Multi-time-scale power utilization dynamic analysis method, device and equipment

Abstract

The invention discloses a multi-time scale electricity consumption dynamic analysis method, device and equipment, and relates to the technical field of power load analysis, and the method comprises the following steps: obtaining a historical daily electricity consumption sequence and a corresponding multi-source covariable; decomposing the historical daily electricity consumption sequence to obtain a trend component, a season component and a residual component; constructing a correction operator, and performing primary correction on the residual component; based on the sub-industry operation capacity, a hysteresis effect is quantized through a kernel function, and a trend component is reconstructed; constructing an impact variable based on the periodic application capacity, and performing secondary correction on the residual component after the primary correction; and fusing prediction results of the first prediction model and the second prediction model to obtain a total electricity consumption prediction value, and mapping the total electricity consumption prediction value into an electricity supply quantity and electricity sale quantity business index. According to the method, accurate quantification of industrial structural changes and collaborative analysis of multi-time scale influence factors are realized, and the precision of medium and short-term electric power quantity prediction and the degree of connection with power grid business are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power load analysis, and more particularly, to a multi-time-scale power consumption dynamic analysis method, device and equipment. BACKGROUND

[0002] Power load forecasting is the core basis for ensuring the safe and stable operation of the power grid, scientific power dispatching and market transaction decision-making. With the large-scale integration of new energy and the rapid development of emerging industries represented by high-end manufacturing and data centers, China's power load structure is undergoing profound changes. The emerging industry has strong growth momentum and unique rules, and its production capacity climbing period has a significant and increasing structural impact on the overall social power consumption, especially the medium and short-term power consumption changes. Traditional forecasting methods often treat the overall social power load as a homogeneous whole and rely on historical total data for trend extrapolation, which makes it difficult to quantify and respond to the new growth pattern caused by the dramatic changes in internal industrial structure, resulting in a serious challenge to the accuracy of the forecast.

[0003] The main defect is that the existing model construction logic is mismatched with the actual dynamics of the current power system. On the one hand, the mainstream forecasting models (whether classic time series models or complex machine learning models) generally lack the ability to model the special development of emerging industries, and cannot describe the unique power consumption climbing rules based on the application capacity and growth cycle, resulting in a prediction bias at the source. On the other hand, the existing methods are insufficient in coordinating the effects of multiple time dimensions, and fail to finely coordinate the effects of key driving factors such as industrial growth (long-term trend), season and holiday (medium-term cycle) and meteorological temperature difference (short-term fluctuation) in a unified framework, making it difficult for the prediction results to have both trend stability and fluctuation sensitivity. In addition, the connection between the prediction output and the actual business of the power grid (such as power supply and sales planning) is often insufficient. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a multi-time-scale power consumption dynamic analysis method, device and equipment to solve the problems of insufficient modeling of industrial structure changes, low prediction accuracy and disconnection with power supply and sales business in traditional methods.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions: The multi-time scale power consumption dynamic analysis method comprises the following steps: obtaining a historical daily power consumption sequence and corresponding multi-source covariants; performing time series decomposition on the historical daily power consumption sequence to obtain a trend component, a seasonal component and a residual component; constructing a modified operator coupled with a holiday and a meteorological effect to modify the residual component; quantifying the hysteresis effect of capacity ramping through a kernel function based on industry-specific operating capacity, and embedding it into the trend component to obtain a reconstructed trend component; constructing an impact variable based on periodic application capacity to modify the residual component; constructing a first prediction model of the total social power consumption based on the reconstructed trend component, the seasonal component and the second modified residual component; constructing a second prediction model of the emerging industry power consumption based on the emerging industry related data selected from the multi-source covariants; and fusing the prediction results of the first prediction model and the second prediction model to obtain the power consumption prediction value and mapping it to the power supply and power sales business indicators.

[0006] In a preferred embodiment, the obtaining of the historical daily power consumption sequence and the corresponding multi-source covariants comprises: the multi-source covariants include holiday, meteorological effect data, industry-specific operating capacity and periodic application capacity; obtaining the historical daily power consumption and aligning it according to a unified daily time reference to form a basic power sequence; obtaining and encoding the holiday state identifier and meteorological effect data corresponding to the time reference; performing spatial aggregation and daily scale sampling on the industry-specific operating capacity data to construct an industry-specific operating capacity sequence; and constructing a periodic application capacity sequence aligned with the time reference based on the periodic application capacity.

[0007] In a preferred embodiment, the time series decomposition of the historical daily power consumption sequence to obtain the trend component, the seasonal component and the residual component comprises: decomposing the basic power sequence to obtain an initial trend component, a seasonal component and an initial residual component; based on the industry-specific operating capacity sequence, testing the stationarity of the initial trend component to identify the trend mutation period caused by industrial structure adjustment; in the trend mutation period, performing robust decomposition to separate the incremental trend component dominated by the capacity ramping effect; fusing the incremental trend component with the initial trend component to form a final trend component; and removing the final trend component and the seasonal component from the basic power sequence to obtain the residual component to be modified.

[0008] In a preferred embodiment, the correction operator coupling the holiday and meteorological effects is constructed to make a first correction to the residual component, specifically: the holiday state identifier obtained by encoding and the meteorological effect data are converted into dummy variables and meteorological variables, respectively; the dummy variables and meteorological variables are interacted to generate interaction terms for quantifying the meteorological sensitivity during the holiday; the dummy variables, meteorological variables and interaction terms are used as joint inputs to simultaneously calibrate the correction operator by regularized regression fitting of the residual component to be corrected; and the correction operator is used to operate on the residual component to be corrected to correct and output a first corrected residual.

[0009] In a preferred embodiment, the lag effect of capacity ramping is quantified based on the industry-specific operating capacity and embedded in the trend component modeling, specifically including: calculating the month-on-month growth rate of the industry-specific operating capacity sequence to obtain a month-on-month growth sequence; performing convolution operation on the growth sequence using a lognormal-exponential mixed kernel function to obtain a driving sequence representing the smooth ramping process of capacity; embedding the driving sequence as an exogenous variable into the modeling equation for generating the trend component; and synchronously determining the modeling equation and the action weight of the exogenous variable through joint parameter estimation to generate the reconstructed trend component that integrates the lag effect of industry-specific capacity ramping.

[0010] In a preferred embodiment, the impact variable is constructed based on the periodic application capacity to make a second correction to the first corrected residual, specifically including: calculating the cumulative amount of the periodic application capacity sequence within a preset statistical period to generate an initial impact indicator; analyzing the time shift correlation between the initial impact indicator and the historical residual sequence to construct a first-order lag transfer function, and inputting the initial impact indicator into the first-order lag transfer function to output a delayed impact sequence; and performing linear regression with constraints on the first corrected residual with the delayed impact sequence as the explanatory variable to estimate and eliminate the systematic bias to obtain a second corrected residual.

[0011] In a preferred embodiment, the construction of the second prediction model specifically includes: extracting the operating capacity sub-sequence and the application capacity sub-sequence corresponding to the emerging industry from the industry-specific operating capacity sequence and the periodic application capacity sequence according to the preset emerging industry classification; taking the operating capacity sub-sequence and the application capacity sub-sequence as input data, and generating the trend component, the seasonal component and the second corrected residual component corresponding to the emerging industry according to the time series decomposition, residual correction and trend reconstruction rules consistent with the first prediction model; and constructing the second prediction model for outputting the emerging industry electricity consumption prediction value based on the trend component, the seasonal component and the second corrected residual component corresponding to the emerging industry.

[0012] In a preferred embodiment, the mapping is to a power supply and power sales business index, specifically: based on historical data, linear regression relationships between the predicted value of the total social power consumption and the actual power supply and the actual power sales are established respectively, and the power supply conversion coefficient and the power sales conversion coefficient are calibrated; the final predicted value of the total social power consumption is multiplied by the power supply conversion coefficient and the power sales conversion coefficient respectively, and the predicted values of the power supply and the power sales are output simultaneously.

[0013] A device of a multi-time scale power consumption dynamic analysis method, comprising: a data module for acquiring historical daily power consumption sequences and corresponding multi-source covariants; a decoupling module for time series decomposition of the historical daily power consumption sequences to obtain a trend component, a seasonal component and a residual component; a correction module for constructing a correction operator coupled with holidays and meteorological effects to correct the residual component once; based on industry-specific operating capacity, quantifying the hysteresis effect of capacity ramping through a kernel function and embedding it in the trend component to obtain a reconstructed trend component; constructing an impact variable based on periodic application capacity to correct the once corrected residual twice; a modeling module for constructing a first prediction model of the total social power consumption caliber based on the reconstructed trend component, the seasonal component and the twice corrected residual, and constructing a second prediction model of emerging industry power consumption caliber based on emerging industry related data selected from the multi-source covariants; an output module for fusing the prediction results of the first prediction model and the second prediction model to obtain a power consumption prediction value and map it to a power supply and power sales business index.

[0014] A multi-time scale power consumption dynamic analysis device, comprising a memory and a processor: the memory is used to store a program; the processor is used to execute the program to realize each step of the multi-time scale power consumption dynamic analysis method.

[0015] The present application can directly identify and strip the short-term composite fluctuation disturbance formed by the interweaving of social activity rules and natural climate conditions by constructing a correction operator coupled with holidays and meteorological effects to correct the residual component once, so as to obtain a more pure and stable residual sequence, which removes the noise barrier for subsequent accurate analysis, and solves the prediction deviation caused by factor mixing and response lag in the traditional method from the model mechanism. At the same time, an impact variable is constructed based on periodic application capacity to correct the once corrected residual component twice, which realizes the identification and removal of systematic deviation driven by potential growth demand, and significantly improves the purity of the prediction result. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 It is a flowchart of the multi-time scale power consumption dynamic analysis method of the present application; Figure 2 It is a structural diagram of the multi-time scale power consumption dynamic analysis method device of the present application; Figure 3 A structural block diagram of an exemplary electronic device that can be used to implement embodiments of the present disclosure is provided for embodiments of the present application. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the protection scope of the present application.

[0018] Embodiment 1, Figure 1 The multi-time scale electricity dynamic analysis method of the present application is given, including the following steps: S1, obtaining a historical daily electricity consumption sequence and corresponding multi-source covariates; In the present embodiment, the obtaining of the historical daily electricity consumption sequence and the corresponding multi-source covariates includes: The multi-source covariates include holiday, meteorological effect data, industry-specific operation capacity and periodic application capacity.

[0019] S11, forming a basic electricity sequence. From the energy management system of the target area, obtain the historical daily electricity consumption original data of the past N years. Use the forward filling method to repair single-day missing values, and ensure that all data points are strictly arranged according to the Gregorian calendar date, forming a continuous basic electricity sequence with day as the basic time unit , wherein t is the date index.

[0020] S12, obtaining and encoding holiday state identifier and meteorological effect data corresponding to the time reference. Based on the published statutory holiday arrangement, generate a holiday state identifier sequence completely corresponding to the date . The specific encoding rule is: 0 for ordinary weekdays, 1 for weekends, 2 for statutory holidays, and 3 for make-up workdays. At the same time, the daily maximum temperature data of the target area in the same period is extracted from the climate monitoring data. In order to quantify its marginal effect (rather than absolute temperature), it is standardized and graded: calculate the temperature percentile (such as 25%, 50%, 75%) of the entire period, divide the daily temperature into 4 grades, as shown in Table 1, and generate a meteorological grade sequence .

[0021] Table 1

[0022] S13, constructing the sub-industry operation capacity sequence. Obtain the monthly statistical operation capacity data of each industry from the relevant database, total the operation capacity of the sub-industry under the same industry category to obtain the monthly total operation capacity of the aggregated industry, for example, the operation capacity of the aggregated "high-end manufacturing industry". Since the daily data is daily, and the operation capacity data is monthly, time scale alignment is required. Linear interpolation method is used to convert the above aggregated monthly operation capacity sequence into a smooth daily sequence. Specifically, the operation capacity value at the end of the Mth month is known , and the operation capacity value of each day in the month is interpolated as a linearly increasing sequence from the beginning of the month to the end of the month. In this way, the sub-industry operation capacity sequence with the same time granularity is constructed, denoted as , where i represents the i-th aggregated industry.

[0023] S14, constructing the periodic application capacity sequence. Extract the original electricity application records from the power grid company. Each record contains the application date, voltage level and application capacity. Sum the capacity of all low-voltage electricity applications per day based on the application date to obtain the daily low-voltage application capacity; similarly, sum to obtain the daily high-voltage application capacity. The sum of the daily low-voltage application capacity and the daily high-voltage application capacity is the daily total application capacity, which constitutes the original periodic application capacity sequence.

[0024] S2, time series decomposition of the historical daily electricity consumption sequence to obtain the trend component, seasonal component and residual component; In this embodiment, the historical daily electricity consumption sequence is decomposed into trend component, seasonal component and residual component, specifically: S21, decomposition of the basic electricity sequence. The initial processing is performed on the basic electricity sequence using the seasonal decomposition method. In a preferred embodiment, the STL (Seasonal and Trend decomposition using Loess) decomposition method is used, with the seasonal period parameters S=7 (to capture the weekly cycle) and S=365 (to capture the annual cycle). Through STL decomposition, the basic electricity sequence is decomposed into three additive components:

[0025] wherein, is the initial trend component, reflecting the long-term change direction of electricity consumption, is the seasonal component, reflecting the fixed weekly and annual periodic fluctuations, is the initial residual component, containing the remaining fluctuations and noise that are not explained by the trend and season. ​​

[0026] S22, Identification of Trend Mutation Period. The industry-specific running capacity sequence is subjected to first-order difference operation to obtain its daily variation sequence, and a 7-day moving average sequence of the daily variation sequence is calculated to smooth random fluctuations and form a smooth indicator sequence reflecting the short-term momentum of industry expansion.

[0027] is taken as an external reference signal, and the initial trend component is adjusted to obtain a preliminary trend component .

[0028] S23, Robust Decomposition within Trend Mutation Period. Within each identified trend mutation period, robust decomposition is performed on the sub-sequence of the base electricity consumption sequence . Traditional STL decomposition may blur the structural changes in the mutation period and distribute them in the trend and residual. This step aims to more cleanly separate the increment driven by the specific reason of “capacity ramp-up”. Typically, decomposition techniques insensitive to outliers are used, for example, using robust local regression based on M-estimators to fit the trend, or using the RANSAC (Random Sample Consensus) regression framework to iteratively fit the trend line, thereby resisting the interference of abnormal fluctuations that may exist within the mutation period. Through robust decomposition, a more pure increment trend component within the trend mutation period is obtained.

[0029] S24, Fusion of Trend Components. The increment trend component obtained by robust decomposition is added to the initial trend component to form the final trend component that can comprehensively reflect long-term changes (including smooth changes and structural mutations). In non-mutation periods, the initial trend component is the final trend component .

[0030] S25, Generation of Residual Component to be Corrected. In the base electricity consumption sequence , the final trend component and the seasonal component are removed to obtain the residual component to be corrected : ​​​

[0031] S3, constructing a correction operator coupling holiday and weather effects, and correcting the residual component once; In the present embodiment, the correction operator coupling holiday and weather effects is constructed, and the residual component is corrected once, in particular as follows: S31, converting the holiday state indicator sequence into dummy variables that can be used in linear models. Since contains multiple categories, a k-1 dummy variable encoding scheme is adopted to avoid multicollinearity. For example, taking ordinary weekdays (0) as the baseline, the following three dummy variables are created, as shown in Table 2. For simplicity of expression, the set of these three dummy variables is denoted as vector .

[0032] Table 2

[0033] To capture the nonlinear effect of different temperature levels, the weather level sequence is treated as an ordinal categorical variable and numerically encoded. In a preferred embodiment, the level number is directly used as the weather variable, i.e. let .

[0034] S32, generation of interaction terms. Recognizing that social behavior patterns change during holidays, which can change people's sensitivity to weather conditions (for example, holiday high temperatures can lead to a greater increase in air conditioning load), the present invention constructs interaction terms to explicitly quantify this coupling effect. Specifically, the dummy variable representing the holiday core state is multiplied with the weather variable to generate the interaction term .

[0035] S33, forming the correction operator. For the historical period to N, a feature matrix F is constructed, each row of which contains: holiday dummy variables , weather variables and interaction terms . With the historical residual component to be corrected as the target variable, and the feature matrix F as the input, a linear regression fitting with regularization is performed. Preferably, ridge regression is used, whose optimization objective is:

[0036] where is the coefficient vector to be estimated, is the regularization strength hyperparameter, which is determined by cross-validation.

[0037] After fitting, the obtained coefficient vector That is, the correction operator is defined. The operator is a linear function For any given feature vector , the output is , that is, the predicted, residual part that should be explained by holiday and weather effects.

[0038] Apply the calibrated correction operator to the entire time series of the residual components to be corrected. For each time point t, generate the corresponding feature vector according to its holiday and weather information, and calculate the predicted effect value of the point:

[0039] Subtract the predicted effect value from the original residual component to be corrected to obtain a corrected residual .

[0040] S4, based on the running capacity of the sub-industry, quantifies the hysteresis effect of capacity climbing through a kernel function, and embeds it into the trend component modeling to obtain the reconstructed trend component; In the embodiment, the kernel function is used to quantify the hysteresis effect of capacity climbing based on the running capacity of the sub-industry, and is embedded into the trend component modeling, which specifically includes: S41, calculating the month-on-month growth sequence. Based on the sub-industry running capacity basic sequence , the month-on-month growth rate of each time point t relative to the running capacity 30 days ago (approximately one month) is calculated, and then the month-on-month growth sequence is obtained. The calculation formula of the month-on-month growth rate is specifically:

[0041] S42, generation of driving sequence. After the capacity is put into operation, the corresponding electricity load does not immediately reach full production, but gradually releases with production line debugging and capacity climbing. The present application adopts a specific convolution kernel function to simulate this hysteresis distribution. In the embodiment, a lognormal-exponential mixed kernel function is adopted, and its mathematical definition is:

[0042] wherein, is the lognormal distribution component, simulating the time lag of the main construction and debugging period of capacity climbing; is the log-scale position parameter, representing the typical time scale (median) of the capacity climbing process; is the log-scale shape parameter, representing the uncertainty or dispersion degree of the climbing period; is the exponential distribution component, simulating the process that the impact of capacity after reaching the peak is gradually attenuated by technical, market or equipment factors; is the decay rate parameter, representing the rate of decay of the capacity effect; is the mixing weight.

[0043] The monthly growth sequence is regarded as a series of "capacity expansion pulses" occurring at different times, and the impact of the diffusion of these pulses over time is simulated by convolution operation. The driving sequence is calculated:

[0044] where L is the effective length of the kernel function (e.g. 180 days). The sequence is the driving sequence that is smooth, continuous and contains the standard lag effect, which represents the potential pressure of "effective capacity ramp-up" on electricity demand.

[0045] S43, the modeling equation of the driving sequence embedded in the trend component as an exogenous variable. An extended autoregressive integrated moving average model (ARIMAX) is used as the basic equation for trend modeling. Let the trend component to be generated be , the following ARIMAX (p, d, q) model structure is constructed:

[0046] where are autoregressive and moving average operator polynomials, respectively, is the difference operator, is a white noise sequence, is a new term, is the to-be-estimated coefficient of the driving sequence, which directly introduces the lag driving effect of industry ramp-up into the trend generation equation.

[0047] S44, joint parameter estimation and trend component generation. The final trend component is used as the training target, and the maximum likelihood estimation or conditional least squares method is used to simultaneously optimize all parameters of the ARIMAX model.

[0048] Using the estimated complete model, the trend is calculated for the in-sample and prediction period. For the prediction period, based on the known or predicted industry operating capacity data, the future driving sequence is generated by repeating S41-S42, and it is substituted into the model to generate the reconstructed trend component that integrates the lag effect of industry capacity ramp-up.

[0049] S5, based on the periodic application capacity to construct the impact variable, the once-revised residual is revised again; In the present embodiment, the impact variable is constructed based on the periodic application capacity, and the once-revised residual is revised again, specifically including: S51, generating an initial impact index. Based on the sequence of periodic application capacity, the total application capacity per day is obtained , which has been embodied in the above S14, and will not be repeated here. In order to characterize the accumulated potential demand pressure, a preset statistical window reflecting the typical period of investment intention converted into actual demand is defined , in a preferred embodiment, days, to characterize the investment and construction period at the annual level. The rolling cumulative amount in the window is calculated to generate the initial impact index :

[0050] Initial impact index represents the total amount of potential demand accumulated in the past days, which has been submitted but has not yet been fully converted into electricity load as of date t, which is a leading pressure indicator of future electricity growth.

[0051] S52, construction of first-order lag transfer function and output of delayed impact sequence. Calculate the initial impact index and the historical residual component to be revised (output from S25) cross-correlation coefficient under different lag orders :

[0052] wherein, is a correlation coefficient calculation function, usually referring to Pearson correlation coefficient, used to calculate the linear correlation degree between two time series.

[0053] Analyzing the trend of cross-correlation coefficient , it is found that its influence is still significant after tens to hundreds of days of lag, and the correlation intensity presents an exponential decay trend with time, which confirms that the influence of application capacity on electricity residual has a delayed and decaying conduction effect. Therefore, a first-order lag transfer function is constructed to quantify this conduction law. The core of the function is a decay coefficient , which physically means the retention and release rate of the impact influence over time. Input the initial impact index into this function, and generate the delayed impact sequence , which is essentially an exponentially weighted moving average of , which smooths the noise of the original index and reasonably delays the influence on the time axis, more accurately characterizing the dynamic release process of "potential demand pressure" over time.

[0054] S53, obtain the second revised residual. Add the first revised residual as the dependent variable, and the delayed impact sequence as the explanatory variable, a linear regression model is established:

[0055] wherein, is the intercept, is the coefficient of , and is the new residual.

[0056] When fitting the model, the regression coefficient of is subjected to a non-negative constraint, which ensures that the revision direction is logical and improves the robustness of the model. Least squares method with this inequality constraint is used for parameter estimation. Using the estimated model, the systematic deviation part that can be explained by is calculated, and the systematic deviation part is removed from the first revised residual to obtain the second revised residual .

[0057] S6, based on the reconstructed trend component, seasonal component and the second revised residual, a first prediction model of the total social electricity consumption is constructed, and based on the emerging industry related data selected from the multi-source covariates, a second prediction model of the emerging industry electricity consumption is constructed; In the embodiment, the construction of the first prediction model and the second prediction model is specifically: S61, the first prediction model is constructed. The reconstructed trend component , seasonal component and second revised residual are added and summed to reconstruct the training sequence of the daily electricity consumption of the total social caliber after the whole process optimization, and the sequence is used as the training target to train a final total social electricity consumption prediction model, i.e. the first prediction model of the total social electricity consumption caliber. The model can use a light machine learning model (such as gradient boosting tree) or a time series model, and the output is the first prediction value .

[0058] S62, constructing a second prediction model. According to a preset emerging industry classification (such as "computer, communication and other electronic equipment manufacturing industry", "electrical machinery and equipment manufacturing industry"), from the sub-industry operation capacity sequence and the period application capacity sequence, the running capacity sub-sequence and the application capacity sub-sequence corresponding to the emerging industry are extracted, the running capacity sub-sequence and the application capacity sub-sequence are taken as input data, and the trend component, the seasonal component and the secondary correction residual component corresponding to the emerging industry are generated according to the time sequence decomposition, residual correction and trend reconstruction rules consistent with the first prediction model; based on the trend component, the seasonal component and the secondary correction residual component corresponding to the emerging industry, the power consumption sequence of the emerging industry is reconstructed, and an independent prediction model, i.e. the second prediction model of the emerging industry power consumption caliber, is trained to obtain a second prediction value . The detailed process of this step is the same as the construction of the first prediction model and the output method of the first prediction value, which will not be repeated here.

[0059] S7, fusing the prediction results of the first prediction model and the second prediction model to obtain the power consumption prediction value and mapping it to the power supply and power sales business indicators.

[0060] In this embodiment, the fusion of the prediction results of the first prediction model and the second prediction model to obtain the power consumption prediction value is specifically: for the first prediction model and the second prediction model, their rolling average absolute percentage errors in the last K days (for example, K=90) are calculated respectively, denoted as and The model with smaller error should obtain higher weight, and the dynamic weight of the first prediction model is calculated :

[0061] The final total social power consumption prediction value :

[0062] In this embodiment, the mapping to the power supply and power sales business indicators is specifically: the final total social power consumption prediction value sequence and the actual statistical daily power supply sequence and daily power sales sequence of the same period (such as the past three years) are collected, two simple linear regression relationships are established respectively, and the power supply conversion coefficient, the power sales conversion coefficient and the intercept are calibrated by least square fitting. For the future prediction day, the final total social power consumption prediction value is input into the calibrated linear relationship, and the power supply prediction value and the power sales prediction value can be obtained.

[0063] Embodiment 2, Figure 2 The device of the multi-time scale power consumption dynamic analysis method is given, which comprises: a data module for acquiring historical daily power consumption sequence and corresponding multi-source covariates; a decoupling module configured to perform time series decomposition on the historical daily electricity consumption sequence to obtain a trend component, a seasonal component, and a residual component; a correction module configured to construct a correction operator coupled with a holiday and a meteorological effect, and perform a first correction on the residual component; based on an industry-specific operation capacity, a kernel function is used to quantify a hysteresis effect of capacity ramping, and the hysteresis effect is embedded into the trend component to obtain a reconstructed trend component; a shock variable is constructed based on a periodic application capacity, and a second correction is performed on the first corrected residual; a modeling module configured to construct a first prediction model of a social electricity consumption caliber based on the reconstructed trend component, the seasonal component, and the second corrected residual, and construct a second prediction model of a new industry electricity caliber based on new industry related data selected from multiple source covariates; an output module configured to fuse prediction results of the first prediction model and the second prediction model to obtain an electricity consumption prediction value and map the electricity consumption prediction value to power supply and power sales business indicators.

[0064] Embodiment 3: A multi-time scale electricity dynamic analysis device, as shown in Figure 3 includes a memory and a processor: the memory is configured to store a program; the processor is configured to execute the program to implement any of the embodiments of embodiment 1.

[0065] Since the multi-time scale electricity dynamic analysis device introduced in the embodiment is a device used to implement the method in embodiment 1 of the present application, the specific implementation of the electronic device of the embodiment and its various forms can be understood by those skilled in the art based on the method introduced in embodiment 1 of the present application. Therefore, the method of the electronic device in the present application will not be described in detail. As long as the device used to implement the method in the present application is implemented by those skilled in the art, it belongs to the scope of the present application.

[0066] The above formulas are dimensionless numerical calculations, the formulas are obtained by collecting a large amount of data to simulate the recent real situation, and the preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0067] The above embodiments can be realized wholly or partially by software, hardware, firmware, or any combination thereof. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product.

[0068] Those skilled in the art can understand that the modules and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0069] In addition, each functional module in each embodiment of the present application can be integrated in one processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0070] The above is merely specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any skilled in the art within the technical scope disclosed in the present application can easily think of changes or replacements, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0071] Finally: the above is only the preferred embodiment of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for dynamic analysis of electricity consumption across multiple time scales, characterized in that, Includes the following steps: Obtain historical daily electricity consumption sequences and corresponding multi-source covariates; The historical daily electricity consumption series is decomposed into trend components, seasonal components, and residual components. A correction operator coupling holidays and meteorological effects is constructed to correct the residual components once. Based on the operating capacity of different industries, the lag effect of capacity ramp-up is quantified by kernel function and embedded into the trend component model to obtain the reconstructed trend component. Based on the periodic application capacity, an impact variable is constructed, and the residual after the first correction is then corrected a second time. Based on the reconstructed trend component, seasonal component, and residual after secondary correction, a first prediction model for total electricity consumption is constructed. Based on emerging industry-related data selected from multi-source covariates, a second prediction model for emerging industry electricity consumption is constructed. By integrating the prediction results of the first and second prediction models, the predicted electricity consumption value is obtained and mapped into electricity supply and sales business indicators.

2. The multi-timescale dynamic analysis method for electricity consumption according to claim 1, characterized in that, The acquisition of historical daily electricity consumption sequences and corresponding multi-source covariates includes: The multi-source covariates include holiday data, meteorological effect data, industry-specific operating capacity, and periodic application capacity. Historical daily electricity consumption is obtained and aligned according to a unified daily time base to form a basic electricity consumption sequence; Acquire and encode holiday status identifiers and meteorological effect data corresponding to the time base; Spatial aggregation and daily-scale sampling are performed on industry-specific operational capacity data to construct industry-specific operational capacity sequences. Based on the periodic application capacity, a periodic application capacity sequence aligned with the time base is constructed.

3. The multi-timescale dynamic electricity consumption analysis method according to claim 2, characterized in that, The time-series decomposition of historical daily electricity consumption sequences yields trend components, seasonal components, and residual components, specifically: The basic electricity sequence is decomposed to obtain the initial trend component, seasonal component, and initial residual component; Based on the industry-specific operating capacity sequence, the stationarity of the initial trend component is tested to identify the period of trend abrupt change caused by industrial restructuring. During periods of abrupt trend change, robustness decomposition is performed to separate the incremental trend component dominated by the capacity ramp-up effect. The incremental trend component is merged with the initial trend component to form the final trend component; The final trend component and seasonal component are removed from the basic electricity series to obtain the residual component to be corrected.

4. The multi-timescale dynamic electricity consumption analysis method according to claim 3, characterized in that, The aforementioned modified operator that couples holidays and meteorological effects performs a correction on the residual components, specifically as follows: The holiday status identifiers and meteorological effect data obtained from the coding are converted into dummy variables and meteorological variables, respectively. Interacting dummy variables with meteorological variables generates interactive terms for quantifying meteorological sensitivity during holidays; Using dummy variables, meteorological variables, and interaction terms as joint inputs, the correction operator is formed by regularizing regression fitting of the residual components to be corrected and simultaneously calibrating. The correction operator is used to perform calculations on the residual components to be corrected, and the corrected residual is output.

5. The multi-timescale dynamic electricity consumption analysis method according to claim 4, characterized in that, The method, based on industry-specific operating capacity, quantifies the lag effect of capacity ramp-up using kernel functions and embeds it into trend component modeling, specifically includes: The month-on-month growth rate is calculated based on the industry-specific operating capacity sequence to obtain the month-on-month growth sequence. The growth sequence is convolved using a log-normal-exponential hybrid kernel function to obtain a driving sequence characterizing the smooth ramp-up process of production capacity; The driving sequence is embedded as an exogenous variable into the modeling equation used to generate the trend component; By simultaneously determining the role weights of the modeling equations and exogenous variables through joint parameter estimation, a reconstructed trend component that integrates the lag effect of capacity ramp-up in different industries is generated.

6. The multi-timescale dynamic analysis method for electricity consumption according to claim 5, characterized in that, The process of constructing an impact variable based on periodic application capacity and then performing a secondary correction on the residuals after the first correction specifically includes: Based on the periodic application capacity sequence, calculate its cumulative amount within a preset statistical period to generate an initial impact index. The time-shift correlation between the initial shock index and the historical residual sequence is analyzed, a first-order lag transfer function is constructed, and the initial shock index is input into the first-order lag transfer function to output the delayed shock sequence. Using the delayed shock sequence as the explanatory variable, a constrained linear regression is performed on the first-correction residual to estimate and remove its systematic bias, thus obtaining the second-correction residual.

7. The multi-timescale dynamic electricity consumption analysis method according to claim 6, characterized in that, The construction of the second prediction model is as follows: Based on the pre-defined classification of emerging industries, extract the corresponding operating capacity subsequence and application capacity subsequence from the industry-specific operating capacity sequence and the periodic application capacity sequence; Using the running capacity subsequence and the requested capacity subsequence as input data, and following the time series decomposition, residual correction and trend reconstruction rules consistent with the first prediction model, the trend component, seasonal component and secondary correction residual component corresponding to the emerging industry are generated. Based on the trend component, seasonal component, and secondary correction residual component corresponding to the emerging industries, a second prediction model is constructed to output the predicted electricity consumption value of the emerging industries.

8. The multi-timescale dynamic analysis method for electricity consumption according to claim 7, characterized in that, The mapping of this to electricity supply and electricity sales business indicators is as follows: Based on historical data, linear regression relationships were established between the predicted total electricity consumption and the actual electricity supply and sales, respectively, and the electricity supply conversion coefficient and the electricity sales conversion coefficient were calibrated. The final predicted total electricity consumption is multiplied by the power supply conversion coefficient and the power sales conversion coefficient, respectively, and the predicted power supply value and the predicted power sales value are output simultaneously.

9. An apparatus for using the multi-timescale dynamic electricity consumption analysis method as described in any one of claims 1-8, characterized in that, include: The data module is used to obtain historical daily electricity consumption sequences and corresponding multi-source covariates; The decoupling module is used to perform time-series decomposition on historical daily electricity consumption sequences to obtain trend components, seasonal components, and residual components. The correction module is used to construct a correction operator that couples holidays and meteorological effects, and performs a correction on the residual components. Based on the operating capacity of different industries, the lag effect of capacity ramp-up is quantified by kernel function and embedded into the trend component model to obtain the reconstructed trend component. Based on the periodic application capacity, an impact variable is constructed, and the residual after the first correction is then corrected a second time. The modeling module is used to construct a first prediction model for total electricity consumption based on the reconstructed trend component, seasonal component, and residual after secondary correction, and to construct a second prediction model for electricity consumption of emerging industries based on emerging industry-related data selected from multi-source covariates. The output module is used to integrate the prediction results of the first prediction model and the second prediction model to obtain the electricity consumption prediction value and map it into the electricity supply and sales business indicators.

10. A multi-timescale dynamic power consumption analysis device, characterized in that, Including memory and processor: The memory is used to store programs; The processor is used to execute the program to implement each step of the multi-timescale dynamic electricity consumption analysis method as described in any one of claims 1-8.

Images