Sub-industry load prediction method and system based on type analysis and feature reconstruction

Abstract

The invention belongs to the technical field of sub-industry load prediction, and provides a sub-industry load prediction method and system based on type analysis and feature reconstruction, and the method comprises the steps: obtaining power load data and meteorological feature data of each industry, and carrying out the preprocessing; classifying electrical loads of different industries; calculating the maximum information coefficient between each meteorological characteristic variable and different types of electrical loads, performing characteristic selection according to the value of the maximum information coefficient, and reconstructing a meteorological characteristic matrix; taking historical load data in a set time period away from the to-be-predicted day as model input features, and combining the model input features into a feature matrix; training a deep learning model by taking the merged feature matrix as an input and the load value as an output; and selecting a corresponding feature matrix according to the power consumption type of a prediction target, and performing day-ahead load prediction by using the trained deep learning model. According to the invention, day-ahead load prediction of different industries can be realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of industry-specific load forecasting, and specifically relates to an industry-specific load forecasting method and system based on type analysis and feature reconstruction. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Load forecasting is of great significance in the power system. It can help power operators accurately predict future load demand, thereby optimizing power production, dispatching and distribution. Through effective load forecasting, it is possible to achieve reasonable allocation of power resources, improve the stability and reliability of the power grid, and reduce operating costs.

[0004] However, the electricity demand of different industries varies significantly, and it is impossible to accurately match the actual electricity consumption of each industry, which is prone to problems such as excessive or insufficient electricity, thereby affecting the production efficiency and cost control of enterprises. Therefore, it is particularly important to predict the electricity load of different industries. Through accurate prediction of the electricity load of different industries, reasonable allocation of power resources, optimization of power purchase plans, effective avoidance of risks brought by price fluctuations, improvement of power purchase efficiency, and accurate satisfaction of industry electricity demand can be ensured.

[0005] Artificial intelligence algorithms based on machine learning and deep learning are currently the main prediction methods. However, current artificial intelligence models usually directly use historical load sequences and their related meteorological data sequences as input features during model training, and usually assume that the same input features at different time points have the same degree of influence, which to a certain extent limits the model's ability to identify load fluctuations caused by meteorological factors. In addition, there are significant differences in the electricity consumption characteristics of different industries. The electricity demand of each industry is greatly affected by factors such as production processes, operating cycles, and equipment loads, resulting in different volatility and stability of their load characteristics.

[0006] Therefore, it is often difficult to achieve ideal results by uniformly using the same model parameters to predict the power load of various industries. Therefore, how to accurately predict the power load according to the power consumption characteristics of different industries has become an urgent problem to be solved in the existing technology. Summary of the invention

[0007] In order to solve the above problems, the present invention proposes an industry-specific load forecasting method and system based on type analysis and feature reconstruction. The present invention can realize day-ahead load forecasting in different industries and provide reference information for power dispatching and production planning, as well as power purchase transactions.

[0008] According to some embodiments, the present invention adopts the following technical solutions: A method for industry-specific load forecasting based on type analysis and feature reconstruction includes the following steps: Obtain the power load data and meteorological characteristic data of each industry and perform normalization processing; Classify the electricity load of different industries; Calculate the maximum information coefficient between each meteorological characteristic variable and different types of loads, perform feature selection based on the value of the maximum information coefficient, and reconstruct the meteorological characteristic matrix; The historical load data within a set time period from the forecast date is used as the model input feature and merged into the feature matrix; The deep learning model is trained with the merged feature matrix as input and the load value as output; The corresponding feature matrix is ​​selected according to the electricity consumption type of the forecast target, and the day-ahead load forecast result is obtained using the trained deep learning model.

[0009] As an optional implementation, the process of classifying the loads of different industries includes classifying the loads of different industries into the following categories according to the load characteristics: Double peak type: within a predetermined time, there are two peaks of electricity consumption exceeding the set value; Continuous power consumption type: The ratio of the average power load in the continuous period to the average load for the whole day exceeds the set value; Lunchtime electricity consumption type: The ratio of the average load during the lunchtime at the set time to the average load during the whole day exceeds the set threshold.

[0010] As an optional implementation, the process of calculating the maximum information coefficient between each meteorological characteristic variable and different types of loads includes: y is the target variable, x The maximum information coefficient between each meteorological characteristic variable and load data is calculated. , to reflect the input variables x For the target variable y degree of impact.

[0011] As an optional implementation, the process of reconstructing the meteorological characteristic matrix includes: Construct load sequences and meteorological characteristic sequences under different electricity consumption types; Selecting meteorological characteristic variables, selecting meteorological characteristic variables whose maximum information coefficient is greater than a predetermined value, and determining an initial meteorological characteristic matrix; The maximum information coefficient of different meteorological characteristic variables at each sampling point is used as the weight to measure the influence of each meteorological characteristic, and a weight matrix is ​​constructed; The corresponding position elements of the initial meteorological feature matrix and the weight matrix are multiplied to obtain the final meteorological feature matrix.

[0012] As a further step, in the process of selecting meteorological characteristic variables whose maximum information coefficient value is greater than a predetermined value, the maximum information coefficient value between different meteorological characteristics and load data is obtained by calculating the data of multiple sampling points within a set time period, and then the average maximum information coefficient value of each meteorological characteristic within the set time period is calculated, and different characteristics are sorted according to the average maximum information coefficient value, and characteristic variables with an average maximum information coefficient value greater than the set value are selected to form the initial meteorological characteristic matrix of the prediction model.

[0013] As an optional implementation, the deep learning model is a bidirectional long short-term memory network model. During the training process of the bidirectional long short-term memory network model, each time the bidirectional long short-term memory network model is trained, a forget gate, an input gate and an output gate are used to update the memory unit and the hidden layer.

[0014] As a further step, the calculation formula of the bidirectional long short-term memory network model is specifically as follows: ; ; ; in, and Respectively in t Load prediction values ​​of forward and backward long short-term memory networks at each moment; for The load forecast value at the moment; and Respectively represent the forward and backward calculation processes of the long short-term memory network; for t Input characteristics at the time; and They are t Forward and backward output weights at each moment; for t The memory unit of the moment; for t The load forecast value finally output by the bidirectional long short-term memory network model.

[0015] An industry-specific load forecasting system based on type analysis and feature reconstruction includes: The data acquisition module is configured to acquire load data and meteorological characteristic data of various industries and perform normalization processing; A type analysis module, configured to classify loads in different industries; A feature reconstruction module is configured to calculate the maximum information coefficient between each meteorological feature variable and different types of loads, perform feature selection according to the value of the maximum information coefficient, and reconstruct the meteorological feature matrix; The data merging module is configured to use the historical load data within a set time period from the forecasting date as the model input feature and merge it into the feature matrix; A prediction model training module is configured to train the deep learning model using the merged feature matrix as input and the load value as output; The day-ahead prediction module is configured to select a corresponding feature matrix according to the electricity consumption type of the prediction target, and use the trained deep learning model to obtain the day-ahead prediction result of the load.

[0016] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the steps in the above method are completed.

[0017] An electronic device comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps in the above method are completed.

[0018] Compared with the prior art, the present invention has the following beneficial effects: The present invention analyzes the electricity consumption characteristics of different industries, identifies different typical electricity consumption types, and formulates corresponding discrimination conditions to achieve the classification of industry loads. The present invention uses the maximum information coefficient method to quantify the impact of various meteorological characteristics on different types of loads, and reconstructs the input characteristics based on this to improve the adaptability of the model to different electricity consumption types.

[0019] To further improve the prediction accuracy, the present invention adopts a bidirectional long short-term memory network (BI-LSTM), making full use of its bidirectional information processing capabilities to effectively capture the nonlinear relationship between load and input features. Experimental results show that the BI-LSTM model after feature reconstruction shows excellent prediction performance under different types of electricity consumption, and can provide reference information for power dispatching and production planning, as well as electricity purchase transactions. In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0021] Figure 1 A schematic diagram of a first manifestation of a bimodal electricity consumption characteristic in an embodiment; Figure 2 A schematic diagram of a second manifestation of a bimodal electricity consumption characteristic in an embodiment; Figure 3 This is a schematic diagram of the continuous power consumption characteristics in one embodiment; Figure 4 A schematic diagram of the midday electricity consumption characteristics in one embodiment; Figure 5 A prediction flow chart of load prediction in an embodiment; Figure 6 It is a schematic diagram of the calculation result of the maximum information coefficient of the first bimodal electricity consumption type in one embodiment; Figure 7 A schematic diagram of the calculation result of the maximum information coefficient of the second bimodal electricity consumption type in an embodiment; Figure 8 A schematic diagram of a maximum information coefficient calculation result of a continuous power consumption type in an embodiment; Fig. 9 It is a schematic diagram of the calculation result of the maximum information coefficient of the midday electricity consumption type in one embodiment; Fig.10 It is a schematic diagram of load prediction results under the first type of double-peak electricity consumption type in an embodiment; Fig.11 It is a schematic diagram of load prediction results under the second type of double-peak electricity consumption type in an embodiment; Fig.12 It is a schematic diagram of load prediction results under the continuous power consumption type in an embodiment; Fig.13 It is a schematic diagram of load prediction results under the midday electricity consumption type in an embodiment. DETAILED DESCRIPTION

[0022] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0023] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0024] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0025] In the absence of conflict, the embodiments in this application and the features in the embodiments may be combined with each other.

[0026] Embodiment 1 Different industries have unique electricity usage characteristics due to their different operating models and demand characteristics. These characteristics not only affect the energy costs of enterprises, but also have an important impact on the power market and grid management. In order to more effectively manage and predict power demand, it is particularly important to classify different types of electricity consumption according to their characteristics and formulate discrimination criteria.

[0027] There are differences in electricity consumption levels in different industries, and different meteorological elements have different units. In order to reduce the impact of the above dimensions and units on load forecasting, the load data of each industry and the data of each meteorological element are first normalized, as shown in formula (1): (1) In the formula, is the value after normalization of each load or meteorological data. For its original data, and They represent the maximum and minimum values ​​of the original data of the corresponding load or meteorological series respectively.

[0028] Due to the wide variety of different industries, they need to be classified to reduce the workload of modeling. According to different load characteristics, they can be divided into the following typical types of electricity consumption: (1) Bimodal type This type of electricity consumption pattern shows two obvious peaks. The first type of peaks occurs between 9:00 and 11:00 a.m. and between 14:00 and 16:00 p.m. This double peak phenomenon is usually consistent with the working and resting patterns of weekdays, such as Figure 1 As shown in the figure, each sample is a power user of this type. The two peaks of electricity consumption in the second form occur from 10 to 14 o'clock and from 18 to 22 o'clock. This electricity consumption characteristic is usually related to the catering and accommodation industry, such as Figure 2 As shown, each sample in the figure is an electricity user of this type.

[0029] The bimodal type can be identified by comparing the two peak loads with the daily average load. The identification criterion is defined by formula (2): (2) In the formula, and are the average loads during the two peak hours, is the average load for the day, is the ratio threshold set to identify two obvious peaks.

[0030] (2) Continuous power consumption The main feature of this type of electricity consumption pattern is that the load usage continues to be at a high level from the afternoon to the evening, such as Figure 3As shown in the figure, each sample in the figure is a power user of this type. This feature is usually related to continuous business activities or production operations, reflecting the high power demand of some industries during this period.

[0031] The continuous power consumption type can be identified by the ratio of the average load of the continuous period to the average load of the whole day. The identification criterion is defined by formula (3): (3) In the formula, is the average load for a continuous period starting in the afternoon and ending in the evening. is a set ratio threshold, which should be set high enough to ensure that true sustained peak power consumption characteristics are distinguished.

[0032] (3) Lunchtime electricity consumption This type of electricity consumption pattern is characterized by a sharp increase in electricity load during the lunch period, that is, between 11:00 and 14:00, such as Figure 4 As shown in the figure, each sample is a power user of this type. The peak power consumption during this period is mainly due to the concentrated activation of power-consuming equipment, such as kitchen equipment and air conditioning systems.

[0033] The midday electricity consumption type can be determined by the ratio of the midday average load to the daily average load. The determination criteria are defined by formula (4): (4) In the formula, It is the average load between 11:00 and 14:00 at noon. is the set ratio threshold.

[0034] Due to the different operation modes of different industries, there are significant differences in meteorological influencing factors under different electricity consumption modes. Therefore, this embodiment adopts the maximum information coefficient method to reconstruct meteorological characteristics, by calculating the maximum information coefficient of electricity load under different meteorological elements and various electricity consumption characteristics, quantifying the degree of influence of meteorological element characteristics on various electricity loads, and sorting meteorological characteristics according to the degree of influence, and selecting a group of characteristic variables with the most significant impact on electricity load as the meteorological input characteristic matrix.

[0035] The maximum information coefficient (MIC) is a correlation measurement index that can widely mine the multidimensional dependency relationship between variables in a large data set. It is defined as follows: is a finite two-dimensional data set, x and y is the current two-dimensional space X, Y The interval divided in the direction; G is the interval formed by dimensional grid; is a set formed by different grid division methods, then define: (5) In the formula, Representation dataset D In the grid G Distribution on ; express The mutual information of . Remember the data sets under different partition intervals D The maximum normalized mutual information matrix is , whose elements are: (6) In the formula, represents the smaller value of the two. Then the maximum information coefficient can be defined as: (7) In the formula, n is the number of samples in the dataset; is the upper limit of the number of grid divisions, and its general value range is , .along with As MIC increases, the accuracy of MIC's measurement of correlation increases, but at the same time, the computational complexity also increases. , this value can take into account both accuracy and computational complexity, thus achieving better results.

[0036] When MIC is used as the correlation metric for selecting meteorological features for load day forecasting, the load value y is the target variable, x The meteorological input variables of the prediction model to be screened are calculated by calculating the maximum information coefficient between each variable and the load data. , to reflect the input variables x For the target variable y , thereby realizing input feature selection.

[0037] MIC overcomes the limitation of the Pearson correlation coefficient that it cannot reveal nonlinear relationships. When processing load data under the influence of multidimensional meteorological characteristics, it can more effectively extract valuable input features. After completing the feature selection, the MIC value is used as the weight of each meteorological input variable, and the meteorological input variables are reconstructed according to the weight. The reconstructed meteorological input matrix contains each meteorological characteristic variable and its corresponding weight, thereby reflecting the degree of influence of different variables on the load in the prediction model. The specific steps of feature selection and reconstruction in this embodiment are: (1) Construct load sequences under different electricity consumption types and meteorological characteristic sequences .

[0038] For the kth type of electricity consumption, extract the kth i The load values ​​of the sampling points are sorted in chronological order to obtain the i The sampling point level load sequence at each moment: (8) In the formula, Represents the k-th type of electricity consumption characteristics on the nth day, ( ) sampling points.

[0039] The meteorological characteristic data that affect the load are reconstructed in the same way. m Extract all date sampling points from feature sequences The value of can be obtained: (9) In the formula, Representative m The characteristics of the n Tiandi Normalized data of sampling points.

[0040] (2) Select meteorological features and determine the initial meteorological feature data set.

[0041] For k The electricity consumption characteristics of the is the target sequence, is the feature sequence to be selected, and and By calculating the data of 24 sampling points in a day, the MIC values ​​between different meteorological characteristics and load data are obtained. Then the average MIC value of each meteorological characteristic within 24 hours is calculated, and different characteristics are sorted according to the average value. The characteristic variables with an average MIC greater than 0.3 are selected to form the initial meteorological input feature set of the prediction model, and the initial meteorological input feature set corresponding to the load at 24 points per day is obtained. .

[0042] (10) In the formula, Representative k Under the electricity consumption characteristics of the q The meteorological characteristics of the day i The normalized value of the sampling points.

[0043] (3) Reconstruct input features based on MIC values.

[0044] The MIC values ​​of different variables at each sampling point are used as weights to measure the impact of each meteorological feature and construct the input weight matrix. qThe input characteristics and loads are i The MIC value of each sampling point is used as the weight value of the input feature at that moment to form the weight matrix of the input feature. M . is defined as: (11) Thus, the weights of all variables form a weight matrix under the kth type of electricity consumption: .

[0045] (12) (4) Determine the input feature data set.

[0046] Take the feature matrix from step 3 and step 4 With the feature weight matrix , the corresponding positions of the two matrices are multiplied to obtain the final meteorological input matrix ,Right now (13) Where * represents the Hadamard product of two matrices.

[0047] Since the electricity consumption of the same industry on nearby dates is similar, the load data between adjacent dates have a large correlation. Therefore, when making load forecasts, the historical load data of the same industry for the three days closest to the predicted day (need to be the same working day or the same non-working day) are also used as model input features and merged into the feature matrix.

[0048] After the input features are determined, the prediction model is trained. In this embodiment, a bidirectional long short-term memory network (BILSTM) is selected as the prediction model.

[0049] LSTM is a commonly used time series prediction model that can effectively capture the nonlinear change relationship between the target variable and all input features from historical data. In view of the load characteristics of different industries, the load value at the current moment is associated with the information at the historical moment and the information at the future moment, so it is suitable for the LSTM model. Through its unique "gating unit" structure, LSTM can flexibly add or delete information, thereby realizing training with memory function. This feature makes LSTM more accurate in learning long-term dependencies, and can also more effectively capture nonlinear dependencies. In the LSTM training process, each training not only depends on the hidden layer of the previous training, but also on the hidden layer of the previous training. , and also introduced the memory unit LSTM can effectively update memory cells through three gating units: forget gate, input gate and output gate. And the hidden layer , thereby realizing the dynamic transmission and storage of information.

[0050] On the basis of the LSTM model, this embodiment further selects BILSTM, which can simultaneously consider bidirectional time information, as the load day-ahead forecasting model. The input data of BI-LSTM is processed by the forward and backward LSTM networks respectively, and the final output result contains the splicing of the two-directional information, thereby improving the prediction accuracy. BI-LSTM is a special recurrent neural network structure that can capture the forward and reverse time dependencies in the sequence at the same time by combining the forward LSTM and backward LSTM networks. BI-LSTM enhances the modeling ability of sequence data by processing past (forward) and future (backward) information in parallel, thereby showing better results in sequence modeling tasks.

[0051] The calculation formula of BILSTM is as follows.

[0052] (14) (15) (16) In the formula, and Respectively in t Load prediction values ​​of forward and backward LSTM units at the moment; for The load forecast value at the moment; and Respectively represent the LSTM forward and backward calculation processes; for t Input characteristics at the time; and They are t Forward and backward output weights at each moment; for t Memory unit of moment; for t The load forecast value finally output by the BILSTM unit at time .

[0053] according to Figure 5 The process shown realizes the day-ahead load forecasting, and the normalized mean absolute error NMAE and normalized root mean square error NRMSE are used to evaluate the forecasting results, whose expressions are shown in Equations (17) and (18).

[0054] (17) (18) In the formula, and fori The actual load and predicted load normalized at each moment; N is the total number of samples.

[0055] Embodiment 2 This embodiment conducts a load day-ahead forecasting example verification based on the industry load data and corresponding meteorological data of a certain city from 2020 to 2022. The load data resolution is 1 hour, and the meteorological data resolution is the same as the load data, including load influencing characteristics such as temperature, humidity, wind speed, and irradiation.

[0056] First, a candidate feature set is constructed based on the existing meteorological data, and feature selection is performed, and then reconstructed into new input features according to the weights. There are 14 different variables in the candidate feature set. The MIC values ​​between different meteorological characteristics and load data under various types of electricity consumption are calculated. Figure 6-Figure 9 shown.

[0057] by The meteorological input characteristics of the threshold screening prediction model are used. When the MIC is less than 0.3, it is considered that the impact of the meteorological characteristics on the load is small and can be ignored. The optimization results of meteorological characteristics under different electricity consumption types are shown in Table 1.

[0058] Table 1 Feature selection results

[0059] It can be seen that the MIC values ​​of various meteorological impact characteristics under different electricity consumption types are different, and the selected high-impact characteristics are also different, reflecting the differences in electricity consumption behaviors in different industries.

[0060] Since the power load characteristics of the nearby dates are similar, the historical load data of the same type (working days or non-working days) for three days closest to the forecast date are also used as input features. The MIC value is used as the weight, and the input features are reconstructed as the input of the final model. For different types of power consumption, the selected meteorological features are combined with historical load information to obtain the input feature matrix. After confirming the input features, the Bi-LSTM network is used to conduct experimental verification of the day-ahead forecast of load data under various types of electricity consumption. For each type of electricity consumption, a training set, a test set, and a validation set are constructed. The Bi-LSTM model is recorded as M1, and the LSTM is used as the control model M2; at the same time, the Bi-LSTM model without feature reconstruction is used as M3. The day-ahead forecast results and errors of load under various types of electricity consumption are compared. Figure 10-13 As shown in Table 2.

[0061] Table 2 Comparison of prediction errors

[0062] It can be seen that under different types of electricity consumption, the proposed model can achieve good prediction results and smaller prediction errors. Compared with M3, which does not use the feature optimization method proposed in this embodiment, NMAE and NRMSE are reduced by 3.009%-6.880% and 3.392%-8.624%, respectively, which proves the effectiveness of the MIC feature reconstruction method proposed in this embodiment.

[0063] In the bimodal prediction, the model proposed in this embodiment can accurately capture the peak and valley values ​​during the morning and evening peak hours. This is because LSTM can effectively learn and memorize long-term time series dependencies, so that the model can not only predict the magnitude of the load, but also accurately predict the time points of load increase and decrease; in the continuous power consumption prediction, the model proposed in this embodiment shows good recognition ability for continuous high load states, can maintain accurate tracking of load trends, and has no significant prediction deviation, which is of great significance for energy management and cost control of continuously operating commercial facilities or industrial production; in the midday power consumption prediction, the model proposed in this embodiment can accurately capture the sharp increase and subsequent decrease in load during the noon period, showing a rapid response capability to the midday power consumption peak.

[0064] Embodiment 3 Industry-specific load forecasting system based on type analysis and feature reconstruction, including: The data acquisition module is configured to acquire the power load data and meteorological characteristic data of various industries and perform normalization processing; A type analysis module is configured to classify the electricity load of different industries; A feature reconstruction module is configured to calculate the maximum information coefficient between each meteorological feature variable and different types of loads, perform feature selection according to the value of the maximum information coefficient, and reconstruct the meteorological feature matrix; The data merging module is configured to use the historical load data within a set time period from the forecasting date as the model input feature and merge it into the feature matrix; A prediction model training module is configured to train the deep learning model using the merged feature matrix as input and the load value as output; The day-ahead prediction module is configured to select a corresponding feature matrix according to the electricity consumption type of the prediction target, and use the trained deep learning model to obtain the day-ahead prediction result of the electricity load.

[0065] It should be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of an all-hardware embodiment, an all-software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of one or more computer-usable storage media (including but not limited to disk storage, CD - ROM , optical storage, etc.).

[0066] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0067] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0068] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0069] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principle of the present invention without creative labor shall be included in the protection scope of the present invention.

Claims

1. A method for load forecasting by industry based on type analysis and feature reconstruction, characterized in that: The following steps are involved: Obtain the power load data and meteorological characteristic data of each industry and perform normalization processing; Classify the electricity load of different industries; Calculate the maximum information coefficient between each meteorological characteristic variable and different types of power load, perform feature selection based on the value of the maximum information coefficient, and reconstruct the meteorological characteristic matrix; The historical load data within a set time period from the forecast date is used as the model input feature and merged into the feature matrix; The deep learning model is trained with the merged feature matrix as input and the load value as output; The corresponding feature matrix is ​​selected according to the electricity consumption type of the forecast target, and the day-ahead forecast result of the electricity load is obtained using the trained deep learning model.

2. The method for industry-specific load forecasting based on type analysis and feature reconstruction as claimed in claim 1, characterized in that: The process of classifying the power load of different industries includes classifying the power load of different industries into the following categories according to the power consumption characteristics: Double peak type: within a predetermined time, there are two peaks of electricity consumption exceeding the set value; Continuous power consumption type: The ratio of the average power load in the continuous period to the average load for the whole day exceeds the set value; Lunchtime electricity consumption type: The ratio of the average load during the lunchtime at the set time to the average load during the whole day exceeds the set threshold.

3. The method for industry-specific load forecasting based on type analysis and feature reconstruction as claimed in claim 1, characterized in that: The process of calculating the maximum information coefficient between each meteorological characteristic variable and different types of loads includes: y is the target variable, x The maximum information coefficient between each meteorological characteristic variable and load data is calculated. , to reflect the input variables x For the target variable y degree of impact.

4. The method for industry-specific load forecasting based on type analysis and feature reconstruction as claimed in claim 1, characterized in that: The process of the meteorological characteristics matrix includes: Construct load sequences and meteorological characteristic sequences under different electricity consumption types; Selecting meteorological characteristic variables, selecting meteorological characteristic variables whose maximum information coefficient is greater than a predetermined value, and determining an initial meteorological characteristic matrix; The maximum information coefficient of different meteorological characteristic variables at each sampling point is used as the weight to measure the influence of each meteorological characteristic, and a weight matrix is ​​constructed; The corresponding position elements of the initial meteorological feature matrix and the weight matrix are multiplied to obtain the final meteorological feature matrix.

5. The method for industry-specific load forecasting based on type analysis and feature reconstruction as claimed in claim 4, characterized in that: In the process of selecting meteorological characteristic variables whose maximum information coefficient value is greater than a predetermined value, the maximum information coefficient value between different meteorological characteristics and load data is obtained by calculating the data of multiple sampling points within the set time period, and then the average maximum information coefficient value of each meteorological characteristic within the set time period is calculated, and different characteristics are sorted according to the average maximum information coefficient value, and characteristic variables whose average maximum information coefficient value is greater than the set value are selected to form the initial meteorological characteristic matrix of the prediction model.

6. The method for industry-specific load forecasting based on type analysis and feature reconstruction as claimed in claim 1, characterized in that: The deep learning model is a bidirectional long short-term memory network model. During the training process of the bidirectional long short-term memory network model, each time the bidirectional long short-term memory network model is trained, a forget gate, an input gate and an output gate are used to update the memory unit and the hidden layer.

7. The method for industry-specific load forecasting based on type analysis and feature reconstruction as claimed in claim 6, characterized in that: The calculation formula of the bidirectional long short-term memory network model is specifically: ； ； ； in, and Respectively in t Load prediction values ​​of forward and backward long short-term memory networks at each moment; for The load forecast value at the moment; and Respectively represent the forward and backward calculation processes of the long short-term memory network; for t Input characteristics at the time; and They are t Forward and backward output weights at each moment; for t The memory unit of the moment; for t The load forecast value finally output by the bidirectional long short-term memory network model.

8. An industry-specific load forecasting system based on type analysis and feature reconstruction, characterized in that: include: The data acquisition module is configured to acquire the power load data and meteorological characteristic data of various industries and perform normalization processing; A type analysis module is configured to classify the electricity load of different industries; A feature reconstruction module is configured to calculate the maximum information coefficient between each meteorological feature variable and different types of power loads, perform feature selection according to the value of the maximum information coefficient, and reconstruct the meteorological feature matrix; The data merging module is configured to use the historical load data within a set time period from the forecasting date as the model input feature and merge it into the feature matrix; A prediction model training module is configured to train the deep learning model using the merged feature matrix as input and the load value as output; The day-ahead prediction module is configured to select a corresponding feature matrix according to the electricity consumption type of the prediction target, and use the trained deep learning model to obtain the day-ahead prediction result of the load.

9. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps of the method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: The method comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the steps in the method according to any one of claims 1 to 7 are completed.

Images