Enterprise power load prediction method based on multi-influence factor STIRPAT model

Abstract

The invention discloses an enterprise power load prediction method based on a multi-influence-factor STIRPAT model, and the method comprises the steps: carrying out the correlation analysis through collecting the related historical data of the power consumption of an enterprise, and screening key influence factors having significant correlation with the power consumption of the enterprise; establishing a relation model between each key influence factor and the enterprise power load by adopting an extended STIRPAT model; and performing time sequence prediction on the data of each influence factor in the future by using the ARMA model to obtain variable values of each key influence factor in the future time period, inputting the variable values into the STIRPAT model, and calculating the power load in the corresponding time period. By adopting the method, the demand prediction and optimization problems of enterprises in medium and long term power load planning and management can be solved.

Description

Technical Field

[0001] The present invention relates to the field of power systems, and in particular to a medium- and long-term curve deduction technology for enterprise-level power load based on a multi-influencing factor ridge regression-STIRPAT model. Background Art

[0002] With the rapid development of the global economy and the acceleration of industrialization, the management of corporate power loads has become increasingly complex. Traditional power load forecasting methods, such as time series analysis and linear regression models based on historical data, have a certain application effect in short-term forecasting, but often face the problem of insufficient accuracy in medium- and long-term forecasting. This is mainly because these traditional methods mainly rely on past power consumption data and fail to fully consider the diverse factors that affect power demand.

[0003] Changes in power load are not only affected by historical power consumption, but also by the combined effects of multiple factors, including the production activities of enterprises, product price fluctuations, raw material price changes, market demand changes, and technological progress. For example, rising product prices may lead to increased production costs, thereby affecting production plans and power demand; fluctuations in raw material prices will also have a direct impact on production costs; technological progress may lead to improved production efficiency, which in turn affects power demand patterns.

[0004] In addition, existing load forecasting methods often have limitations when dealing with these complex, multi-dimensional factors. For example, although time series models can capture certain time dependencies, it is difficult to model the impact of complex economic, social and environmental factors; although regression analysis can consider multiple influencing factors, it has limited effectiveness when faced with nonlinear and high-dimensional data.

[0005] As a flexible environmental impact assessment tool, the STIRPAT (Stochastic Impacts by Regression on Population, Affluence, and Technology) model can simultaneously consider multiple influencing factors such as population, wealth, and technology, and can handle complex nonlinear relationships. However, the existing STIRPAT model applications are mainly concentrated in the field of environmental impact assessment, and are less used in the medium- and long-term forecasting of power load. The advantage of the STIRPAT model is that it can comprehensively consider the impact of multiple variables and make predictions and analyses on this basis, but it has not been fully utilized in corporate power load forecasting. Summary of the invention

[0006] The purpose of the present invention is to improve the accuracy and reliability of power load forecasting and to be able to effectively deduce the medium- and long-term change trend of enterprise power load.

[0007] In order to achieve the above object, the technical solution provided by the present invention is as follows:

[0008] The enterprise power load forecasting method based on the multi-influencing factor STIRPAT model includes the following steps:

[0009] S1. Analysis of factors affecting electricity consumption: Collect relevant historical data on the electricity consumption of enterprises, conduct statistical analysis on it, use correlation analysis to screen key factors that have significant correlation with the electricity consumption of enterprises, and select several of them as multi-dimensional factors for STIRPAT modeling;

[0010] S2. STIRPAT model construction: Based on the collected historical data, the extended STIRPAT model is used to establish the relationship model between various influencing factors and enterprise power load;

[0011] S3, ARMA forecast: Use the ARMA model to perform time series forecasting on the data of various influencing factors in the future to obtain the variable values ​​of various influencing factors in the future time period;

[0012] S4. Input the variable values ​​of each influencing factor into the STIRPAT model and calculate the power load in the corresponding time period.

[0013] This paper integrates the historical data of enterprises, such as product output, product price, raw material price, energy consumption and other factors, and combines the power consumption data of enterprises to obtain key influencing factors through correlation analysis, and selects several of them as multi-dimensional factors for modeling and analysis using the STIRPAT model. Subsequently, the ARMA model is used to predict each key influencing factor, and these prediction results are combined with the STIRPAT model to deduce the medium- and long-term change trend of the enterprise's power load.

[0014] The present invention uses the ARMA model to perform time series forecasting on the future influencing factor variables and introduces the STIRPAT model, which can effectively capture the multi-dimensional influencing factors in the enterprise development process and combine them with the changing trend of power load, thereby realizing the accurate deduction of the medium- and long-term curve of the enterprise's future power load.

[0015] Furthermore, the enterprise power load forecasting method further includes the following steps:

[0016] S4. Generation of power load curve: According to the above steps, the variable values ​​of the influencing factors in the future time period are obtained based on ARMA, and input into the STIRPAT model to generate the medium- and long-term change curve of the enterprise power load in the future time period.

[0017] The extended STIRPAT model formula in step S2 is as follows:

[0018] E=Pa a ×Pb b ×PC c ×...×β α

[0019] Where E is the charge, Pa, Pb, Pc, ... are the influencing factors, β is the error term, and a, b, c, ..., α are the percentages of charge change caused by a 1% change in the corresponding influencing factor.

[0020] The ARMA model in step S3 is constructed using the following method:

[0021] S301. Data preparation: Conduct time series analysis on the influencing factor variables that need to be predicted to ensure the stability of the data and handle missing data;

[0022] S302, model selection: select the appropriate order (p, q) by AIC / BIC criteria, where p represents the number of lag periods and q represents the number of lag error periods;

[0023] S303, model fitting: fitting the parameters of the ARMA model by maximum likelihood estimation method.

[0024] The following method is used to ensure the stability of the data in step S301: ADF test is performed using the following formula:

[0025]

[0026] ∈ t ΔY t =Y t -Y t-1 p is required

[0027] Among them, t is the time trend term, is the error term, is the difference value of the data, and is the key parameter for ADF detection. If p is greater than 0.05, it means that the sequence is non-stationary;

[0028] If the time series is non-stationary, use the second-order difference method to transform it into a stationary series.

[0029] In the above step S302, the ARMA model order (p, q) is selected by the following method:

[0030] The autoregressive term AR(p) models the linear relationship between the current value and the value of the previous p periods, where p represents the number of lag periods:

[0031] Y t =φ1Y t-1 +φ2Y t-2 +...+φ p Y t-p +∈ t ;

[0032] The moving average term MA(q) models the linear relationship between the current error and the error of the previous q periods, where q represents the number of lagged error periods:

[0033] Y t = t +θ 1t-1 +θ 2t-2 +...+θ qt-q

[0034] The best model order is selected by AIC and BIC criteria:

[0035]

[0036]

[0037] Where k is the number of parameters in the model, n is the sample size, is the maximum likelihood estimate.

[0038] Compared with the prior art, the present invention has the following advantages:

[0039] 1. Integration of STIRPAT model and power load forecasting: This paper applies the STIRPAT (Stochastic Impacts by Regression on Population, Affluence, and Technology) model to the medium- and long-term forecasting of enterprise-level power load for the first time. By incorporating the multi-dimensional influencing factors of the STIRPAT model into the power load forecasting, it is possible to comprehensively consider the comprehensive impact of various factors such as product output, product price or technology (energy consumption) on power demand, thereby improving the accuracy and scientificity of the forecast.

[0040] 2. Comprehensive modeling of multiple data sources: The present invention combines the company's historical product output, product prices, raw material prices and other economic data as well as the company's electricity consumption to establish a forecasting model that comprehensively considers multiple factors. Compared with the traditional method that relies solely on historical electricity consumption data, the present invention can more comprehensively reflect the company's actual electricity demand and changing trends.

[0041] 3. Combination of ARMA time series prediction: The present invention uses the ARMA (Auto-Regressive Moving Average) model to perform time series prediction on future key influencing factor variables (such as product output, product price, and raw material price). Combining the ARMA model with the STIRPAT model can take into account the autocorrelation and moving average effect of time series data in the prediction process, further improving the accuracy of the prediction.

[0042] 4. Dynamic adjustment and real-time optimization: The method of the present invention not only provides static power load forecasting, but also has the ability of dynamic adjustment and real-time optimization. Through the real-time update of the key influencing factor variable values ​​and the adjustment of the STIRPAT model, it can timely reflect the changes in the enterprise's operating environment, support dynamic adjustment of power demand forecasting, and help enterprises respond to market and environmental changes more flexibly.

[0043] 4. Enhance the practical application value of prediction: This technology provides a practical and operational prediction solution, which provides a scientific basis for the medium- and long-term planning of the power system by comprehensively considering the enterprise's operating data and external economic factors. Its application can optimize the configuration of the enterprise's power resources, improve the efficiency of power dispatching, and reduce electricity costs, thereby enhancing the power system's ability to respond to the actual needs of the enterprise.

[0044] The above innovations not only make up for the shortcomings of traditional power load forecasting methods, but also apply advanced statistical models and data analysis technologies to actual power load management. They can better adapt to complex enterprise operating environments, provide more accurate power load forecasting results, and help achieve reasonable scheduling and allocation of enterprise power resources. They have significant technological progress and practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a flow chart of the enterprise power load forecasting method of the present invention;

[0046] Figure 2 The comparison results of the STIRPAT model and the IPAT model in Example 2 of the present invention;

[0047] In the figure, the red color is the prediction result of the STIRPAT model, the blue color is the prediction result of the IPAT model, and the green curve is the actual electricity consumption;

[0048] Figure 3 The product price curve predicted by the ARMA model in Example 3 of the present invention;

[0049] Figure 4 The ARMA model predicts the yield curve of product 6 in Example 3 of the present invention;

[0050] Figure 5 The enterprise electricity consumption prediction curve in Example 3 of the present invention. DETAILED DESCRIPTION

[0051] The present invention is described in detail below with reference to the accompanying drawings.

[0052] Example 1

[0053] The present invention provides an enterprise power load forecasting method based on a multi-factor STIRPAT model, such as Figure 1As shown, the specific steps are as follows:

[0054] 1. Determination of influencing factors

[0055] Before predicting the power load of an enterprise, it is necessary to conduct a systematic analysis of the main factors affecting the power load to ensure that the variables considered in the model are reasonable and representative.

[0056] Historical data analysis: Through statistical analysis of the company's historical data and using correlation analysis, we can preliminarily determine which factors have a strong correlation with electricity consumption and screen out variables that have a significant impact on power load.

[0057] Specific process:

[0058] (1) Data preparation

[0059] First, it is necessary to sort out the historical data related to the company's electricity consumption. This data usually includes the following two categories:

[0060] Electricity consumption data: The actual electricity consumption of an enterprise over different time periods (e.g. monthly or annual electricity consumption).

[0061] Data on potential influencing factors: including product output, product price, raw material price, market demand, technology level, production scale, etc.

[0062] (2) Data preprocessing

[0063] Before conducting correlation analysis, ensure the completeness and accuracy of the data:

[0064] Missing value handling: filling or removing missing values ​​in data.

[0065] Outlier detection: Identify and handle possible outliers that may significantly affect the results of correlation analysis.

[0066] Standardization or normalization (optional): For data with inconsistent dimensions (such as price and output), the impact of the dimensions can be eliminated by standardization or normalization. This can make variables in different units comparable during analysis.

[0067] (3) Calculate the Pearson correlation coefficient

[0068] This coefficient measures the linear relationship between two variables and its value ranges from -1 to 1:

[0069] 1 indicates a perfect positive correlation, meaning that when one variable increases, the other also increases.

[0070] -1 indicates a perfect negative correlation, meaning that when one variable increases, the other decreases.

[0071] 0 means no correlation.

[0072] The Pearson correlation coefficient calculation formula is:

[0073]

[0074] Among them, X i and Y i Represents the sample data of two variables respectively. and are the means of the two variables.

[0075] Calculate the Pearson correlation coefficient between each influencing factor (such as product output, price, raw material price) and electricity consumption.

[0076] By generating a correlation coefficient matrix, we can know the correlation results of each factor.

[0077] (4) Analyze the correlation results

[0078] Based on the results of the correlation analysis, it is evaluated which factors have a strong correlation with electricity consumption.

[0079] Positively correlated factors: If the correlation coefficient of a variable is close to 1, it means that there is a positive correlation between the factor and electricity consumption. For example, there may be a strong positive correlation between product output and electricity consumption. When output increases, the company's electricity consumption will usually increase.

[0080] Negatively correlated factors: If the correlation coefficient is close to -1, it means that there is a negative correlation between the factor and electricity consumption. For example, some companies may choose to reduce production when product prices are high, so prices and electricity consumption may be negatively correlated.

[0081] No significant correlation: If the correlation coefficient is close to 0, it means that there is no significant linear relationship between the factor and electricity consumption. At this time, you can consider removing these variables from the model or further testing their nonlinear relationship.

[0082] Common influencing factors:

[0083] Enterprise production type: Different production processes and product types consume different amounts of electricity. For example, high-energy-consuming industries such as steel and chemicals tend to have greater demand for electricity, while light industries have relatively lower demand for electricity.

[0084] Product output: Usually product output is positively correlated with power consumption, and the correlation is large, but it is not a linear relationship.

[0085] Market environment: The market demand for products directly affects the production plan of enterprises. When the market demand is strong, enterprises increase production, thereby increasing electricity consumption.

[0086] Price fluctuations: Price fluctuations of products and raw materials have a significant impact on the production plans of enterprises, which in turn affects the power load.

[0087] Energy consumption per unit product: Generally, technological updates will reduce the energy consumption per unit product, further reducing the company's total electricity consumption.

[0088] Policies and regulations: Government energy policies, environmental standards, etc. also affect the electricity consumption of enterprises. For example, mandatory energy conservation and emission reduction policies will prompt enterprises to adopt energy-saving technologies, thereby reducing electricity load.

[0089] Based on the analysis of historical data of various enterprises, the analysis conclusions of relevant influencing factors are as follows:

[0090] Product output: It is positively correlated with electricity consumption, and the correlation coefficient is significantly positive (the correlation coefficient of some enterprises may be close to 0.8 or higher). When the output increases, the electricity demand increases accordingly. Product output is one of the important driving factors of electricity load.

[0091] Product price: Product price has a direct impact on the production activities of enterprises, especially when market demand is sensitive. When prices are high, enterprises may reduce production, and the correlation is generally negative (range is between -0.4 and -0.6). Therefore, price fluctuations have a certain negative impact on power load.

[0092] Raw material prices: Raw material prices affect the power load by affecting the production costs of enterprises. Generally speaking, the correlation between raw material prices and power load may be low (0.05 to 0.1), but in some industries (such as the chemical industry), the correlation is relatively high (0.3 to 0.5).

[0093] Market demand: Market demand directly affects the production planning and operation mode of enterprises. Increased demand usually leads to increased power load, and the correlation coefficient is usually high, close to 0.7-0.9.

[0094] Technological progress and energy consumption level: Technological progress has a significant impact on power load, especially the introduction of energy-saving technologies may significantly reduce power consumption. This paper quantifies the technological level by the number of patent applications per year. The correlation of technological progress is generally negative (-0.5 to -0.7) because technological progress usually reduces the demand for electricity. However, scientific and technological progress may lead to a decrease in overall energy consumption, but because the company's electricity replaces coal, oil and other resources, the power consumption increases, so the power consumption level also needs to be considered comprehensively.

[0095] Policy factors: These are long-term plans at the national or provincial level. In key industries, policies will limit and affect the energy consumption, industrial scale, development trends, etc. of enterprises. The impact of policies on different industries may vary greatly. For example, the steel industry, petrochemical industry and other industries are greatly affected by policy factors (the steel industry controls the production capacity to decrease by about 1% per year during the 14th Five-Year Plan period), while the electronics, metal products and other industries are less affected.

[0096] 2. Model establishment

[0097] 2.1 Establishment of the STIRPAT model

[0098] The STIRPAT (Stochastic Impacts by Regression on Population, Affluence, and Technology) model is a multidimensional factor model that allows the impact of changes in different influencing factors on the power load of an enterprise to be quantified. Based on the collected historical data, the extended STIRPAT model is used to establish a relationship model between each influencing factor (selected from the key influencing factors in the above steps based on the historical data related to the enterprise's electricity consumption) and the enterprise's power load. The STIRPAT model can flexibly handle multiple influencing factors and can reflect nonlinear relationships. The formula for the extended STIRPAT model is:

[0099] The specific steps are as follows:

[0100] Selection of model form: The STIRPAT model describes the relationship between influencing factors and target variables (enterprise power load) in the form of regression. The basic form is:

[0101] E=Pa a ×Pb b ×PC c ×...×β α ;

[0102] Among them, E is the electricity, Pa, Pb, Pc, ... are the output / price / energy consumption levels of different types of products (i.e., the key influencing factor variables), β is the error term, a, b, c, ..., α are the model index parameter terms that need to be estimated, which are statistical regression coefficients, that is, the percentage of electricity change caused by a 1% change in the variable, and its value can be negative. The extended STIRPAT model only needs to satisfy the multiplication form, and does not need to strictly satisfy the Kaya identity.

[0103] Extend the STIRPAT model to decompose LMDI, base period (initial time electricity E 0 ) and period t (the electricity consumption E in the tth year t ) is:

[0104] ΔE=E T -E 0 =ΔE P +ΔE r +...

[0105]

[0106]

[0107] In the formula, Pa and Pr represent the output and product price of product a respectively. As an example, the change in electricity is not only related to the ratio of the current period to the base period of each variable, but also to the values ​​of the regression coefficients a, b, c...

[0108] Logarithmic transformation: In order to simplify calculations and regression analysis, the multiplicative form of the model is usually logarithmically transformed into a linear form:

[0109] ln(E)=a·ln(Pa)+b·ln(Pb)+c·ln(Pc)+...+αln(β)

[0110] This makes it easy to estimate the elasticity of each influencing factor, that is, the corresponding change in power load when each factor changes by 1%.

[0111] The increment of electricity of each variable in the extended STIRPAT model is related to the numerical values ​​of the statistical regression coefficients a, b, c, d, e, f, g, etc. in the model; compared with the IPAT model, in order to satisfy the Kaya identity, the IPAT model needs to add new variables based on the original assumptions, while the STIRPAT model does not need this; the statistical regression method incorporates influencing factors other than the assumed variables into the statistical regression constant and regression coefficient, and adds the ln(β) term. The IPAT model that satisfies the Kaya identity does not have statistical errors, and some of the newly added variables often lack practical significance.

[0112] Parameter estimation: Use the company's historical data to estimate the parameters in the above model. Ordinary least squares (OLS) or ridge regression methods can be used to fit the parameters to improve the model's predictive power.

[0113] Model testing: After the model is established, statistical tests are performed on the model to evaluate the model's fitting effect, explanatory power, and prediction accuracy.

[0114] 2.2 Establishment of ARMA model

[0115] The ARMA (Auto-Regressive Moving Average) model is used to make time series forecasts for key influencing variables such as product output, product price, and energy consumption level of an enterprise. The steps are as follows:

[0116] Data preparation: Conduct time series analysis on key influencing factor variables that need to be predicted (such as product output, product price, energy consumption level, etc.) to ensure the stability of the data (stationarity test, such as ADF test) and handle missing data.

[0117] Use the ADF (Augmented Dickey-Fuller) test to determine whether the sequence has a unit root, that is, whether it is non-stationary. If the p value is less than 0.05, it means that the sequence is stationary. If the p value is greater than 0.05, it means that the sequence is non-stationary.

[0118] The ADF detection formula with constant term and trend term is as follows:

[0119]

[0120] Among them, t is the time trend term, α i represents the lagged difference term coefficient, ∈ t is the error term, ΔY t =Y t -Y t-1

[0121] p is a key parameter that requires ADF detection. If p is greater than 0.05, the sequence has a single data difference value, a root, indicating that the sequence is non-stationary.

[0122] If the time series is non-stationary, the second-order difference method can be used to convert it into a stationary series. The differenced series can be calculated to make the mean and variance stable.

[0123] Second order difference: Δ 2 Y t =ΔY t -ΔY t-1 .

[0124] Model selection: The ARMA model consists of an autoregressive (AR) term and a moving average (MA) term. It is necessary to select the appropriate ARMA model order (p, q) based on the characteristics of the time series.

[0125] Autoregressive term (AR): The AR(p) part models the linear relationship between the current value and the value of the previous p periods. p represents the number of lag periods.

[0126] Y t =φ1Y t-1 +φ2Y t-2 +...+φ p Y t-p +∈ t ,

[0127] Where Y tis the time series data, φ1, φ2, ... are the autoregressive coefficients, p is the order of the autoregressive model, ∈ t is a random error term.

[0128] Moving Average (MA): The MA(q) part models the linear relationship between the current error and the error from the previous q periods. q represents the number of periods of lagged error.

[0129] Y t = t +θ 1t-1 +θ 2t-2 +...+θ qt-q .

[0130] Where θ1, θ2, ... are moving average coefficients, q is the order of the moving average model, ∈ t is a random error term.

[0131] Order determination method: The best model order is selected by Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). We usually choose the model with the smallest AIC / BIC value.

[0132]

[0133]

[0134] Where k is the number of parameters in the model, n is the sample size, is the maximum likelihood estimate.

[0135] Model Fitting:

[0136] After determining the order of AR and MA, use historical data to estimate the parameters of the model:

[0137] The parameters of the ARMA model are fitted by the maximum likelihood estimation (MLE) method. This step finds the optimal parameters by maximizing the likelihood function of the model on the known data.

[0138]

[0139] Among them, L(θ) is the likelihood function, f(Y t |θ) is the data Y under given θ t The probability density function of .

[0140] After fitting the model, you need to check the residuals (i.e. the difference between the predicted value and the actual value). Ideally, the residuals should be white noise, i.e. no autocorrelation and zero mean. This can be verified by the following test method:

[0141] Ljung-Box test: used to test whether the residuals are white noise. If the p-value is greater than 0.05, it can be considered that the residuals are random and the model fit is good.

[0142] Forecast: Based on the fitted ARMA model, forecast future product output, price and other variables. The ARMA model is suitable for processing time series data with autocorrelation and random disturbances, and can better capture trends and cyclical changes.

[0143] 3. Derivation of prediction results

[0144] 3.1 Inputting ARMA forecast results into the STIRPAT model

[0145] After predicting the future output of multiple products, product prices, raw material prices, energy consumption levels and other related influencing factors through the ARMA model, these prediction results are used as input variables of the STIRPAT model. Since the STIRPAT model can handle the interaction between multiple influencing factors, the change trend of the company's future power load can be deduced through this step.

[0146] 3.2 Derivation of medium- and long-term power load curve

[0147] Combining the prediction results of the STIRPAT model and the ARMA model, the medium- and long-term power load curve of the enterprise is deduced. The specific process is as follows:

[0148] Prediction of future economic variables: Based on the ARMA model, the output of various products, product prices, raw material prices, energy consumption levels and other data are predicted year by year in the future to obtain the variable values ​​of the influencing factors in the future time period.

[0149] Input the year-by-year forecast variables into the STIRPAT model: Input the year-by-year forecast variables obtained by ARMA forecast into the fitted STIRPAT model to calculate the power load in the corresponding time period.

[0150] Generation of power load curve: Through the above steps, a medium- and long-term change curve of the enterprise power load in the future is generated. This curve can better reflect the impact of changes in economic variables on power load.

[0151] Example 2

[0152] Considering the actual data of a certain enterprise from 2011 to 2021, the IPAT model and the STIRPAT model are used to predict the electricity consumption. Figure 2As shown in Figure 1, the forecast error of the STIRPAT model is generally smaller and closer to the actual electricity consumption. For example, in 2011, the forecast error of the STIRPAT model was only -0.46%, while the forecast error of the IPAT model reached 6.75%. Throughout the time series, the error of the STIRPAT model never exceeded 5%, while the error of the IPAT model remained around 25%.

[0153] The reason for choosing the STIRPAT model is its higher prediction accuracy and flexibility. As a multi-factor random impact model, the STIRPAT model is able to explore the impact of changes in multiple factors on electricity consumption. Compared with the IPAT model, the STIRPAT model not only focuses on the multiplicative effect of factors, but also considers the interaction and nonlinear relationship between factors, providing a more complex and detailed analysis framework. The advantage of the STIRPAT model is that it adapts to different statistical analyses and hypothesis tests, making the model more flexible and adaptable. It is able to include multiple explanatory variables and allow interactions between variables, thereby more accurately capturing the complex factors that affect power load. In addition, through the random error term, the STIRPAT model is able to consider other influencing factors that are not directly modeled, further improving the accuracy of the prediction.

[0154] Example 3

[0155] Make electricity consumption forecast based on the specific data of another company.

[0156] This example selects the prices and output of various types of products among the key influencing factors as multi-dimensional factors for STIRPAT modeling.

[0157] It is known that the company's products include nine categories, and considering the price changes of the products (using the average futures price). The following formula is the SPIRPAT model fitted by the company data from 2018 to 2023: ln(E) = -0.009585·ln(Pa)+0.010215·ln(Pb)-0.064285·ln(Pc)

[0158] +0.184061ln(Pd)+0.021976ln(Pe)-0.022901ln(Pf)

[0159] +0.064891ln(Pg)-0.070493ln(Ph)+0.050975ln(Pi)

[0160] -0.050511ln(Pr)+0.0009101ln(1.258)

[0161] The ARMA method is used to perform time series forecasting of each product category for each product output and product price. Figure 3 and Figure 4 The forecast results of product price and output of product 6 are shown respectively. It can be seen that the ARMA technology can effectively reflect the cyclical changes of products.

[0162] By substituting the predicted product price and product output data for 2024-2030 into the obtained STIRPAT model, we can get the electricity forecast for 2024-2030, as follows: Figure 5 shown.

Claims

1. The enterprise power load forecasting method based on the multi-influencing factor STIRPAT model is characterized by: The following steps are involved: S1. Analysis of factors affecting electricity consumption: Collect relevant historical data on the electricity consumption of enterprises, conduct statistical analysis on it, use correlation analysis to screen key factors that have significant correlation with the electricity consumption of enterprises, and select several of them as multi-dimensional factors for STIRPAT modeling; S2. STIRPAT model construction: Based on the collected historical data, the extended STIRPAT model is used to establish the relationship model between various influencing factors and enterprise power load; S3, ARMA forecast: Use the ARMA model to perform time series forecasting on the data of various influencing factors in the future to obtain the variable values ​​of various influencing factors in the future time period; S4. Input the variable values ​​of each influencing factor into the STIRPAT model and calculate the power load in the corresponding time period.

2. The enterprise power load forecasting method according to claim 1, characterized in that: The enterprise power load forecasting method further comprises the following steps: S4. Generation of power load curve: According to the above steps, the variable values ​​of the influencing factors in the future time period are obtained based on ARMA, and input into the STIRPAT model to generate the medium- and long-term change curve of the enterprise power load in the future time period.

3. The enterprise power load forecasting method according to claim 2, characterized in that: The extended STIRPAT model formula in step S2 is as follows: E=Pa a ×Pb b ×PC c ×...×β α Where E is the charge, Pa, Pb, Pc, ... are the influencing factors, β is the error term, and a, b, c, ..., α are the percentages of charge change caused by a 1% change in the corresponding influencing factor.

4. The enterprise power load forecasting method according to claim 3 is characterized in that: The ARMA model in step S3 is constructed using the following method: S301. Data preparation: Conduct time series analysis on the influencing factor variables that need to be predicted to ensure the stability of the data and handle missing data; S302, model selection: select the appropriate order (p, q) by AIC / BIC criteria, where p represents the number of lag periods and q represents the number of lag error periods; S303, model fitting: fitting the parameters of the ARMA model by maximum likelihood estimation method.

5. The enterprise power load forecasting method according to claim 4, characterized in that: The following method is used to ensure the stability of the data in step S301: ADF test is performed using the following formula: Where t is the time trend term ∈ t is the error term, ΔY t =Y t -Y t-1 is the difference value of the data, p is the key parameter that needs ADF detection, if p is greater than 0.05, it means that the sequence is non-stationary; If the time series is non-stationary, use the second-order difference method to transform it into a stationary series.

6. The enterprise power load forecasting method according to claim 5, characterized in that: In step S302, the ARMA model order (p, q) is selected by the following method: The autoregressive term AR(p) models the linear relationship between the current value and the value of the previous p periods, where p represents the number of lag periods: Y t =φ1Y t-1 +φ2Y t-2 +...+φ p Y t-p +∈ t ; The moving average term MA(q) models the linear relationship between the current error and the error of the previous q periods, where q represents the number of lagged error periods: Y t = t +θ 1t-1 +θ 2t-2 +...+θ q t-q The best model order is selected by AIC and BIC criteria: Where k is the number of parameters in the model, n is the sample size, is the maximum likelihood estimate.

Images