Power load prediction model establishment method based on convolutional bidirectional long-short-term memory network and attention mechanism

Abstract

The invention discloses a power load prediction model establishment method based on a convolutional bidirectional long-short-term memory network and an attention mechanism, and the method comprises the steps: carrying out the full feature extraction of original complex power load data through a one-dimensional convolutional network; the factors influencing the load and the characteristics of trend, periodicity, randomness and the like of load change can be more fully mastered by considering the multivariate characteristics; compared with the traditional LSTM, the bidirectional structure of the BiLSTM can more comprehensively capture the bidirectional dependency relationship in the load sequence data through forward propagation and backward propagation, and the understanding of the model on the data is improved. Besides, an attention mechanism widely applied in the field of natural language processing is applied to research on multi-element short-term power load prediction, key information in a load sequence can be focused by introducing the attention mechanism, and the prediction precision is further improved; experiments are carried out on a real multivariate load data set, and the precision of multivariate short-term load prediction can be effectively improved by adding the one-dimensional convolutional network and the attention mechanism.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent power engineering, and specifically relates to a method for establishing a power load forecasting model based on a convolutional bidirectional long short-term memory network and an attention mechanism. Background Art

[0002] In the power system, short-term load forecasting is a crucial task, which is of great significance to the stable operation of the power system, optimal resource allocation and improvement of economic benefits. Traditional short-term load forecasting methods mainly include time series analysis and machine learning methods.

[0003] Time series analysis methods, such as the autoregressive integrated moving average model (ARIMA) and seasonal decomposed time series (STL), mainly rely on historical data and seasonal patterns for prediction. Such methods perform well when dealing with stable and highly cyclical loads, but their prediction accuracy is often limited when faced with complex and changeable power load data; machine learning methods, such as support vector machines (SVM) and random forests (RF), improve prediction accuracy compared to time series analysis methods by constructing complex algorithm models to handle more variables and nonlinear relationships. However, with the continuous expansion of the scale of power systems and the rapid growth of data volume, machine learning models face problems such as high computational complexity and long training time when dealing with large-scale data sets. In addition, machine learning models are also insufficient in capturing long-term dependencies in time series data, which limits their further application in short-term load forecasting.

[0004] The existing power load data is complex and diverse, and the existing technologies mostly perform learning and prediction on single load data, which makes it difficult to fully explore the characteristics of power load data; there are still defects to a certain extent. Summary of the invention

[0005] In order to solve the above technical problems, the present invention designs a method for establishing a power load forecasting model based on a convolutional bidirectional long short-term memory network and an attention mechanism, comprising the following steps:

[0006] S1: Data preprocessing: including processing outliers and missing data, then standardizing the data, and then constructing time series data through sliding windows with 24-hour slices;

[0007] S2: Model construction: It includes the following steps:

[0008] S2.1: Use one-dimensional convolutional network (1D-CNN) to extract temperature and humidity multivariate features from the original power load data by sliding in the time dimension;

[0009] S2.2: The new feature vector after convolution is used as the input of the maximum pooling layer for pooling operation, which reduces the dimension of the time series data while retaining the most important features, shortens the training time, prevents overfitting, and improves the generalization ability of the model;

[0010] S2.3: Flatten the pooled feature vector as the input of the bidirectional long short-term memory network;

[0011] S2.4: Capture historical and future information of data through a bidirectional long short-term memory network (BiLSTM) through a gating mechanism and a bidirectional network structure;

[0012] S2.5: Focus on the important parts of the load sequence data through the weight allocation of the attention mechanism, improve the model's ability to recognize key information, and thus improve the accuracy of short-term power load forecasting.

[0013] Furthermore, in S1, the outliers are processed by horizontal processing method to deal with data anomalies caused by sudden random events; missing data are processed by interpolation method, and z-score standardization is used to process the data during standardization;

[0014] The beneficial effects of the present invention are:

[0015] The present invention uses a one-dimensional convolutional network to fully extract features from the original complex power load data, rather than using only a single load data for training and prediction. Considering multiple features can more fully grasp the factors affecting the load and the characteristics of the trend, periodicity and randomness of load changes; compared with traditional LSTM, the bidirectional structure of BiLSTM can more comprehensively capture the bidirectional dependencies in the load sequence data through forward propagation and backward propagation, improving the model's understanding of the data.

[0016] In addition, the present invention applies the attention mechanism widely used in the field of natural language processing to the research of multivariate short-term power load forecasting. The introduction of the attention mechanism can focus on the key information in the load sequence and further improve the prediction accuracy. Experiments are conducted on real multivariate load data sets. The addition of one-dimensional convolutional networks and attention mechanisms can effectively improve the accuracy of multivariate short-term load forecasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for describing the embodiments are briefly introduced below.

[0018] Figure 1 It is the overall structure diagram of the prediction model in the present invention;

[0019] Figure 2 It is a comparison chart of 24-hour load forecast of the existing model in the present invention. DETAILED DESCRIPTION

[0020] The present invention discloses a method for establishing a power load forecasting model based on a convolutional bidirectional long short-term memory network and an attention mechanism, comprising:

[0021] S1: Data preprocessing: including processing of outliers and missing data, followed by data standardization, and then constructing time series data using 24-hour slices through sliding windows; outliers are processed using the horizontal processing method to address data anomalies caused by sudden random events; missing data are processed using the interpolation method, and z-score standardization is used to process data during standardization;

[0022] S2: Model construction: It includes the following steps:

[0023] S2.1: Use one-dimensional convolutional network (1D-CNN) to extract temperature and humidity multivariate features from the original power load data in the time dimension; let Xt, t = 1, 2, ..., n be the time series, X t ∈R k is the k-dimensional vector at time t. During the convolution operation, a convolution kernel with a height of h and the same width as the time series is used to perform the convolution operation to obtain the new feature c t The one-dimensional convolution operation is calculated as follows: t =f(W·X t:t+h-1 + b);

[0024] S2.2: The new feature vector after convolution is used as the input of the maximum pooling layer for pooling operation, which reduces the dimension of the time series data while retaining the most important features, shortens the training time, prevents overfitting, and improves the generalization ability of the model;

[0025] S2.3: Flatten the pooled feature vector as the input of the bidirectional long short-term memory network; S2.4: Capture the historical and future information of the data through the bidirectional long short-term memory network (BiLSTM) through the gating mechanism and bidirectional network structure; The calculation formula for each node is as follows:

[0026]

[0027] S2.5: Focus on the important parts of the load sequence data through the weight allocation of the attention mechanism, improve the model's ability to identify key information, and thus improve the accuracy of short-term power load forecasting; for the output hidden state h of the BiLSTM network t , importance score e t , calculated by the following formula: t =h tW. Then the importance score is normalized using the softmax function:

[0028]

[0029] The final attention vector v is obtained by weighted sum:

[0030]

[0031] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only the specific implementation methods described.

Claims

1. A method for establishing a power load forecasting model based on a convolutional bidirectional long short-term memory network and an attention mechanism, characterized in that: The following steps are involved: S1: Data preprocessing: including processing outliers and missing data, then standardizing the data, and then constructing time series data through sliding windows with 24-hour slices; S2: Model construction: It includes the following steps: S2.1: Use one-dimensional convolutional network (1D-CNN) to extract temperature and humidity multivariate features from the original power load data by sliding in the time dimension; S2.2: The new feature vector after convolution is used as the input of the maximum pooling layer for pooling operation, which reduces the dimension of the time series data while retaining the most important features, shortens the training time, prevents overfitting, and improves the generalization ability of the model; S2.3: Flatten the pooled feature vector as the input of the bidirectional long short-term memory network; S2.4: Capture historical and future information of data through a bidirectional long short-term memory network (BiLSTM) through a gating mechanism and a bidirectional network structure; S2.5: Focus on the important parts of the load sequence data through the weight allocation of the attention mechanism, improve the model's ability to recognize key information, and thus improve the accuracy of short-term power load forecasting.

2. The method for establishing a power load forecasting model based on a convolutional bidirectional long short-term memory network and an attention mechanism according to claim 1 is characterized in that: In S1, the horizontal processing method is used for outlier processing, which is aimed at data anomalies caused by sudden random events; the missing data are processed by interpolation method, and the z-score standardization is used for data processing during standardization.

Images