power load time series data.
2. A method for predicting power load according to claim 1, characterized in that: Preprocessing the historical power load time series data specifically includes the following steps: Normalization processing: normalize the historical power load time series data to remove the impact of the dimension; Data outlier processing: Use the Hampel filtering algorithm to remove outliers in historical power load time series data.
3. The method for predicting power load according to claim 1, characterized in that: The 1D-CNN-Adaboost power load forecasting model is trained through the following steps: The preprocessed historical power load time series data is used as a data set and divided into a training set and a test set; Assign the same initial weight to each sample of the power load time series data in the training set; The samples assigned with initial weights are input into each 1D-CNN neural network model in the 1D-CNN-Adaboost power load forecasting model for training; The Adaboost iterative model is used to calculate the error rate of each 1D-CNN neural network model on the training set; According to the error rate of each 1D-CNN neural network model on the training set, the model weight corresponding to each 1D-CNN neural network model is calculated; According to the error rate of each 1D-CNN neural network model on the training set, the initial weight of each sample is iteratively updated; The samples after the initial weight update are input into each 1D-CNN neural network model again for training, and the model weight corresponding to the 1D-CNN neural network model is calculated according to the error rate of the trained 1D-CNN neural network model on the training set, until the number of iterations meets the set threshold; When the number of iterations meets the set threshold requirement, multiple 1D-CNN neural network models are weightedly combined according to the model weights and used as strong learners for predicting the power load in the target area.
4. A method for predicting power load according to claim 3, characterized in that: According to the error rate of each 1D-CNN neural network model on the training set, the model weight corresponding to each 1D-CNN neural network model is calculated using the following calculation formula: Among them, αi represents the model weight of the i-th 1D-CNN neural network model; εi represents the error rate of the i-th 1D-CNN neural network model.
5. A method for predicting power load according to claim 3, characterized in that: According to the error rate of each 1D-CNN neural network model on the training set, the initial weight of each sample is iteratively updated, specifically through the following formula: in, represents the sample weight of the jth sample in the training set after update, represents the sample weight of the jth sample in the training set before updating, represents the model weight of the j-th 1D-CNN neural network model; yj is the real sample before training, Represents the predicted data after training, and t represents the number of update iterations.
6. A power load prediction system, characterized in that: include: A data acquisition module is used to obtain the historical power load time series data of the target area; A data preprocessing module, used for preprocessing the historical power load time series data; The data prediction module is used to use the 1D-CNN-Adaboost power load prediction model to predict the preprocessed historical power load time series data to obtain the power load prediction result. The 1D-CNN-Adaboost power load prediction model includes multiple 1D-CNN neural network models and an Adaboost iterative model, wherein the Adaboost iterative model is used to weightedly combine multiple 1D-CNN neural network models to generate a strong learner for predicting the power load in the target area.
7. A power load prediction system according to claim 6, characterized in that: It also includes a model training module for training a 1D-CNN-Adaboost power load forecasting model.
8. The power load prediction system according to claim 7, characterized in that: The model training module is trained by the following steps: The preprocessed historical power load time series data is used as a data set and divided into a training set and a test set; Assign the same initial weight to each sample of the power load time series data in the training set; The samples assigned with initial weights are input into each 1D-CNN neural network model in the 1D-CNN-Adaboost power load forecasting model for training; The Adaboost iterative model is used to calculate the error rate of each 1D-CNN neural network model on the training set; According to the error rate of each 1D-CNN neural network model on the training set, the model weight corresponding to each 1D-CNN neural network model is calculated; According to the error rate of each 1D-CNN neural network model on the training set, the initial weight of each sample is iteratively updated; The samples after the initial weight update are input into each 1D-CNN neural network model again for training, and the model weight corresponding to the 1D-CNN neural network model is calculated according to the error rate of the trained 1D-CNN neural network model on the training set, until the number of iterations meets the set threshold; When the number of iterations meets the set threshold requirement, multiple 1D-CNN neural network models are weightedly combined according to the model weights and used as strong learners for predicting the power load in the target area.
9. The power load prediction system according to claim 6, characterized in that: It also includes a visualization display module for displaying the power load forecasting results.
10. A storage medium, characterized in that: The storage medium stores a computer program, and the computer program is executed by a processor to implement the power load prediction method described in any one of claims 1-5.