Social, industrial and enterprise electricity consumption analysis method based on DDPM and GRU

Abstract

The invention discloses a method for analyzing electricity consumption of society, industry and enterprise based on DDPM and GRU. The method comprises the following steps: collecting electricity consumption data of society, industry and enterprise; constructing a multi-dimensional social development characteristic index system covering economic thermometer indexes, per capita power consumption, power consumption elastic coefficients and the like; performing standardization, forward diffusion and reverse de-noising processing on the industry power consumption data by using a DDPM algorithm, predicting the yield of a terminal product by combining the upstream and downstream industry chain power consumption data, eliminating noise interference and capturing dynamic relevance; and a GRU model is adopted to carry out time sequence modeling on the power consumption data and the non-power indexes of the enterprise. According to the method, the advantages of DDPM and GRU are fused, the accuracy and robustness of power consumption analysis are remarkably improved, scientific data support is provided for policy making, industrial planning and enterprise green transformation, and energy management optimization, carbon emission reduction and green energy utilization efficiency improvement are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis and machine learning, and particularly to a method for analyzing electricity consumption of society, industries, and enterprises based on DDPM and GRU. Background Art

[0002] Analysis of the electricity consumption characteristics of society-industry-enterprise based on electricity consumption data can not only improve the overall operation efficiency of the power system, but also promote the green and low-carbon development of industrial enterprises, provide scientific decision-making support for the government, industries, and enterprises, and achieve the coordinated unity of economic benefits and environmental protection goals. The existing methods have the following deficiencies: 1) Lack of a comprehensive electricity consumption characteristic index system for social and economic development: At present, social and economic development mainly relies on traditional indicators such as GDP and industrial output value, and fails to make full use of electricity consumption data as a direct signal to measure economic vitality and industrial structure changes. Existing analyses mostly stay at the level of electricity quantity statistics, lacking comprehensive characteristic indicators of social and economic development based on electricity consumption data, and it is difficult to accurately reflect changes such as economic cycles and industrial transformations. 2) Low accuracy of industry output prediction and failure to consider the influence of the industrial chain: Traditional industry electricity consumption analysis only focuses on the load characteristics of the industry itself, without combining the electricity consumption patterns of the upstream and downstream industrial chains, resulting in low accuracy of terminal product output prediction. Existing machine learning methods have limitations in dealing with the conduction effect of the industrial chain and are difficult to capture the in-depth impact of power data on the production link. 3) Insufficient evaluation method for the adjustment ability of enterprise non-electricity indicators: At present, the analysis of the load adjustment ability of enterprises mainly focuses on the power load level, while the actual adjustment ability of enterprises is often closely related to non-electricity indicators such as production plans, equipment operation status, and supply chain conditions. Existing methods lack the joint modeling of non-electricity indicators (such as inventory levels, order quantities, equipment status, etc.) and electricity consumption behaviors, and it is difficult to accurately predict the load adjustment ability of enterprises.

[0003] Therefore, there is an urgent need for an electricity consumption characteristic analysis method to improve the accuracy of social and economic development analysis, optimize industrial electricity consumption prediction, and enhance the analysis of enterprise load adjustment ability, so as to provide more scientific decision-making support for the government, power grid enterprises, and industry users. Summary of the Invention

[0004] The purpose of the present invention is to provide a social, industrial and corporate electricity consumption analysis method based on DDPM and GRU, so as to solve the limitations of the existing technology in that the feature system is single, industry prediction ignores the correlation of the industrial chain, and the evaluation of corporate non-electricity indicators is insufficient. By integrating the noise robust modeling of DDPM with the time series dynamic analysis of GRU, the accuracy of electricity consumption analysis is significantly improved: a multi-dimensional feature system (such as economic thermometer, electricity consumption elasticity coefficient) is constructed to comprehensively reflect social development; DDPM is used to explore the correlation of electricity consumption in the industrial chain and accurately predict terminal output; combined with GRU to quantify the impact of corporate non-electricity indicators (supply chain, equipment status) on low-carbon regulation, green electricity utilization and emission reduction capabilities, providing data-driven support for policy formulation, industrial planning and corporate green transformation, and promoting the synergistic effect of energy optimization and carbon emission reduction.

[0005] In order to achieve the above object, the present invention is implemented through the following technical solutions:

[0006] The social, industry, and enterprise electricity consumption analysis methods based on DDPM and GRU include:

[0007] Step 1: Data Collection

[0008] Collect electricity consumption data from society, industry, and enterprises, including social electricity consumption data, industry electricity consumption data, and enterprise electricity consumption data. Enterprise electricity consumption data includes but is not limited to total enterprise electricity consumption, enterprise green power consumption, coal-fired power consumption, electricity load of different industries, carbon emissions, sulfur dioxide emissions, nitrogen oxide emissions, government carbon trading subsidies, enterprise production scale, and enterprise production hours;

[0009] Step 2: Construct a social development characteristic indicator system

[0010] Based on the electricity consumption data, a social development characteristic indicator system is constructed, including economic thermometer indicators, per capita electricity consumption indicators, the proportion of residential air-conditioning electricity consumption, electricity consumption elasticity coefficient and scientific and technological development indicators;

[0011] Step 3: DDPM algorithm predicts the output of end products

[0012] Based on industry electricity consumption data and electricity consumption data from upstream and downstream industrial chains, the DDPM algorithm is used to predict the output of end products. This includes: standardizing industry electricity consumption data, generating diffusion data through a forward diffusion process, recovering the standardized output forecast value through a reverse denoising process, and recovering the actual output through a denormalization process.

[0013] Step 4: GRU algorithm predicts enterprise regulation capabilities

[0014] Based on the enterprise's electricity consumption data and non-electricity indicators, the GRU algorithm is used to predict the enterprise's non-electricity indicator regulation capability, including low-carbon regulation capability, green electricity regulation capability and pollutant emission regulation capability.

[0015] As a preferred embodiment of the present invention, collecting the electricity consumption data of society, industry, and enterprises includes the following steps:

[0016] The expression formula for the electricity consumption data of society is:

[0017] X i (k) = [E t , E green,t , E cold,t , E industry,t (1)

[0018] The expression formula for the electricity consumption data of the industry is:

[0019] X i (l) = [CO 2,t , SO 2,t , NO x,t (2)

[0020] The expression formula for the electricity consumption data of the enterprise is:

[0021] X i (m) = [R t , S t , T t (3)

[0022] Where: X i (k), X i (l), X i (m) respectively represent the electricity consumption data of society, the electricity consumption data of the industry, and the electricity consumption data of the enterprise at time t; E t is the total electricity consumption of the enterprise, E green,t is the green electricity consumption of the enterprise, E cold,t is the coal-fired electricity consumption, E industry,t is the electricity load of different industries; CO 2,t is the carbon emission of the enterprise, SO 2,t is the sulfur dioxide emission, NO x,t is the nitrogen oxide emission; R t is the government carbon trading subsidy, S t is the production scale of the enterprise, T t is the production duration of the enterprise.

[0023] As a preferred embodiment of the present invention, constructing the social development characteristic index system includes the following indicators:

[0024] The expression formula for the Economic Thermometer Index (ETI) is:

[0025]

[0026] The expression formula for the Per Capita Electricity Consumption Index (PEC) is:

[0027]

[0028] The expression of the proportion of residential air - conditioner electricity consumption ACR is:

[0029]

[0030] The expression of the elasticity coefficient of electricity consumption ECE is:

[0031]

[0032] The expression of the technology development index TDI is:

[0033]

[0034] Among them: ΔGDP is the increment of the gross domestic product GDP (unit: yuan), ΔE is the increment of the total electricity consumption of the whole society during the same period (unit: kWh); E total is the total electricity consumption (unit: kWh), including industrial, commercial, residential and agricultural electricity consumption, P is the total population (unit: person); E AC is the electricity consumption of residential air - conditioners (cooling + heating) (unit: kWh), E res is the total residential electricity consumption (unit: kWh); E DC is the electricity consumption of the data center (unit: kWh), representing the energy consumption of the computing center, cloud computing, big data, and AI industries.

[0035] As a preferred embodiment of the present invention, the standardization process of the industry electricity consumption data includes the following steps:

[0036] The historical output of end - products Y0 is standardized by the following formula:

[0037]

[0038] The electricity consumption of the upstream industry E up is standardized by the following formula:

[0039]

[0040] The electricity consumption of the mid - stream industry E mid is standardized by the following formula:

[0041]

[0042] The electricity consumption of the downstream industry E down is standardized by the following formula:

[0043]

[0044] Where: Y hist is the historical output of end - products, μ Y , σ Y are the mean and standard deviation of the historical output of end - products; E up,raw is the data to be standardized for the electricity consumption of the upstream industry, are the mean and standard deviation of the electricity consumption of the upstream industry; E mid,raw is the data to be standardized for the electricity consumption of the mid - stream industry, are the mean and standard deviation of the electricity consumption of the mid - stream industry; E down,raw is the data to be standardized for the electricity consumption of the downstream industry, are the mean and standard deviation of the electricity consumption of the downstream industry.

[0045] As a preferred embodiment of the present invention, the generation of diffusion data through the forward diffusion process includes the following steps:

[0046] Gradually add Gaussian noise to the standardized historical output of end - products Y0, and the diffusion formula is:

[0047]

[0048] Accumulated noise parameter The expression is:

[0049]

[0050] The diffused product output Y t The expression is:

[0051]

[0052] Where: β t is the diffusion noise coefficient, which increases with time; I is the unit diagonal matrix, representing independent noise terms, ∈t is the random Gaussian noise (used to simulate uncertainty), q is the product output diffusion function, Y t-1 is the diffused product output at time t - 1, N is the Gaussian noise function, is the accumulated noise calculation model, which is calculated through the diffusion noise coefficient, is the basis of the accumulated noise parameter, is the accumulated noise error term

[0053] As a preferred embodiment of the present invention, the reverse denoising process includes the following steps:

[0054] Define the denoised output distribution through the following formula:

[0055] p θ (Y t-1 |Yt ) = N(Y t-1 ; μ θ (Y t , t), σ θ (t) 2 I) (16)

[0056] The expression is the mean of the denoised output:

[0057]

[0058] Predict the noise term through the following formula:

[0059]

[0060] Generate the predicted value of the denoised output through the following formula:

[0061] Y t-1 = μ θ (Y t , t) + σ t z t , z t ~N(0, I) (19)

[0062] In the forward diffusion process, the expression of the diffused product output is:

[0063]

[0064] Define the diffusion coefficient through the following formula:

[0065] α t = 1 - β t (21)

[0066] The expression is the cumulative diffusion coefficient:

[0067]

[0068] Define the standard deviation of the denoising step size through the following formula:

[0069]

[0070] Where: μ θ (Y t , t) is the mean of the denoised output, g θ is the noise prediction network constructed based on Transformer / LSTM; p θ (Y t-1 |Y t ) is the probability distribution of calculating the denoising result of the (t - 1)th round after t rounds of denoising, which follows a Gaussian distribution with a mean of μ θ (Y t, t), with a standard deviation of, E up , E mid , E down are the normalized electricity consumption of the upstream, midstream, and downstream industries respectively; M is the market factor data, including supply and demand relationship, policy changes; β t is the noise intensity parameter at the t-th step in the forward diffusion process; is the cumulative diffusion coefficient, reflecting the total noise accumulation of forward diffusion; z t is the standard normal distribution noise, used to prevent the model from overfitting.

[0071] As a preferred embodiment of the present invention, the restoration of the true output through inverse normalization processing includes the following steps:

[0072] The expression is the denoised normalized predicted value:

[0073]

[0074] The true output is restored through the following formula:

[0075]

[0076] The expression is the original value of the normalized data:

[0077]

[0078] Where: is the predicted normalized output of the end product, calculated during the denoising process, and is the predicted value under the normalized scale; Y T is the noise state at the final step of the diffusion process, which is a Gaussian noise sample generated at the last time step T of the forward diffusion process; is the final predicted output of the end product; T is the total number of steps in the diffusion process, indicating the number of denoising steps required to regress from Gaussian noise to real data; is the noise term predicted by the denoising network; is the final predicted output of the end product, which is restored to the original scale after the denoising process and inverse normalization; σ Y is the standard deviation of the historical end product output; μ Y is the mean of the historical end product output.

[0079] As a preferred embodiment of the present invention, the training loss of the DDPM algorithm is calculated by minimizing the noise prediction error, and the calculation formula is as follows:

[0080]

[0081] Where: Calculate the mathematical expectations of the standardized output of the end product, the diffusion steps, and the noise term to ensure that the loss function is applicable to the entire training data distribution; ε is the true noise ε θ (Y t , t) is the noise predicted by the model; Y t is the state of the product output at the t-th step during the diffusion process.

[0082] As a preferred embodiment of the present invention, predicting the non-electric index adjustment ability of an enterprise using the GRU algorithm includes the following steps:

[0083] Standardize the enterprise's electricity consumption data and non-electric indexes through the following formula:

[0084]

[0085] Update the hidden state through the following formula to capture the dynamic evolution of the enterprise's adjustment ability:

[0086] h t = GRU(W h X t + U h h t-1 + b h ) (29)

[0087] The expression is the enterprise's ability to reduce carbon emissions:

[0088] C t = W c h t + b c (30)

[0089] The expression is the enterprise's ability to increase the proportion of green electricity used:

[0090] G t = W g C t + b g (31)

[0091] Calculate the enterprise's ability to reduce pollutant emissions through the formula:

[0092] P t = W p G t + b p (32)

[0093] Restore the original scale of the final predicted value through the following formula:

[0094]

[0095] where: X t is the input variable, μX is the historical mean, σ X is the standard deviation, h t is the hidden state representing the time evolution of the enterprise's low-carbon regulation capability, GRU(·) is the GRU calculation unit, and W h 、U h , b h is the weight and bias of GRU, C t is the predicted value of the enterprise’s low-carbon regulation capability, W c , b c is the mapping weight and bias of low carbon regulation capability, G t is the green electricity regulation capacity, that is, the ability of enterprises to increase the proportion of green electricity, W g , b g is the linear regression weight, G t As input, ensuring the consistency of the forecasting process, is the final predicted value after denormalization (C t , G t , P t ), σ Y 、μ Y is the historical mean and standard deviation;

[0096] The non-electrical indicators include supply chain status, equipment operating status, and inventory levels.

[0097] Compared with the existing technology, the beneficial effects of the present invention are as follows: by combining the denoising diffusion probability model (DDPM) and the gated recurrent unit (GRU) algorithm, the present invention can effectively integrate multi-source electricity consumption data and non-electrical indicators of society, industry and enterprises, and significantly improve the accuracy and robustness of electricity consumption analysis and prediction. First, based on DDPM, the industry electricity consumption data is standardized, diffused and reversely denoised, which can effectively eliminate data noise interference and capture the dynamic correlation between the upstream and downstream of the industrial chain, thereby accurately predicting the output of terminal products and solving the problem of insufficient modeling of complex nonlinear relationships by traditional methods. Secondly, by constructing a social development characteristic indicator system covering multiple dimensions such as economic thermometer, per capita electricity consumption, and electricity consumption elasticity coefficient, it can comprehensively reflect the comprehensive status of social and economic development and provide more refined data support for policy making. In addition, the GRU model is used to perform time-series modeling of corporate electricity consumption data and non-electricity indicators, which can dynamically capture the evolution of the company's low-carbon regulation capabilities, green electricity usage capabilities, and pollutant emission regulation capabilities. Combined with denormalization processing, the true scale is restored, and the adaptability to non-electricity indicators (such as supply chain status and equipment operating status) is enhanced. Ultimately, the precise optimization of the company's energy management strategy can be achieved, reducing carbon emissions and pollutant emissions, and improving the efficiency of green energy utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings. Among them:

[0099] Figure 1 It is a schematic flow chart of the method of the present invention. Specific embodiments

[0100] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions of the embodiments of the present invention with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the described embodiments of the present invention fall within the scope of protection of the present invention.

[0101] As Figure 1 shown, it is an embodiment of the present invention. This embodiment provides a method for analyzing the electricity consumption of society, industries, and enterprises based on DDPM and GRU, including:

[0102] Step 1: Collect electricity consumption data of society, industries, and enterprises, including social electricity consumption data, industrial electricity consumption data, and enterprise electricity consumption data. Among them, the enterprise electricity consumption data includes, but is not limited to, the total electricity consumption of the enterprise, the consumption of green electricity by the enterprise, the consumption of coal-fired electricity, the electricity load of different industries, carbon emissions, sulfur dioxide emissions, nitrogen oxide emissions, government carbon trading subsidies, the production scale of the enterprise, and the production duration of the enterprise;

[0103] Among them, the collection of electricity consumption data of society, industries, and enterprises includes the following steps:

[0104] The expression of social electricity consumption data is:

[0105]

[0106] The expression of industrial electricity consumption data is:

[0107]

[0108] The expression of enterprise electricity consumption data is:

[0109] Among them: X i (k), X i (l), X i (m) respectively represent the social electricity consumption data, industrial electricity consumption data, and enterprise electricity consumption data at time t; E t is the total electricity consumption of the enterprise, Egreen,t is the enterprise's green electricity consumption, E cold,t is the coal-fired electricity consumption, E industry,t is the electricity load of different industries; CO 2,t is the enterprise's carbon emissions, SO 2,t is the sulfur dioxide emissions, NO x,t is the nitrogen oxide emissions; R t is the government carbon trading subsidy, S t is the enterprise's production scale, T t is the enterprise's production duration.

[0110] In this embodiment, the purpose of data collection: to collect the basic data of social, industrial and enterprise electricity consumption, and provide raw materials for subsequent analysis.

[0111] Collect social electricity consumption (such as residential and commercial electricity consumption), industrial electricity consumption (such as manufacturing and service industry electricity consumption), and enterprise electricity consumption (such as total electricity consumption, green electricity usage, coal consumption, etc.).

[0112] At the same time, collect the enterprise's environmental protection data, such as carbon emissions, pollutant emissions (sulfur dioxide, nitrogen oxides), as well as information such as government subsidies and production scale. This example needs to first collect all the "raw material data" related to electricity consumption.

[0113] Furthermore, step 2: construct a social development characteristic index system based on the electricity consumption data, including an economic thermometer index, a per capita electricity consumption index, the proportion of residential air-conditioning electricity consumption, the electricity consumption elasticity coefficient, and a science and technology development index;

[0114] Specifically, the construction of the social development characteristic index system includes the following indexes:

[0115] The expression of the economic thermometer index ETI is:

[0116]

[0117] The expression of the per capita electricity consumption index PEC is:

[0118]

[0119] The expression of the proportion of residential air-conditioning electricity consumption ACR is:

[0120]

[0121] The expression of the electricity consumption elasticity coefficient ECE is:

[0122]

[0123] The expression of the science and technology development index TDI is:

[0124]

[0125] Where: ΔGDP is the increment of GDP (Gross Domestic Product) (unit: yuan), and ΔE is the increment of the total electricity consumption of the whole society during the same period (unit: kWh); E total is the total electricity consumption (unit: kWh), including industrial, commercial, residential, and agricultural electricity consumption, and P is the total population (unit: person); E AC is the electricity consumption of residential air conditioners (cooling + heating) (unit: kWh), and E res is the total residential electricity consumption (unit: kWh); E DC is the electricity consumption of the data center (unit: kWh), representing the energy consumption of the computing center, cloud computing, big data, and AI industries.

[0126] In this example, the purpose of constructing the social development index system is to convert the collected data into indicators that can reflect the social development status.

[0127] Calculate the economic thermometer index (the relationship between electricity growth and GDP growth), per capita electricity consumption (total electricity consumption divided by the population), the proportion of residential air conditioner electricity consumption (the proportion of air conditioner power consumption in total residential electricity consumption), the electricity consumption elasticity coefficient (the ratio of electricity consumption growth rate to economic growth rate), the science and technology development index (such as the electricity consumption of the data center), etc.

[0128] In this embodiment, the scattered data is organized into several "scorecards". For example, the living standard can be seen through per capita electricity consumption, and the climate impact can be understood through the proportion of air conditioner electricity consumption.

[0129] Furthermore, step 3: Based on the industry electricity consumption data and the electricity consumption data of the upstream and downstream industrial chains, use the DDPM algorithm to predict the output of the end products, including: standardizing the industry electricity consumption data, generating diffusion data through the forward diffusion process, restoring the standardized output prediction value based on the reverse denoising process, and restoring the real output through anti-standardization processing;

[0130] The standardization processing of the industry electricity consumption data includes the following steps:

[0131] The historical output Y0 of the end product is standardized by the following formula:

[0132]

[0133] The electricity consumption E of the upstream industry up is standardized by the following formula:

[0134]

[0135] The electricity consumption E of the midstream industry mid is standardized by the following formula:

[0136]

[0137] Electricity consumption E of downstream industries down Standardized by the following formula:

[0138]

[0139] Where: Y hist is the historical output of end products, μ Y , σ Y are the mean and standard deviation of the historical output of end products; E up,raw is the data of electricity consumption of upstream industries to be standardized, are the mean and standard deviation of the electricity consumption of upstream industries; E mid,raw is the data of electricity consumption of midstream industries to be standardized, are the mean and standard deviation of the electricity consumption of midstream industries; E down,raw is the data of electricity consumption of downstream industries to be standardized, are the mean and standard deviation of the electricity consumption of downstream industries.

[0140] Specifically, the generation of diffusion data through the forward diffusion process includes the following steps:

[0141] Gradually add Gaussian noise to the standardized historical output Y0 of end products. The diffusion formula is:

[0142]

[0143] Accumulated noise parameter The expression is:

[0144]

[0145] Diffused product output Y t The expression is:

[0146]

[0147] Where: β t is the diffusion noise coefficient, increasing with time; I is the unit diagonal matrix, representing independent noise terms, ∈t is the random Gaussian noise (used to simulate uncertainty), q is the product output diffusion function, Y t-1 is the diffused product output at time t-1, N is the Gaussian noise function, is the accumulated noise calculation model, calculated through the diffusion noise coefficient, is the basis of the accumulated noise parameter, is the accumulated noise error term

[0148] Specifically, the reverse denoising process includes the following steps:

[0149] Define the denoised production distribution through the following formula:

[0150] p θ (Y t-1 |Y t ) = N(Y t-1 ; μ θ (Y t , t), σ θ (t) 2 I) (16)

[0151] The expression is the mean of the denoised production:

[0152]

[0153] Predict the noise term through the following formula:

[0154]

[0155] Generate the predicted value of the denoised production through the following formula:

[0156] Y t-1 = μ θ (Y t , t)+σ t z t , z t ~N(0, I) (19)

[0157] In the forward diffusion process, the expression of the diffused product production is:

[0158]

[0159] Define the diffusion coefficient through the following formula:

[0160] α t = 1 - β t (21)

[0161] The expression is the cumulative diffusion coefficient:

[0162]

[0163] Define the standard deviation of the denoising step size through the following formula:

[0164]

[0165] Where: μ θ (Y t , t) is the mean of the denoised production, gθ is a noise prediction network built based on Transformer / LSTM; p θ (Y t-1 |Y t ) is the probability distribution of the denoising result at the (t - 1)-th round calculated after denoising at the t-th round, which follows a Gaussian distribution with a mean of μ θ (Y t ,t), and a standard deviation of, E up ,E mid ,E down are the electricity consumption of the upstream, midstream, and downstream industries after standardization respectively; M is the market factor data, including supply-demand relationship and policy changes; β t is the noise intensity parameter at the t-th step in the forward diffusion process; is the cumulative diffusion coefficient, reflecting the total noise accumulation of the forward diffusion; zt is the standard normal distribution noise, used to prevent the model from overfitting.

[0166] Specifically, the restoration of the true output through inverse standardization processing includes the following steps:

[0167] The expression is the denoised standardized predicted value:

[0168]

[0169] The true output is restored through the following formula:

[0170]

[0171] The expression is the original value of the standardized data:

[0172]

[0173] Where: is the predicted standardized terminal product output, calculated during the denoising process, and is the predicted value under the standardized scale; Y T is the noise state at the final step of the diffusion process, which is a Gaussian noise sample generated at the last time step T in the forward diffusion process; is the finally predicted terminal product output; T is the total number of steps in the diffusion process, indicating the number of denoising steps required to regress from Gaussian noise to real data; is the noise term predicted by the denoising network; is the finally predicted terminal product output, which is restored to the original scale after the denoising process and inverse normalization; σ Y is the standard deviation of the historical terminal product output; μ Y is the mean of the historical terminal product output.

[0174] In this example, the training loss of the DDPM algorithm is calculated by minimizing the noise prediction error, and the formula is as follows:

[0175]

[0176] where: E Y0,t,ε Calculate the mathematical expectation of the standardized end-product output, diffusion steps, and noise terms to ensure that the loss function can be applied to the entire training data distribution; ε is the true noise ε θ (Y t ,t) is the noise predicted by the model; Y t is the product output state at the t-th step in the diffusion process.

[0177] In this embodiment, the industry electricity consumption data is used to predict the output of future end-products (such as automobiles and household appliances).

[0178] Implementation:

[0179] Standardization processing: Uniformly scale the electricity consumption data of different industries to the same range (for example, with a mean of 0 and a standard deviation of 1) to facilitate model processing.

[0180] Forward diffusion: Gradually add random noise to the data (simulating the uncertainty in reality) to generate "fuzzy data" with noise.

[0181] Reverse denoising: Use a model (such as Transformer or LSTM) to gradually remove the noise and restore the "clean" predicted output.

[0182] Inverse standardization: Convert the prediction result back to the original data range to obtain the final true output value.

[0183] Just like deliberately blurring a photo first (adding noise), then training the AI to restore the clear image step by step, and finally obtaining an accurate prediction result.

[0184] Furthermore, the adoption of the GRU algorithm to predict the non-electric index adjustment ability of enterprises includes the following steps:

[0185] Standardize the enterprise electricity consumption data and non-electric indexes through the following formula:

[0186]

[0187] Update the hidden state through the following formula to capture the dynamic evolution of the enterprise adjustment ability:

[0188] h t = GRU(W h X t + U h h t-1 + b h) (29)

[0189] Expression for the enterprise's ability to reduce carbon emissions:

[0190] C t =W c h t +b c (30)

[0191] Expression for the enterprise's ability to increase the proportion of green electricity used:

[0192] G t =W g C t +b g (31)

[0193] Calculate the enterprise's ability to reduce pollutant emissions through the following formula:

[0194] P t =W p G t +b p (32)

[0195] Restore the original scale of the final predicted value through the following formula:

[0196]

[0197] Where: X t is the input variable, μ X is the historical mean, σ X is the standard deviation, h t is the time evolution of the hidden state representing the enterprise's low-carbon adjustment ability, GRU(·) is the GRU calculation unit, W h 、U h 、b h are the weights and biases of the GRU, C t is the predicted value of the enterprise's low-carbon adjustment ability, W c 、b c are the mapping weights and biases of the low-carbon adjustment ability, G t is the green electricity adjustment ability, that is, the enterprise's ability to increase the proportion of green electricity, W g 、b g are the linear regression weights, G t is used as the input to ensure the coherence of the prediction process, is the final predicted value after anti-normalization (C t ,G t ,P t ), σ Y 、μ Y are the historical mean and standard deviation;

[0198] The non - electrical indicators include supply chain status, equipment operation status, and inventory level.

[0199] In this embodiment, analyze the adjustment ability of the enterprise in terms of environmental protection, such as whether it can reduce carbon emissions and increase the use of green electricity.

[0200] Standardized data: Uniformly scale the electricity consumption and non - electrical indicators (such as production scale, subsidies) of the enterprise.

[0201] GRU model training: Use the gated recurrent unit (GRU) to analyze time - series data and capture the dynamic changes in the enterprise's adjustment ability.

[0202] Prediction ability: Output the predicted values of low - carbon adjustment ability (reduce carbon emissions), green electricity adjustment ability (increase the use of green electricity), and pollutant adjustment ability (reduce pollutant emissions).

[0203] Restore the original scale: Convert the prediction results back to the actual units for the enterprise to refer to.

[0204] In this embodiment, use AI to learn the past behavior patterns of the enterprise and predict whether it can be more environmentally friendly in the future, such as "can it use more solar energy and less coal next year".

[0205] In summary, the present invention innovatively integrates the denoising diffusion probability model (DDPM) with the gated recurrent unit (GRU) to construct a multi - level and multi - dimensional social - industry - enterprise electricity consumption analysis framework. This method combines the noise - robust modeling ability of DDPM with the time - series dynamic capture advantage of GRU for the first time, achieving accurate analysis and prediction of electricity consumption data. By constructing a comprehensive characteristic index system covering economic thermometers, electricity consumption elasticity coefficients, etc., it solves the limitation of traditional methods relying on a single economic indicator; uses DDPM to perform diffusion - denoising modeling on the electricity consumption data of the upstream and downstream of the industrial chain, effectively eliminating noise interference and revealing complex non - linear correlations, significantly improving the accuracy of terminal product output prediction; combines GRU for time - series dynamic modeling of the enterprise's non - electrical indicators, realizing the quantitative evaluation of key indicators such as low - carbon adjustment ability and green electricity utilization efficiency, and providing data - driven decision - making support for enterprise energy optimization and emission reduction paths. The technical solution of the present invention not only fills the gaps in the existing technology in industrial chain conduction effect modeling and non - electrical indicator integration analysis, but also provides scientific and refined technical tools for the government to formulate carbon emission reduction policies, for the industry to plan production capacity layouts, and for enterprises to implement green transformation, with outstanding practicality and broad social and economic benefits.

[0206] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0207] Any process or method description shown in a flowchart or described in other ways herein can be understood as representing a module, segment, or portion of code including one or more executable instructions for implementing a specific logical function or process. And the scope of the preferred embodiments of the present application includes additional implementations, where the functions may be performed in a substantially simultaneous manner or in a reverse order according to the functions involved, without being performed in the order shown or discussed.

[0208] As described above, this is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of various changes or substitutions, and these should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A method for analyzing electricity consumption of society, industries and enterprises based on DDPM and GRU, characterized in that, The method includes: Collecting social, industrial, and enterprise electricity consumption data, including social electricity consumption data, industrial electricity consumption data, and enterprise electricity consumption data. Among them, enterprise electricity consumption data includes but is not limited to total enterprise electricity consumption, enterprise green electricity consumption, coal-fired electricity consumption, electricity load of different industries, carbon emissions, sulfur dioxide emissions, nitrogen oxide emissions, government carbon trading subsidies, enterprise production scale, and enterprise production duration; Constructing a social development characteristic index system based on the electricity consumption data, including an economic thermometer index, per capita electricity consumption index, proportion of residential air-conditioning electricity consumption, electricity consumption elasticity coefficient, and science and technology development index; Predicting the output of end products using the DDPM algorithm based on industrial electricity consumption data and upstream and downstream industrial chain electricity consumption data, including: standardizing the industrial electricity consumption data, generating diffusion data through the forward diffusion process, restoring the standardized output prediction value based on the reverse denoising process, and restoring the real output through anti-standardization processing; Predicting the non-electric index adjustment ability of enterprises using the GRU algorithm based on enterprise electricity consumption data and non-electric indexes, including low-carbon adjustment ability, green electricity adjustment ability, and pollutant emission adjustment ability.

2. The method for analyzing electricity consumption of society, industry and enterprises based on DDPM and GRU according to claim 1, characterized in that, The collection of social, industrial, and enterprise electricity consumption data includes the following steps: The expression of social electricity consumption data is: X i (k) = [E t , E green,t , E cold,t , E industry,t (1) The expression of industrial electricity consumption data is: X i (l) = [CO 2,t , SO 2,t , NO x,t (2) The expression of enterprise electricity consumption data is: X i (m) = [R t , S t , T t (3) Among them: X i (k), X i (l), X i (m) respectively represent the social electricity consumption data, industrial electricity consumption data, and enterprise electricity consumption data at time t; E t is the total electricity consumption of the enterprise, E green,t is the green electricity consumption of the enterprise, E cold,t is the coal-fired electricity consumption, E industry,t is the electricity load of different industries; CO 2,t is the carbon emissions of the enterprise, SO 2,t is the sulfur dioxide emissions, NO x,t is the nitrogen oxide emissions; R t is the government carbon trading subsidy, S t is the production scale of the enterprise, T t is the production duration of the enterprise.

3. The method for analyzing electricity consumption of society, industry and enterprises based on DDPM and GRU according to claim 1, characterized in that, The construction of the social development characteristic index system includes the following indexes: The expression of the economic thermometer index ETI is: The expression of the per capita electricity consumption index PEC is: The expression of the proportion of residential air-conditioning electricity consumption ACR is: The expression of the electricity consumption elasticity coefficient ECE is: The expression of the science and technology development index TDI is: Where: ΔGDP is the increment of GDP (Gross Domestic Product), and ΔE is the increment of the total electricity consumption of the whole society during the same period; E total is the total electricity consumption, including industrial, commercial, residential, and agricultural electricity consumption, P is the total population; E AC is the electricity consumption of residential air conditioners, E res is the total residential electricity consumption; E DC is the electricity consumption of data centers, representing the energy consumption of computing centers, cloud computing, big data, and the AI industry.

4. The method for analyzing electricity consumption of society, industries and enterprises based on DDPM and GRU according to claim 1, wherein The standardization process of the industrial electricity consumption data includes the following steps: The historical output of end products Y0 is standardized by the following formula: Electricity consumption \(E\) of upstream industries up Normalized by the following formula: Midstream industry electricity consumption E mid Normalized by the following formula: Downstream industry electricity consumption E down Normalized by the following formula: Where: Y hist is the historical output of end products, μ Y , σ Y are the mean and standard deviation of the historical output of end products; E up,raw is the data to be standardized for the electricity consumption of the upstream industry, are the mean and standard deviation of the electricity consumption of the upstream industry; E mid,raw is the data to be standardized for the electricity consumption of the midstream industry, are the mean and standard deviation of the electricity consumption of the midstream industry; E down,raw is the data to be standardized for the electricity consumption of the downstream industry, are the mean and standard deviation of the electricity consumption of the downstream industry.

5. The method for analyzing electricity consumption of society, industry and enterprises based on DDPM and GRU according to claim 4, characterized in that, The generation of diffusion data through the forward diffusion process includes the following steps: Gradually adding Gaussian noise to the standardized historical output of end products Y0, and the expression is: Cumulative noise parameter The expression is: The output Y of the product after diffusion t The expression is: Where: β t is the diffusion noise coefficient, which increases with time; I is the unit diagonal matrix, representing independent noise terms, ∈t is the random Gaussian noise; q is the product output diffusion function, Y t-1 is the product output after diffusion at time t - 1, N is the Gaussian noise function, is the cumulative noise calculation model, which is calculated through the diffusion noise coefficient, is the cumulative noise parameter reference value, is the cumulative noise error term.

6. The method for analyzing electricity consumption of society, industry and enterprises based on DDPM and GRU according to claim 5, wherein The reverse denoising process includes the following steps: Defining the denoised output distribution by the following formula: p θ (Y t-1 |Y t )=N(Y t-1 ;μ θ (Y t ,t),σ θ (t) 2 I) (16) The expression is the mean of the denoised output: Predicting the noise term by the following formula: ò θ (Y t ,t)=g θ (E up ,E mid ,E down ,M,t) (18) Generating the denoised output prediction value by the following formula: Y t-1 = μ θ (Y t , t) + σ t z t , z t ~ N(0, I) (19) In the forward diffusion process, the expression of the diffused product output is: Defining the diffusion coefficient by the following formula: α t ＝1-β t (21) The expression is the cumulative diffusion coefficient: Defining the standard deviation of the denoising step size by the following formula: Where: μ θ (Y t , t) is the mean of the output after denoising, and g θ is the noise prediction network constructed based on Transformer / LSTM; p θ (Y t-1 |Y t ) is the probability distribution of the denoising result at the (t - 1)-th round calculated after the t-th round of denoising, which follows a Gaussian distribution with a mean of μ θ (Y t , t) and a standard deviation of, E up , E mid , E down are the electricity consumption of the upstream, midstream, and downstream industries after standardization, respectively; M is the market factor data, including supply and demand relationships and policy changes; β t is the noise intensity parameter at the t-th step in the forward diffusion process; is the cumulative diffusion coefficient, reflecting the total noise accumulation of the forward diffusion; z t is the standard normal distribution noise, used to prevent the model from overfitting.

7. The method for analyzing electricity consumption of society, industry and enterprises based on DDPM and GRU according to claim 6, characterized in that The restoration of the real output through anti-standardization processing includes the following steps: The expression is the denoised standardized prediction value: Restoring the real output by the following formula: The expression is the original value of the standardized data: Wherein: is the predicted standardized terminal product output, calculated during the denoising process, and is the predicted value under the standardized scale; Y T is the noise state at the final step of the diffusion process, and is a Gaussian noise sample generated at the last time step T of the forward diffusion process; is the finally predicted terminal product output; T is the total number of steps in the diffusion process, indicating the number of denoising steps required to regress from Gaussian noise to real data; ò θ (Y t , t) is the noise term predicted by the denoising network; is the finally predicted terminal product output, which is restored to the original scale after the denoising process and inverse normalization; σ Y is the standard deviation of the historical terminal product output; μ Y is the mean of the historical terminal product output.

8. The method for analyzing electricity consumption of society, industry and enterprises based on DDPM and GRU according to claim 1, wherein The training loss of the DDPM algorithm is calculated by minimizing the noise prediction error, and the calculation formula is as follows: Wherein: Calculate the mathematical expectations of the standardized output of the end product, the number of diffusion steps, and the noise term to ensure that the loss function can be applied to the entire training data distribution; ε is the true noise ε θ (Y t , t) is the noise predicted by the model; Y t is the output state of the product at the t-th step in the diffusion process.

9. The method for analyzing electricity consumption of society, industry and enterprises based on DDPM and GRU according to claim 1, wherein The prediction of the non-electric index adjustment ability of enterprises using the GRU algorithm includes the following steps: standardizing the enterprise electricity consumption data and non-electric indexes by the following formula: Updating the hidden state by the following formula to capture the dynamic evolution of the enterprise adjustment ability: h t = GRU(W h X t + U h h t-1 + b h ) (29) The expression is the ability of the enterprise to reduce carbon emissions: C t = W c h t + b c (30) The expression is the ability of the enterprise to increase the proportion of green electricity use: G t = W g C t + b g (31) Calculating the ability of the enterprise to reduce pollutant emissions by the formula: P t = W p G t + b p (32) Restore the original scale of the final predicted value using the following formula: Where: X t is the input variable, μ X is the historical mean, σ X is the standard deviation, h t is the hidden state representing the time evolution of the enterprise's low-carbon adjustment ability, GRU(·) is the GRU calculation unit, W h 、U h 、b h are the weights and biases of GRU, C t is the predicted value of the enterprise's low-carbon adjustment ability, W c 、b c are the mapping weights and biases of the low-carbon adjustment ability, G t is the green power adjustment ability, that is, the ability of the enterprise to increase the proportion of green power, W g 、b g are the linear regression weights, G t is used as the input to ensure the coherence of the prediction process, is the final predicted value after inverse standardization (C t ,G t ,P t ), σ Y 、μ Y are the historical mean and standard deviation; The non-electrical indicators include supply chain status, equipment operating status, and inventory level.

Images