Power grid load prediction improving method and device based on AI intelligent algorithm

Abstract

The invention provides a power grid load prediction improvement method and device based on an AI intelligent algorithm. The method comprises the following steps: S1, constructing an AI intelligent algorithm library for load prediction; s2, performing industry clustering on the power grid load, and dividing the power grid load into different industry classifications; s3, according to the industry classification, obtaining historical load data of each industry, selecting an algorithm of the AI intelligent algorithm library to carry out load prediction, and obtaining sub-industry load prediction data; s4, constructing a temperature correction model of each industry classification; and S5, based on the sub-industry load prediction data and the temperature correction model of each industry, correcting the sub-industry load prediction data in real time through an artificial intelligence algorithm model in combination with predicted meteorological data. The problem that the power grid load prediction precision is reduced in a large-scale new energy grid-connected scene can be solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power grid load prediction, and in particular relates to a method and device for improving power grid load prediction based on an AI intelligent algorithm. Background Art

[0002] With the rapid development of new power systems based on new energy, photovoltaic and wind power have become an indispensable part of the power system, which can effectively solve the problem of power loss during voltage boosting and long-distance transportation, and significantly improve energy efficiency. However, with the rapid increase in the total installed capacity of photovoltaic and wind power, it also brings huge challenges to the safe operation of the power grid.

[0003] Before the development of new energy, the distribution network was a passive power grid, and the power grid load forecasting technology was relatively mature, with the accuracy of the entire network reaching 98%-99%. However, after the large-scale access of new energy to the power grid, the distribution network has changed from a passive power grid to an active power grid. The load forecasting results have been reduced by 2%-3% or even more due to the access of new energy. Traditional load forecasting methods can no longer meet the needs of modern power systems. Although a variety of load forecasting methods for new energy have been proposed in the prior art, there is still a problem of relatively low load forecasting accuracy. Summary of the invention

[0004] The present invention proposes a method and device for improving power grid load forecasting based on AI intelligent algorithm to solve the problem of reduced power grid load forecasting accuracy in large-scale renewable energy grid-connected scenarios.

[0005] To achieve the above object, the technical solution of the present invention is achieved as follows: A method for improving power grid load forecasting based on AI intelligent algorithm, comprising: S1. Build an AI intelligent algorithm library for load forecasting; S2, cluster the power grid load by industry and divide it into different industry classifications; S3. According to the industry classification, respectively obtain the historical load data of each industry, select the algorithm of the AI ​​intelligent algorithm library to perform load forecasting, and obtain the load forecast data by industry; S4. Construct temperature correction models for each industry classification; S5. Based on the load forecast data for each industry and the temperature correction model of each industry, the load forecast data for each industry is corrected in real time through an artificial intelligence algorithm model in combination with the predicted meteorological data.

[0006] Furthermore, the AI ​​intelligent algorithm library in step S1 includes traditional statistical algorithms, machine learning algorithms and deep neural network algorithms.

[0007] Furthermore, the method of selecting the algorithm of the AI ​​intelligent algorithm library in step S3 includes: For each industry classification, 1-N algorithms are selected for load forecasting. For the same initial conditions, different prediction results are obtained using different algorithms. A fixed-period accuracy comparison is set, and the optimal load forecasting algorithm for different industry classifications is selected based on the comparison results.

[0008] Furthermore, the method for constructing the temperature correction model for each industry classification in step S4 includes: S401. Obtain the temperature load capacity affected by temperature factors through historical load data classified by each industry and historical measured meteorological data; S402, calculating the historical temperature sensitive load by combining the temperature load capacity with the numerical weather forecast data; S403, obtaining the temperature coefficient of each industry classification according to the ratio of the historical temperature sensitive load to the total industry load; S404, the temperature coefficient of each industry classification is calculated in real time at a frequency of 15 minutes per day, and stored in units of 96 points per day, ultimately forming the temperature correction model data of each industry classification in a year, and is continuously corrected over time.

[0009] Furthermore, the method for correcting the industry-specific load forecast data in real time in step S5 includes: S501, using weather characteristic data of predicted meteorological data, historical load data of various industry classifications, and historical temperature correction model data as inputs of the artificial intelligence algorithm model; S502. Weighting the input weather characteristic data of the forecast meteorological data, the historical load data of each industry classification, and the historical temperature correction model data; reducing the impact of long time series on the overfitting of the artificial intelligence algorithm model; reducing the impact of differences between different variables on the artificial intelligence algorithm model; S503, performing nonlinear mapping output on the multivariate labels processed in step S502 through a multilayer perceptron to obtain a correction result of the load forecast data by industry; S504. Add up the revised results of the load forecast data for each industry to obtain the load forecast result of the power grid system.

[0010] On the other hand, the present invention also proposes a power grid load prediction and improvement device based on AI intelligent algorithm, comprising: Algorithm library module: builds AI intelligent algorithm library for load forecasting; Clustering module: clusters the power grid load by industry and divides it into different industry categories; Load forecasting module: according to the industry classification, the historical load data of each industry is obtained respectively, and the algorithm of the AI ​​intelligent algorithm library is selected to perform load forecasting to obtain load forecasting data by industry; Temperature correction model module: build temperature correction models for various industry classifications; Correction module: Based on the load forecast data for each industry, the temperature correction model of each industry, and the predicted meteorological data, the load forecast data for each industry is corrected in real time through an artificial intelligence algorithm model.

[0011] Furthermore, the AI ​​intelligent algorithm library in the algorithm library module includes traditional statistical algorithms, machine learning algorithms and deep neural network algorithms.

[0012] Furthermore, the load forecasting module includes: For each industry classification, 1-N algorithms are selected for load forecasting. For the same initial conditions, different prediction results are obtained using different algorithms. A fixed-period accuracy comparison is set, and the optimal load forecasting algorithm for different industry classifications is selected based on the comparison results.

[0013] Furthermore, the temperature correction model module includes: Temperature load capacity unit: obtains the temperature load capacity affected by temperature factors through historical load data classified by each industry and historical measured meteorological data; Temperature sensitive load unit: Calculate historical temperature sensitive loads by combining temperature load capacity with numerical weather forecast data; Temperature coefficient unit: the temperature coefficient of each industry classification is obtained by the ratio of the historical temperature sensitive load to the total industry load; Model data unit: The temperature coefficient of each industry classification is calculated in real time at a frequency of 15 minutes per day, and stored in units of 96 points per day, ultimately forming the temperature correction model data of each industry classification in a year, and is continuously corrected over time.

[0014] Furthermore, the correction module includes: Input embedding module: The weather characteristic data of the predicted meteorological data, the historical load data of each industry classification, and the historical temperature correction model data are used as the input of the artificial intelligence algorithm model; Encoder module: weights the input weather characteristic data of forecast meteorological data, historical load data of various industry classifications, and historical temperature correction model data; reduces the impact of long time series on overfitting of artificial intelligence algorithm models; reduces the impact of differences between different variables on artificial intelligence algorithm models; Projection layer module: The multivariate labels processed by the encoder module are nonlinearly mapped and output through a multi-layer perceptron to obtain the correction results of the load forecast data by industry; Result module: The revised results of the load forecast data for each industry are added together to obtain the load forecast results of the power grid system.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention proposes load forecasting by industry, realizes separate classification of new energy loads, and largely avoids the load forecast being affected by the output of new energy.

[0016] 2. The present invention realizes temperature correction by industry, distinguishes temperature-sensitive industries from temperature-insensitive industries by coefficients, helps to eliminate temperature-sensitive loads in load forecasting by industry, and makes load forecasting more targeted.

[0017] 3. The present invention adopts an artificial intelligence algorithm model to correct load forecasting: based on the temperature correction model, combined with predicted meteorological data, the artificial intelligence algorithm is used to correct the industry load forecasting data results in real time, thereby improving the load forecasting accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a schematic diagram of the method flow of Example 1 of the present invention; Figure 2 is a schematic diagram of a flow chart of load forecast correction according to Embodiment 1 of the present invention; Figure 3 It is a schematic diagram of the overall framework structure of the real-time modified industry-specific load forecasting model based on iTransformer in Example 1 of the present invention. DETAILED DESCRIPTION

[0019] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0020] In order to solve the problem of reduced load forecasting accuracy after large-scale access of new energy to the power grid, the present invention proposes a load forecasting improvement design method based on AI intelligent algorithm.

[0021] In order to make the purpose and features of the present invention more obvious and understandable, the present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0022] Embodiment 1: The load forecasting improvement method based on AI intelligent algorithm in this embodiment is as follows: Figure 1 As shown, including: 1. Construction of AI intelligent algorithm library.

[0023] To achieve load forecasting improvement based on AI intelligent algorithms, the first thing is to build an AI intelligent algorithm library, which includes dozens of algorithms including traditional statistical algorithms, machine learning algorithms and deep neural networks. Traditional algorithms include: univariate regression, binary regression, multivariate regression, polynomial regression, least squares method, etc.; machine learning algorithms include support vector machines, Xgboost, Lasso regression, etc.; deep neural networks include recurrent neural networks, feedforward neural networks, long-term and short-term neural networks, etc.

[0024] 2. Industry clustering.

[0025] Traditional power grid loads are generally divided into commercial loads, industrial loads, agricultural loads, civil loads and other loads. Based on the traditional classification, in order to solve the impact of new energy on the power grid load, this embodiment regards new energy loads as a new category of power grid loads. Therefore, the power grid loads are divided into the following industry classifications in this embodiment: Category 1: Commercial loads.

[0026] Category 2: Industrial load.

[0027] Category 3: Agricultural loads.

[0028] Category 4: Civilian loads.

[0029] Category 5: New energy load.

[0030] Category 6: Other loads.

[0031] 3. Load forecasting temperature correction model.

[0032] 3.1 Load forecast by industry.

[0033] According to the historical load data of different industry classifications, the algorithms in the AI ​​intelligent algorithm library are selected, and the industry load forecast is carried out according to the industry classification to obtain the historical industry load forecast data.

[0034] The algorithm selection rules in the AI ​​intelligent algorithm library are as follows: There are at least 1-N algorithms that can predict the load forecast of each industry in the industry classification. For the same initial conditions, different algorithms will have different results. A fixed period accuracy comparison is set to select the optimal load forecasting algorithm for different industries.

[0035] 3.2 Load forecasting temperature correction model.

[0036] Temperature can easily affect people's behavior and the way businesses operate. For example, when the temperature is high, users tend to turn on air conditioners to cool down, which increases the civilian load. When the temperature is low, users in the south also tend to turn on air conditioners, which increases the electricity load. Different industries have different degrees of sensitivity to temperature.

[0037] Different temperature coefficients are set according to the impact of temperature in each industry classification.

[0038] The temperature coefficient is calculated using the following steps: (1) First, the temperature load capacity affected by temperature factors is obtained through the historical load data of each industry classification and the historical measured meteorological data; the temperature load capacity refers to the sum of the load capacities that are sensitive to temperature, which can be generated according to different seasons.

[0039] For example, when the temperature is high in summer, the air conditioner needs to be turned on due to the temperature, and the air conditioner uses electricity to generate load. The temperature load capacity in summer is: ;in represents the temperature load capacity in summer, Indicates the total amount of load generated by all air conditioners.

[0040] In winter, when the temperature is low, heating is required due to the temperature, and heating electricity generates load. The temperature load capacity in winter is: ;in represents the temperature load capacity in winter, Indicates the total amount of all heating loads.

[0041] (2) Then the historical temperature sensitive load is calculated by combining the temperature load capacity with the numerical weather forecast data; ; All coefficients are determined by linear regression method.

[0042] To judge the calculated temperature sensitive load, the value must be greater than or equal to 0. Values ​​less than 0 are automatically set to 0.

[0043] (3) The temperature coefficient of each industry is obtained by comparing the ratio of historical temperature-sensitive load to the total industry load.

[0044] (4) The temperature coefficient is calculated in real time at a frequency of 15 minutes per day and stored in units of 96 points per day. The load forecast temperature correction model for each industry in the industry classification is finally formed in a year and is continuously corrected over time. The temperature correction model for each industry is expressed in the form of a matrix, and each industry is stored in units of years.

[0045] 4. AI load forecast correction.

[0046] like Figure 2 As shown, the correction method in this step includes correcting the industry-specific load forecast data in real time through an artificial intelligence algorithm based on the industry-specific load forecast data and the load forecast temperature correction model, combined with predicted meteorological data. The correction result takes into account the impact of temperature-sensitive loads and can correct the temperature correction model in real time according to changes in actual meteorological conditions. This will improve the accuracy of the load forecasting algorithm that only considers historical data.

[0047] In this embodiment, the artificial intelligence algorithm adopts iTransformer, and the real-time correction of the industry load forecasting model based on iTransformer is as follows: Figure 3 As shown in the figure, its overall framework structure can be divided into three modules: input embedding, encoder, and projection layer.

[0048] (1) Input embedding module, input the weather characteristics of the forecast meteorological data, the historical industry-specific load forecast data, and the historical load forecast temperature correction model (time series in matrix form) into the iTransformer model. The formula is as follows: h 0 n =Embedding(X); Where: X represents the weather characteristics of the input forecast meteorological data, the historical industry-specific load forecast data, and the historical load forecast temperature correction model (a time series in matrix form); h 0 n Embedded tags representing weather characteristics of forecasted meteorological data, historical industry-specific load forecast data, and historical load forecast temperature correction models (time series in matrix form); Embedding represents the embedding operation.

[0049] (2) The encoder module uses a multi-head attention mechanism to calculate the attention distribution value of independently embedded multiple variables (i.e., weather characteristics of predicted meteorological data, historical industry-specific load forecast data, and historical load forecast temperature correction model), and assigns high weights to important features, thereby achieving deep mining of the nonlinear features of weather data that have a greater impact on load forecasting. A feedforward neural network is used to reduce the impact of long time series on model overfitting; the normalization layer normalizes the data to reduce the impact of differences between different variables on the model.

[0050] T FFN (x)=max(0,xW1+b1)W2+b2; X′=T L (X+T M(Q,K,V) ); X out =T L (X′+T FFN (X)); Where: X′ represents the normalized weather feature input; x represents an element in X; max represents the activation function; T L Indicates normalization processing; T FFN Represents the linear processing of a feedforward neural network; X out represents the output matrix; W1 and W2 represent the weights of the model; b1 and b2 represent the bias of the model; T M(Q,K,V) It means that in the encoder module of iTransformer, the transformation or operation based on Query (Q), Key (K) and Value (V) is the core part of the attention mechanism.

[0051] (3) The projection layer module is composed of a multi-layer perceptron (MLP). This module performs nonlinear mapping output on the multi-variable labels that have been independently processed by the encoder module.

[0052] The method described in this embodiment can achieve separate classification of new energy loads, help eliminate temperature-sensitive loads in industry-specific load forecasting, and based on the temperature correction model and combined with predicted meteorological data, use artificial intelligence algorithms to correct industry-specific load forecasting data results in real time, thereby improving load forecasting accuracy.

[0053] Embodiment 2: This embodiment proposes a power grid load prediction and improvement device based on AI intelligent algorithm, including: Algorithm library module: builds AI intelligent algorithm library for load forecasting; Clustering module: clusters the power grid load by industry and divides it into different industry categories; Load forecasting module: according to the industry classification, the historical load data of each industry is obtained respectively, and the algorithm of the AI ​​intelligent algorithm library is selected to perform load forecasting to obtain load forecasting data by industry; Temperature correction model module: build temperature correction models for various industry classifications; Correction module: Based on the load forecast data for each industry, the temperature correction model of each industry, and the predicted meteorological data, the load forecast data for each industry is corrected in real time through an artificial intelligence algorithm model.

[0054] Among them, the AI ​​intelligent algorithm library in the algorithm library module includes traditional statistical algorithms, machine learning algorithms and deep neural network algorithms.

[0055] The load forecasting module includes: For each industry classification, 1-N algorithms are selected for load forecasting. For the same initial conditions, different prediction results are obtained using different algorithms. A fixed-period accuracy comparison is set, and the optimal load forecasting algorithm for different industry classifications is selected based on the comparison results.

[0056] The temperature correction model module includes: Temperature load capacity unit: obtains the temperature load capacity affected by temperature factors through historical load data classified by each industry and historical measured meteorological data; Temperature sensitive load unit: Calculate historical temperature sensitive loads by combining temperature load capacity with numerical weather forecast data; Temperature coefficient unit: the temperature coefficient of each industry classification is obtained by the ratio of the historical temperature sensitive load to the total industry load; Model data unit: The temperature coefficient of each industry classification is calculated in real time at a frequency of 15 minutes per day, and stored in units of 96 points per day, ultimately forming the temperature correction model data of each industry classification in a year, and is continuously corrected over time.

[0057] The correction modules include: Input embedding module: The weather characteristic data of the predicted meteorological data, the historical load data of each industry classification, and the historical temperature correction model data are used as the input of the artificial intelligence algorithm model; Encoder module: weights the input weather characteristic data of forecast meteorological data, historical load data of various industry classifications, and historical temperature correction model data; reduces the impact of long time series on overfitting of artificial intelligence algorithm models; reduces the impact of differences between different variables on artificial intelligence algorithm models; Projection layer module: The multivariate labels processed by the encoder module are nonlinearly mapped and output through a multi-layer perceptron to obtain the correction results of the load forecast data by industry; Result module: The revised results of the load forecast data for each industry are added together to obtain the load forecast results of the power grid system.

[0058] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for improving power grid load forecasting based on AI intelligent algorithm, characterized in that: include: S1. Build an AI intelligent algorithm library for load forecasting; S2, cluster the power grid load by industry and divide it into different industry classifications; S3. According to the industry classification, respectively obtain the historical load data of each industry, select the algorithm of the AI ​​intelligent algorithm library to perform load forecasting, and obtain the load forecast data by industry; S4. Construct temperature correction models for each industry classification; S5. Based on the load forecast data for each industry and the temperature correction model of each industry, the load forecast data for each industry is corrected in real time through an artificial intelligence algorithm model in combination with the predicted meteorological data.

2. The method for improving power grid load prediction based on AI intelligent algorithm according to claim 1 is characterized in that: The AI ​​intelligent algorithm library described in step S1 includes traditional statistical algorithms, machine learning algorithms and deep neural network algorithms.

3. The method for improving power grid load prediction based on AI intelligent algorithm according to claim 1 is characterized in that: The method for selecting the algorithm of the AI ​​intelligent algorithm library in step S3 includes: For each industry classification, 1-N algorithms are selected for load forecasting. For the same initial conditions, different prediction results are obtained using different algorithms. A fixed-period accuracy comparison is set, and the optimal load forecasting algorithm for different industry classifications is selected based on the comparison results.

4. The method for improving power grid load prediction based on AI intelligent algorithm according to claim 1 is characterized in that: The method for constructing the temperature correction model for each industry classification in step S4 includes: S401, obtaining the temperature load capacity affected by temperature factors through historical load data classified by each industry and historical measured meteorological data; S402, calculating the historical temperature sensitive load by combining the temperature load capacity with the numerical weather forecast data; S403, obtaining the temperature coefficient of each industry classification according to the ratio of the historical temperature sensitive load to the total industry load; S404, the temperature coefficient of each industry classification is calculated in real time at a frequency of 15 minutes per day, and stored in units of 96 points per day, ultimately forming the temperature correction model data of each industry classification in a year, and is continuously corrected over time.

5. The method for improving power grid load prediction based on AI intelligent algorithm according to claim 1 is characterized in that: The method for real-time correction of the industry-specific load forecast data in step S5 includes: S501, using weather characteristic data of predicted meteorological data, historical load data of various industry classifications, and historical temperature correction model data as inputs of the artificial intelligence algorithm model; S502. Weighting the input weather characteristic data of the forecast meteorological data, the historical load data of each industry classification, and the historical temperature correction model data; reducing the impact of long time series on the overfitting of the artificial intelligence algorithm model; reducing the impact of differences between different variables on the artificial intelligence algorithm model; S503, performing nonlinear mapping output on the multivariate labels processed in step S502 through a multilayer perceptron to obtain a correction result of the load forecast data by industry; S504. Add up the revised results of the load forecast data for each industry to obtain the load forecast result of the power grid system.

6. A power grid load prediction and improvement device based on AI intelligent algorithm, characterized in that: include: Algorithm library module: builds AI intelligent algorithm library for load forecasting; Clustering module: clusters the power grid load by industry and divides it into different industry categories; Load forecasting module: according to the industry classification, the historical load data of each industry is obtained respectively, and the algorithm of the AI ​​intelligent algorithm library is selected to perform load forecasting to obtain load forecasting data by industry; Temperature correction model module: build temperature correction models for various industry classifications; Correction module: Based on the load forecast data for each industry, the temperature correction model of each industry, and the predicted meteorological data, the load forecast data for each industry is corrected in real time through an artificial intelligence algorithm model.

7. The power grid load prediction and improvement device based on AI intelligent algorithm according to claim 6 is characterized in that: The AI ​​intelligent algorithm library in the algorithm library module includes traditional statistical algorithms, machine learning algorithms and deep neural network algorithms.

8. The power grid load prediction and improvement device based on AI intelligent algorithm according to claim 6 is characterized in that: The load forecasting module includes: For each industry classification, 1-N algorithms are selected for load forecasting. For the same initial conditions, different prediction results are obtained using different algorithms. A fixed-period accuracy comparison is set, and the optimal load forecasting algorithm for different industry classifications is selected based on the comparison results.

9. The power grid load prediction and improvement device based on AI intelligent algorithm according to claim 6 is characterized in that: The temperature correction model module includes: Temperature load capacity unit: obtains the temperature load capacity affected by temperature factors through historical load data classified by each industry and historical measured meteorological data; Temperature sensitive load unit: Calculate historical temperature sensitive loads by combining temperature load capacity with numerical weather forecast data; Temperature coefficient unit: the temperature coefficient of each industry classification is obtained by the ratio of the historical temperature sensitive load to the total industry load; Model data unit: The temperature coefficient of each industry classification is calculated in real time at a frequency of 15 minutes per day, and stored in units of 96 points per day, ultimately forming the temperature correction model data of each industry classification in a year, and is continuously corrected over time.

10. The power grid load prediction and improvement device based on AI intelligent algorithm according to claim 6 is characterized in that: The correction modules include: Input embedding module: The weather characteristic data of the predicted meteorological data, the historical load data of each industry classification, and the historical temperature correction model data are used as the input of the artificial intelligence algorithm model; Encoder module: weights the input weather characteristic data of forecast meteorological data, historical load data of various industry classifications, and historical temperature correction model data; reduces the impact of long time series on overfitting of artificial intelligence algorithm models; reduces the impact of differences between different variables on artificial intelligence algorithm models; Projection layer module: The multivariate labels processed by the encoder module are nonlinearly mapped and output through a multi-layer perceptron to obtain the correction results of the load forecast data by industry; Result module: The revised results of the load forecast data for each industry are added together to obtain the load forecast results of the power grid system.

Images