A medium- and long-term load forecasting method taking into account industry correlation

Through iterative interpolation and correlation analysis methods, combined with LSTM neural network, the problem of unconsidered correlation between industries is solved, the accuracy of medium- and long-term power system load prediction is improved, and efficient data recovery and prediction effects are achieved.

CN113962456BActive Publication Date: 2024-11-22JIANGSU FRONTIER ELECTRIC TECH +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111215938.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-10-19
Publication Date
2024-11-22
Estimated Expiration
2041-10-19

AI Technical Summary

Technical Problem

The existing medium- and long-term power system load prediction methods fail to effectively consider the correlation between industries, resulting in low load prediction accuracy.

Method used

The industry's monthly load data repair method based on iterative interpolation and min-max standardization processing were used, and industry correlation was analyzed in combination with Shapiro-Wilk test and Pearson correlation coefficient, and long-term memory neural network (LSTM) was used for medium- and long-term load prediction.

Benefits of technology

The industry's monthly load missing data has been achieved and the improvement of medium- and long-term load prediction accuracy has been improved, with the error being controlled within 4%.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113962456B_ABST
    Figure CN113962456B_ABST
Patent Text Reader

Abstract

The present invention discloses a medium- and long-term load forecasting method taking into account industry correlation. The method of the present invention is as follows: first, in view of the problems of missing industry load data and large differences in data magnitude, the industry load data is preprocessed, and an industry monthly load data repair method based on iterative interpolation is proposed to restore the missing monthly load data of typical industries, and the load data is normalized using the min‑max standardization method. On this basis, the industry correlation is analyzed and quantified based on the Shapiro‑Wilk test and the Pearson correlation coefficient, and strongly correlated industries are screened according to the correlation index. Finally, based on the industry load data, temperature data and industry correlation, the long short-term memory (LSTM) neural network is used to predict the medium- and long-term load of the industry. The method of the present invention can achieve high-precision medium- and long-term load forecasting by introducing industry correlation analysis.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and more specifically, to a medium- and long-term load forecasting method taking into account industry relevance. Background Art

[0002] With the rapid development of my country's power industry, the scale of power system is becoming larger and larger, smart grid is booming, and power load is affected by more and more uncertain factors, such as climate, GDP, industrial structure, policy, etc., which poses greater challenges to medium- and long-term power system load forecasting. In addition to these objective factors, the production and operation activities of an industry in an industrial chain will inevitably be affected by upstream and downstream industries, so the impact of the correlation between industries on industry load forecasting cannot be ignored. Comprehensively considering the above multiple influencing factors to improve the accuracy of load forecasting has become a hot topic in the current power system research field.

[0003] The results of medium- and long-term load forecasts are of great significance to the resource scheduling and planning of power systems. For example, when building a new power plant, the construction site, power generation scale, and project period can be selected based on the medium- and long-term load forecast results in the region. High-precision load forecasting is of great value to the operation planning of the power grid, and the development of big data and artificial intelligence technology also provides certain technical support for high-precision load forecasting. At present, many studies have analyzed the impact of various natural and social factors on load forecasting based on big data and deep learning technology, and proposed a variety of load forecasting models. However, most of the research on medium- and long-term industry load forecasting is mainly based on multivariate data such as load data, meteorological data, regional data, and economic data, and does not consider the impact of the correlation between industries on load forecasting, resulting in the low accuracy of industry load forecasting.

[0004] It can be seen that the forecasting method that takes into account the impact of industry correlation on load forecasting and improves the accuracy of medium- and long-term load forecasting needs to be studied. Summary of the invention

[0005] In order to overcome the defects of the above-mentioned prior art, the present invention provides a medium- and long-term load forecasting method taking into account industry correlation, which can consider the impact of industry correlation on load forecasting, thereby improving the accuracy of industry load forecasting.

[0006] The present invention adopts the following technical solutions:

[0007] A medium- and long-term load forecasting method taking into account industry relevance comprises the following steps:

[0008] Aiming at the problems of missing industry load data and large differences in data magnitude, a method for repairing industry monthly load data based on iterative interpolation is proposed to restore the missing monthly load data, and the min-max normalization method is used to normalize the load data.

[0009] Based on the Shapiro-Wilk test and Pearson correlation coefficient, the correlation between industries is analyzed and quantified, and the strongly correlated industries are screened out based on the correlation index (i.e., Pearson correlation coefficient);

[0010] Based on industry load data, temperature data and industry correlation, long short-term memory neural network is used to predict the medium and long-term industry load.

[0011] In the above technical solution, further, in order to solve the problem of missing industry load data and large data magnitude difference, an industry monthly load data repair method based on iterative interpolation is proposed to restore the missing monthly load data, and the min-max normalization method is used to normalize the load data, including the following steps:

[0012] Since the original load data usually has problems such as missing data and large magnitude differences, it is necessary to repair and preprocess the original load data. The medium- and long-term load data of the industry is collected at monthly intervals, and the data input matrix formed can be expressed as:

[0013]

[0014] In the formula, y ij is the monthly load value of industry i in month j; M is the total number of industries; N is the total number of months for data collection; T j represents the meteorological parameters of the jth month.

[0015] To solve the problem of missing load data,

[0016] The iterative interpolation method is used to complete the data. That is, through multiple iterations, the load data of the industry where the missing data is located is regarded as the dependent variable y, and the remaining relevant characteristic data is regarded as the independent variable x. The linear regressor is used for fitting modeling to interpolate the missing y values. After multiple iterations, the complete load data is output. The calculation of the y value is shown in the following formula:

[0017] y=a 1 x 1 +a 2 x 2 +...a M x M

[0018] The load magnitudes of different industries may differ greatly, which will increase the difficulty of model training. Using normalization can speed up gradient descent and thus improve the convergence speed of the model. At the same time, the normalized data will be limited between 0 and 1, which can unify the evaluation standards and improve the calculation speed. The min-max normalization method is used to normalize the load data:

[0019]

[0020] In the formula, x i ' represents the load data of the ith month after min-max standardization; x i represents the original load data of month i; x min Indicates the minimum value of the element in the original load data; x max Indicates the maximum value of the elements in the raw load data.

[0021] Furthermore, the correlation between industries is analyzed and quantified based on the Shapiro-Wilk test and Pearson correlation coefficient, and strongly correlated industries are screened out based on the correlation index. The steps are as follows:

[0022] The correlation coefficient is used to evaluate the strength and direction of the linear relationship between two random variables. The Pearson correlation coefficient is used to analyze the correlation between random variables. The sample data must be normally distributed. The normality test of the load data of each industry is performed based on the Shapiro-Wilk normality test method. Assume that H 0 : The sample data follows a normal distribution; H 1 :The sample data does not follow a normal distribution. By calculating the statistic W, we can determine the similarity between the distribution of the sample to be tested and the normal distribution:

[0023]

[0024] Where W is the statistic returned by the Shapiro-Wilk test. The closer the value of W is to 1, the closer the data is to the standard normal distribution. i Represents the i-th data in the sample; Represents the average value of sample data; a i is the coefficient for the Shapiro-Wilk test, which can be obtained by looking up the table.

[0025] Based on the statistic W, set a significance level α, and then obtain its quantile W α If W<W α , then reject H 0 , otherwise accept H 0 The Shapiro-Wilk test can also calculate the P value of the data sample, which indicates the probability of 0If the P value is less than the significance level α (usually 0.05 in statistics), then reject H 0 .

[0026] The Pearson correlation coefficient r can be used to analyze the correlation between industry X and industry Y. For the industry to be predicted X and the related industry Y, the Pearson correlation coefficient calculation formula is as follows:

[0027]

[0028] In the formula, r represents the correlation coefficient between industry X and industry Y, and its value is between -1 and +1. i and Y i Represent the electricity load data of industry X and industry Y in the i-th month, and Represent the monthly average electricity load of industries X and Y respectively.

[0029] The value range of the correlation coefficient r is between -1 and +1, that is, |r|≤1. The closer |r| is to 1, the higher the linear correlation between X and Y. r=-1 indicates that there is a complete negative linear correlation between X and Y; r=+1 indicates that there is a complete positive linear correlation between X and Y; r=0 indicates that there is no linear correlation between the two.

[0030] Furthermore, based on the industry load data, temperature data and industry correlation, the long short-term memory neural network is used to predict the industry's medium- and long-term load. The steps are as follows:

[0031] LSTM neural network is an improved recurrent neural network (RNN). Compared with RNN, the biggest advantage of LSTM is that it solves the long-term dependency problem, enabling the neural network to remember the hidden load characteristic information of earlier time nodes. Therefore, LSTM has better performance when processing long time series and is suitable for medium- and long-term load forecasting. The core concepts of long short-term memory neural network lie in the state memory unit and the "gate" structure. The state memory unit constitutes the path for load forecasting information transmission, allowing information to be transmitted in the time series; the "gate" structure determines the degree of memory of the load characteristic information during the transmission process. The long short-term memory neural network realizes the transmission of information through three basic "gate" structures, namely the forget gate, input gate and output gate. The transmission process and update steps of industry load characteristic information in the LSTM unit are divided into the following 4 steps:

[0032] (1) Forget Gate

[0033] The function of the forget gate in LSTM is to determine the degree of feature forgetting of the state memory unit. The forget gate takes the load forecast value h from the previous moment ast-1 and the current input feature x t At the same time, the sigmoid function is passed in, and the sigmoid function outputs f t As the forgetting coefficient of the state memory unit. t It can be expressed as:

[0034] f t =σ(W f [h t-1 ,x t ]+b f )

[0035] Where W f h is the output of the forget gate at the previous moment t-1 and the current input feature x t The multiplied weight matrix; b f is the forget gate bias term; σ represents the sigmoid function; h t-1 represents the load forecast output at the previous moment; x t Represents the current input load forecast characteristic factors, including industry correlation characteristics, temperature characteristics, etc.

[0036] (2) Input Gate

[0037] The function of the input gate is to generate the intermediate value of the state memory unit. First, the input gate converts the predicted output value h of the previous moment into t-1 and the new input x t Pass the sigmoid function together and output i t As input node state g t Update coefficient of input node state g t By h t-1 and x t Joint decision.

[0038] The sigmoid function of the input gate outputs i t The expression is as follows:

[0039] i t =σ(W i [h t-1 ,x t ]+b i )

[0040] Where W i h is the input gate that passes the sigmoid function t-1 And the input feature x t The multiplied weight matrix; b i is the bias term of the input gate sigmoid function.

[0041] Input node status g tThe expression is as follows:

[0042] g t =tanh(W g [h t-1 ,x t ]+b g )

[0043] Where W g h is the input gate for the tanh function t-1 And the input feature x t The multiplied weight matrix; b g is the bias term of the input gate tanh function.

[0044] (3) Update state memory unit

[0045] The third step of the LSTM workflow is to update the state memory unit, that is, to update C in the figure t First, the state of the state memory unit at the previous moment is combined with the output of the forget gate f t Multiply and output the selective memory result. Then add this value to the intermediate value of the state memory unit point by point to update the state memory unit C together. t , C t The expression is as follows:

[0046]

[0047] In the formula, is the Hadamard product.

[0048] (4) Output gate

[0049] The function of the output gate is to determine the load forecast value h at the current moment t First, the load forecast value h at the previous moment is t-1 And the input feature x t Pass in the sigmoid function and output o t . Secondly, the updated state memory unit C t Pass it to the tanh function and compare the output with o t Multiply them to get the current load forecast value h t Among them, o t The expression is as follows:

[0050] o t =σ(W o [h t-1 ,x t ]+b o )

[0051] Where W o h t-1 And the input feature xt The multiplied weight matrix; b o is the output gate bias term.

[0052] Current load forecast value h t The expression is as follows:

[0053]

[0054] Based on the above LSTM neural network model, the target industry load data, strongly related industry load data, and temperature data are used as data sets to train the model. In order to evaluate the accuracy of the prediction model, the following evaluation indicators are used:

[0055]

[0056]

[0057] In the formula, y k is the real industry load data, y k ' is the load data predicted by the model, R 1 is the mean absolute percentage error, R 2 is the root mean square error.

[0058] When the evaluation index meets the requirements, the model training is completed. If it does not meet the requirements, the LSTM model parameters are modified to continue training.

[0059] The beneficial effects of the present invention are:

[0060] The present invention adopts an industry monthly load data repair method based on iterative interpolation to realize the recovery of missing industry monthly load data, and the recovery accuracy can reach more than 95%, which effectively solves the data missing problem in existing projects; a quantification method of industry correlation based on the Pearson correlation coefficient is proposed, and strongly correlated industries can be screened out according to the quantification results, and the industry correlation is used as a feature to construct a load forecasting model in combination with the LSTM neural network to predict the medium and long-term load of the industry, which effectively improves the accuracy of medium and long-term load forecasting, and the error is kept within 4%. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a schematic diagram of the overall process of the present invention.

[0062] Figure 2 It is a schematic diagram of the LSTM network unit of the present invention.

[0063] Figure 3 This is the load forecast result curve of the "agricultural and sideline food processing industry" based on LSTM of the present invention.

[0064] Figure 4This is the load forecast result curve of the "food manufacturing industry" based on LSTM of the present invention.

[0065] Figure 5 This is the load forecast result curve of the "wine, beverage and refined tea manufacturing industry" based on LSTM of the present invention. DETAILED DESCRIPTION

[0066] In order to better understand the purpose, technical scheme and technical effect of the present invention, the present invention is further explained below in conjunction with the accompanying drawings.

[0067] The present invention proposes a medium- and long-term load forecasting method taking into account industry relevance, and its implementation process includes the following detailed steps:

[0068] Step 1: A monthly load data repair method based on iterative interpolation is proposed to restore the missing monthly load data of typical industries, and the load data is normalized using the min-max standardization method.

[0069] Since the original load data usually has problems such as missing data and large magnitude differences, it is necessary to repair and preprocess the original load data. The medium- and long-term load data of the industry is collected at monthly intervals, and the data input matrix formed can be expressed as:

[0070]

[0071] In the formula, y ij is the monthly load value of industry i in month j; M is the total number of industries; N is the total number of months for data collection; T j represents the meteorological parameters of the jth month.

[0072] To solve the problem of missing load data,

[0073] The iterative interpolation method is used to complete the data. That is, through multiple iterations, the load data of the industry where the missing data is located is regarded as the dependent variable y, and the remaining relevant characteristic data is regarded as the independent variable x. The linear regressor is used for fitting modeling to interpolate the missing y values. After multiple iterations, the complete load data is output. The calculation of the y value is shown in the following formula:

[0074] y=a 1 x 1 +a 2 x 2 +...+a M x M

[0075] The load magnitudes of different industries may differ greatly, which will increase the difficulty of model training. Using normalization can speed up gradient descent and thus improve the convergence speed of the model. At the same time, the normalized data will be limited between 0 and 1, which can unify the evaluation standards and improve the calculation speed. The min-max normalization method is used to normalize the load data:

[0076]

[0077] In the formula, x i ' represents the load data of the ith month after min-max standardization; x i represents the original load data of month i; x min Indicates the minimum value of the element in the original load data; x max Indicates the maximum value of the elements in the raw load data.

[0078] Step 2: Based on the Shapiro-Wilk test and Pearson correlation coefficient, analyze and quantify the correlation between industries, and screen out strongly correlated industries based on correlation indicators.

[0079] The correlation coefficient is used to evaluate the strength and direction of the linear relationship between two random variables. The Pearson correlation coefficient is used to analyze the correlation between random variables. The sample data must be normally distributed. The normality test of the load data of each industry is performed based on the Shapiro-Wilk normality test method. Assume that H 0 : The sample data follows a normal distribution; H 1 :The sample data does not follow a normal distribution. By calculating the statistic W, we can determine the similarity between the distribution of the sample to be tested and the normal distribution:

[0080]

[0081] Where W is the statistic returned by the Shapiro-Wilk test. The closer the value of W is to 1, the closer the data is to the standard normal distribution. i Represents the i-th data in the sample; Represents the average value of sample data; a i is the coefficient for the Shapiro-Wilk test, which can be obtained by looking up the table.

[0082] Based on the statistic W, set a significance level α, and then obtain its quantile W α If W<W α , then reject H 0 , otherwise accept H 0 The Shapiro-Wilk test can also calculate the P value of the data sample, which indicates the probability of 0If the P value is less than the significance level α (usually 0.05 in statistics), then reject H 0 .

[0083] The Pearson correlation coefficient r can be used to analyze the correlation between industry X and industry Y. For the industry to be predicted X and the related industry Y, the Pearson correlation coefficient calculation formula is as follows:

[0084]

[0085] In the formula, r represents the correlation coefficient between industry X and industry Y, and its value is between -1 and +1. i and Y i Represent the electricity load data of industry X and industry Y in the i-th month, and Represent the monthly average electricity load of industries X and Y respectively.

[0086] The value range of the correlation coefficient r is between -1 and +1, that is, |r|≤1. The closer |r| is to 1, the higher the linear correlation between X and Y. r=-1 indicates that there is a complete negative linear correlation between X and Y; r=+1 indicates that there is a complete positive linear correlation between X and Y; r=0 indicates that there is no linear correlation between the two.

[0087] Based on the above correlation coefficient theory, the industry correlation evaluation index is formulated as shown in Table 1.

[0088] Table 1 Industry relevance evaluation indicators

[0089] Pearson correlation coefficient range |r| Industry relevance assessment 0.8~1.0 Very strong correlation 0.6~0.8 Strong correlation 0.4~0.6 Moderately related 0.2~0.4 Weak correlation 0~0.2 Very weak correlation

[0090] Based on the calculated Pearson correlation coefficient, industry correlation is evaluated and industries with extremely strong correlation are screened out.

[0091] Step 3: Based on industry load data, temperature data and industry correlation, use long short-term memory neural network to predict the medium and long-term load of the industry.

[0092] LSTM neural network is an improved recurrent neural network (RNN). Compared with RNN, the biggest advantage of LSTM is that it solves the long-term dependency problem and enables the neural network to remember the hidden load characteristic information of earlier time nodes. Therefore, LSTM has better performance when processing long time series and is suitable for medium and long-term load forecasting. The core concepts of long short-term memory neural network lie in the state memory unit and the "gate" structure. The state memory unit constitutes the path for load forecasting information transmission, allowing information to be transmitted in the time series; the "gate" structure determines the degree of memory of the load characteristic information during the transmission process. The long short-term memory neural network realizes the transmission of information through three basic "gate" structures, namely the forget gate, input gate and output gate, such as Figure 2 The transmission process and update steps of industry load characteristic information in LSTM units are divided into the following four steps:

[0093] (1) Forget Gate

[0094] The function of the forget gate in LSTM is to determine the degree of feature forgetting of the state memory unit. The forget gate takes the load forecast value h from the previous moment as t-1 and the current input feature x t At the same time, the sigmoid function is passed in, and the sigmoid function outputs f t As the forgetting coefficient of the state memory unit. t It can be expressed as:

[0095] f t =σ(W f [h t-1 ,x t ]+b f )

[0096] Where W f h is the output of the forget gate at the previous moment t-1 and the current input feature x t The multiplied weight matrix; b f is the forget gate bias term; σ represents the sigmoid function; h t-1 represents the load forecast output at the previous moment; x t Represents the current input load forecast characteristic factors, including industry correlation characteristics, temperature characteristics, etc.

[0097] (2) Input Gate

[0098] The function of the input gate is to generate the intermediate value of the state memory unit. First, the input gate converts the predicted output value h of the previous moment into t-1 and the new input x t Pass the sigmoid function together and output i tAs input node state g t Update coefficient of input node state g t By h t-1 and x t Joint decision.

[0099] The sigmoid function of the input gate outputs i t The expression is as follows:

[0100] i t =σ(W i [h t-1 ,x t ]+b i )

[0101] Where W i h is the input gate that passes the sigmoid function t-1 And the input feature x t The multiplied weight matrix; b i is the bias term of the input gate sigmoid function.

[0102] The tanh function output g of the input gate t The expression is as follows:

[0103] g t =tanh(W g [h t-1 ,x t ]+b g )

[0104] Where W g h is the input gate for the tanh function t-1 And the input feature x t The multiplied weight matrix; b g is the bias term of the input gate tanh function.

[0105] (3) Update state memory unit

[0106] The third step of the LSTM workflow is to update the state memory unit, that is, to update C in the figure t First, the state of the state memory unit at the previous moment is combined with the output of the forget gate f t Multiply and output the selective memory result. Then add this value to the intermediate value of the state memory unit point by point to update the state memory unit C together. t , C t The expression is as follows:

[0107]

[0108] In the formula, is the Hadamard product.

[0109] (4) Output gate

[0110] The function of the output gate is to determine the load forecast value h at the current moment t The output gate first converts the load forecast value h t-1 And the input feature x t Pass in the sigmoid function and output o t Secondly, the output gate updates the state memory unit C t Pass it to the tanh function and compare the output with o t Multiply them to get the current load forecast value h t Among them, o t The expression is as follows:

[0111] o t =σ(W o [h t-1 ,x t ]+b o )

[0112] Where W o h t-1 And the input feature x t The multiplied weight matrix; b o is the output gate bias term.

[0113] Current load forecast value h t The expression is as follows:

[0114]

[0115] In the formula, C t is the updated state memory unit.

[0116] Based on the above LSTM neural network model, the target industry load data, strongly related industry load data, and temperature data are used as data sets to train the model. In order to evaluate the accuracy of the prediction model, the following evaluation indicators are used:

[0117]

[0118]

[0119] In the formula, y k is the real industry load data, y k ' is the load data predicted by the model, R 1 is the mean absolute percentage error, R 2 is the root mean square error.

[0120] When the evaluation index meets the requirements, the model training is completed. If it does not meet the requirements, the LSTM model parameters are modified to continue training.

[0121] In order to verify the effectiveness and accuracy of the proposed medium- and long-term load forecasting method taking into account industry correlation, a simulation example was carried out using the monthly load data of a certain city from 2016 to 2020, which included 2 key industries and more than 30 related industries.

[0122] Taking the “food manufacturing industry” load data as an example, two months of load data are randomly selected as missing data points for iterative interpolation. The interpolated load data are compared with the real load data for error analysis. The results are shown in Table 2.

[0123] Table 2 Comparison between interpolated data and real data

[0124] Data missing point 1 Data missing point 2 Interpolation data 16566.41 10648.95 Real data 15944.91 10277.74 Relative error 3.90% 3.61%

[0125] It can be seen from the above table that the results of iterative interpolation match the actual situation well.

[0126] Next, taking food-related industries as an example, we conducted an industry load correlation analysis on "agricultural and sideline food processing industry", "food manufacturing industry", "wine, beverage and refined tea manufacturing industry" and "tobacco manufacturing industry". First, we conducted a Shapiro-Wilk test on the above industry load data, and the results are shown in Table 3.

[0127] Table 3 Normal distribution test results of industry load data

[0128]

[0129]

[0130] From Table 3, we can see that "agricultural and sideline food processing industry", "food manufacturing industry" and "wine, beverage and refined tea manufacturing industry" obey the normal distribution and can be analyzed using the Pearson correlation coefficient method.

[0131] After verifying that the industry load data meets the normal distribution conditions, the Pearson correlation coefficient method is used to calculate the correlation coefficient matrix between industries. The correlation analysis is carried out with a 12-month cycle, and finally four correlation coefficients between each industry from 2016 to 2019 are obtained. The average value is taken as the final correlation coefficient matrix element, as shown in Table 4.

[0132] Table 4 Industry load data correlation coefficient matrix

[0133] industry Agricultural and sideline food processing industry Food Manufacturing Wine, beverage and refined tea manufacturing industry Agricultural and sideline food processing industry 1 0.962799 0.916886 Food Manufacturing 0.962799 1 0.792809 Wine, beverage and refined tea manufacturing industry 0.916886 0.792809 1

[0134] Based on the LSTM neural network prediction method taking into account industry relevance, the industry load data is divided into a training set and a test set, and the model is trained to obtain the following Figures 3 to 5 The industry load forecast curve is shown in Figure 2 and the industry load forecast error analysis table is shown in Table 5.

[0135] According to Table 5, after considering the industry correlation, the error of the load forecast results of the food-related industry is small, and the forecast results meet the accuracy requirements well.

[0136] Table 5 Industry load forecast error analysis

[0137] industry <![CDATA[R 1 ]]> <![CDATA[R 2 ]]> Agricultural and sideline food processing industry 3.963% 839.881 Food Manufacturing 1.166% 232.081 Wine, beverage and refined tea manufacturing industry 0.987% 81.246

[0138] In order to evaluate the value of considering industry correlation for medium- and long-term industry load forecasting, based on the load data of "agricultural and sideline food processing industry", "food manufacturing industry" and "wine, beverage and refined tea manufacturing industry", the load forecasting method that does not consider industry correlation is compared with the above method. The comparative forecast results are shown in Table 6.

[0139] Table 6 Comparison of prediction results with and without considering industry correlation

[0140]

[0141] It can be seen from Table 6 that the prediction evaluation indicators considering industry correlation are better than those not considering industry correlation, which proves that considering industry correlation can effectively improve the accuracy of medium- and long-term load forecasting in the industry.

[0142] In summary, the method of the present invention realizes the recovery of missing monthly load data in typical industries, with a recovery accuracy of more than 96%, solving the problem of missing data in actual business. At the same time, the error of medium- and long-term load forecasting is kept within 5%, effectively improving the accuracy of medium- and long-term load forecasting.

Claims

1. A medium- and long-term load forecasting method taking into account industry relevance, characterized in that: The following steps are involved: (1) The monthly load data repair method based on iterative interpolation is used to restore the missing data of monthly load, and the load data is normalized using the min-max normalization method; (2) Analyze and quantify the correlation between industries based on the Shapiro-Wilk test and Pearson correlation coefficient, and select highly correlated industries based on the correlation index; (3) Based on industry load data, temperature data and industry correlation, long-term and short-term memory neural networks are used to predict the medium- and long-term load of the industry; The step (2) is specifically as follows: The Pearson correlation coefficient is used to analyze the correlation between random variables. The sample data must be normally distributed. The normality test of the load data of each industry is performed based on the Shapiro-Wilk test. Assumptions H0: The sample data obeys the normal distribution; H1: The sample data does not obey the normal distribution. The statistic W is calculated to determine the similarity between the sample distribution to be tested and the normal distribution: Where W is the statistic returned by the Shapiro-Wilk test. The closer the value of W is to 1, the closer the data is to the standard normal distribution. i Represents the i-th data in the sample; Represents the average value of sample data; a i is the coefficient for the Shapiro-Wilk test, which can be obtained by looking up the table; Based on the statistic W, set a significance level α, and then obtain its quantile W α ; If W<W α , then reject H0, otherwise accept H0; and use Shapiro-Wilk test to calculate the P value of the data sample. If the P value is less than the significance level α, then reject H0; The Pearson correlation coefficient r is used to analyze the correlation between industry X and industry Y. For the predicted industry X and related industry Y, the Pearson correlation coefficient calculation formula is as follows: In the formula, r represents the correlation coefficient between industry X and industry Y; i and Y i Represent the electricity load data of industry X and industry Y in the i-th month, and Represent the monthly average electricity load of industries X and Y respectively; The value range of the correlation coefficient r is between -1 and +1, that is, |r|≤1; the closer |r| is to 1, the higher the linear correlation between X and Y; r = -1, indicating that there is a complete negative linear correlation between X and Y; r = +1, indicating that there is a complete positive linear correlation between X and Y; r = 0, indicating that there is no linear correlation between the two; The step (3) is specifically as follows: (1) The load forecast value h from the previous moment t-1 and the current input feature x t At the same time, the sigmoid function is passed in, and the sigmoid function outputs f t As the forgetting coefficient of the state memory unit; f t It can be expressed as: f t =σ(W f [h t-1 ,x t ]+b f ) Where W f h is the output of the forget gate at the previous moment t-1 and the current input feature x t The multiplied weight matrix; b f is the forget gate bias term; σ represents the sigmoid function; h t-1 represents the load forecast value at the previous moment; x t Represents the current input load forecast characteristic factors, including industry correlation characteristics and temperature characteristics; (2) The predicted output value h at the previous moment t-1 and the new input x t Pass the sigmoid function together and output i t As input node state g t The update coefficient of Output t The expression is as follows: i t =σ(W i [h t-1 ,x t ]+b i ) Where W i h is the input gate that passes the sigmoid function t-1 And the input feature x t The multiplied weight matrix; b i is the bias term of the input gate sigmoid function; Input node status g t The expression is as follows: g t =tanh(W g [h t-1 ,x t ]+b g ) Where W g h is the input gate for the tanh function t-1 And the input feature x t The multiplied weight matrix; b g is the bias term of the input gate tanh function; (3) Update state memory unit C t :First, the state of the state memory unit at the previous moment and the output f of the forget gate t Multiply them and output the selective memory result; then add the selective memory result and the intermediate value of the state memory unit point by point to update the state memory unit C together t , C t The expression is as follows: In the formula, is the Hadamard product; (4) Determine the current load forecast value h t First, the load forecast value h at the previous moment is t-1 And the input feature x t Pass in the sigmoid function and output o t ; Secondly, update the state memory unit C t Pass it to the tanh function and compare the output with o t Multiply them to get the current load forecast value h t ; Among them, t The expression is as follows: the t =σ(W o [h t-1 ,x t ]+b o ) Where W o is the load forecast value at the previous moment h t-1 And the input feature x t The multiplied weight matrix; b o is the output gate bias term; Current load forecast value h t The expression is as follows:

2. A medium- and long-term load forecasting method taking into account industry relevance according to claim 1, characterized in that: The step (1) is specifically as follows: The medium- and long-term load data of the industry is collected at monthly intervals, and the resulting data input matrix can be expressed as: In the formula, y ij is the monthly load value of industry i in month j; M is the total number of industries; N is the total number of months for data collection; T j represents the meteorological parameters of the jth month; In order to solve the problem of missing load data, the iterative interpolation method is used to complete the data. That is, through multiple iterations, the load data of the industry where the missing data is located is regarded as the dependent variable y, and the remaining relevant characteristic data is regarded as the independent variable x. The linear regressor is used for fitting modeling to interpolate the missing y value, and multiple iterations are repeated to output the complete load data. The calculation of the y value is shown in the following formula: y=a1x1+a2x2+...a M x M The load data is normalized using the min-max standardization method: In the formula, x i ' represents the load data of the ith month after min-max standardization; x i represents the original load data of month i; x min Indicates the minimum value of the element in the original load data; x max Indicates the maximum value of the elements in the raw load data.