A semi-normalized method and system for predicting electricity load regression

Abstract

This invention discloses a semi-normalized electricity load regression prediction method and system, comprising: explicitly defining and collecting known influencing factors affecting electricity load; constructing an electricity load prediction model, wherein the electricity load prediction model is used to calculate the predicted electricity load value; performing a first-step training, based on historical electricity load data and historical known influencing factors, training the weights of each known influencing factor and the random error term in the electricity load prediction model; performing a second-step training, based on the results of the first-step training, extracting the residuals, training the coefficients of unknown influencing factors in the electricity load prediction model, and summarizing the statistical regularities of the unknown influencing factors; inputting the known influencing factors at the time to be predicted into the trained electricity load prediction model, and calculating the predicted electricity load value at the time to be predicted. This invention reduces the data length requirement, enables small-sample learning, can explicitly integrate known regularities, and enhances flexibility and interpretability.

Description

Technical Field

[0001] This invention belongs to the field of power system load forecasting, specifically relating to a semi-normalized power load regression forecasting method and system. Background Technology

[0002] Electricity load forecasting is a core technology for power system planning, dispatching, and demand-side management, and its accuracy directly affects the economy and reliability of power grid operation. In power systems, electricity load forecasting is used not only for short-term dispatching to ensure stable power supply but also for long-term planning to optimize resource allocation, such as avoiding overload or waste in peak-valley load management. Traditional load forecasting methods mainly rely on statistical learning (such as linear regression and ARIMA) or machine learning (such as neural networks and random forests). These methods perform well when historical data is abundant, but they all have the following limitations:

[0003] 1. Strong Data Dependence: Traditional models require at least 1-3 years of historical load data (including complete weekly, monthly, and yearly cycles) to capture the cyclical patterns of load (such as weekday / weekend differences, monthly electricity consumption fluctuations, and seasonal temperature effects). For example, capturing weekly cycles requires sufficient weekday and weekend data to distinguish electricity consumption patterns; in monthly fluctuations, multiple months must be covered to identify seasonal changes. However, real-world scenarios often face problems such as missing data (e.g., no historical data for newly built industrial parks), low data quality (e.g., outliers caused by equipment malfunctions), or delayed data updates (e.g., rapid modeling is required after temporary adjustments to electricity pricing policies). These problems may stem from untimely equipment maintenance or external policy changes, leading to model training failures or decreased prediction accuracy. Specifically, in the case of missing data, the model cannot fully learn the cyclical patterns, resulting in amplified prediction bias; with low data quality, outliers interfere with parameter estimation; and with delayed updates, the model cannot adapt to the new environment in a timely manner.

[0004] 2. Insufficient Utilization of Known Patterns: Weekly, monthly, and annual load variations, as well as the influence of temperature, are recognized key factors in the field. However, traditional models need to implicitly learn these patterns through large amounts of data, and cannot actively "tell" the model of the existence of known patterns. This makes it difficult for the model to effectively extract these features when data is insufficient, resulting in large prediction biases. For example, weekly patterns such as peak weekday electricity consumption usually need to appear repeatedly in the data to be captured by the model, but if data is scarce, this implicit learning cannot be fully carried out. Similarly, the influence of temperature, as a known factor, may be ignored by the model in small samples if it is not actively incorporated, leading to an underestimation of load in predictions during hot weather.

[0005] 3. Poor flexibility and interpretability: While machine learning models (such as neural networks) can capture complex nonlinear relationships, they are black-box models, unable to clearly distinguish the influence weights of known patterns (such as weekly cycles) and unknown factors (such as sudden events). This makes debugging and optimization difficult, and users struggle to understand the model's decision-making basis. Statistical models (such as linear regression), while interpretable, require manual addition of features based on known patterns (such as "day of the week" and "month"), and cannot automatically summarize unknown factors, resulting in insufficient flexibility and failing to meet the needs of rapid modeling. For example, in linear regression, users must pre-design feature variables, but if unknown factors such as sudden events occur, the model cannot adaptively adjust, leading to poor performance in dynamic environments. Summary of the Invention

[0006] The purpose of this invention is to provide a semi-normalized electricity load regression prediction method and system, which aims to solve the problems of existing load prediction methods having high requirements for the length of historical data, difficulty in effectively utilizing known periodic patterns when data is insufficient, the need for implicit learning of known patterns through data rather than active integration of known patterns, low prediction accuracy in small data scenarios, and poor model flexibility and interpretability.

[0007] In view of the above problems, this application provides a semi-normalized electricity load regression prediction method and system.

[0008] The first aspect disclosed in this application provides a semi-normalized electricity load regression prediction method, the method comprising the following steps:

[0009] Step 1: Explicitly define and collect known influencing factors affecting electricity load. Each known influencing factor corresponds to a quantifiable characteristic variable. The known influencing factors include factors with periodic patterns and external environmental characteristics.

[0010] Step 2: Construct an electricity load forecasting model. The electricity load forecasting model calculates the predicted electricity load value by weighting and summing each known influencing factor, and combining the self-inductive unknown influencing factors, the coefficients of the unknown influencing factors, and the random error term.

[0011] Step 3: Perform the first step of training. Based on historical electricity load data and historical known influencing factors, train the weights of each known influencing factor and the random error term in the electricity load prediction model.

[0012] Step 4: Perform the second training step, extract the residuals based on the results of the first training step, train the coefficients of unknown influencing factors in the electricity load prediction model, and summarize the statistical laws of unknown influencing factors.

[0013] Step 5: Input the known influencing factors at the time to be predicted into the trained electricity load prediction model, and calculate the predicted electricity load value at the time to be predicted.

[0014] Preferably, step 2 specifically includes:

[0015] Constructing an electricity load prediction model of form (1):

[0016] Equation (1)

[0017] in, This represents the weight of the i-th known influencing factor. Let m represent the i-th known influencing factor at time t, and m represent the number of known influencing factors. The coefficient representing the unknown influencing factor. Indicates unknown influencing factors. Represents the random error term. This represents the predicted electricity load at time t.

[0018] Preferably, step 3 specifically includes the following steps:

[0019] Step 3.1: Initialize the weights of each known influencing factor in the electricity load forecasting model;

[0020] Step 3.2: Degenerate the electricity load prediction model into the form of equation (2):

[0021] Equation (2)

[0022] Step 3.3: Based on historical electricity load data and known historical influencing factors, Equation (3) is used as the optimization objective, and the weights of the known influencing factors are solved using the least squares method. and random error term This minimizes the mean square error between the electricity load calculated by equation (2) and the historical electricity load data.

[0023] Equation (3)

[0024] in, Let t represent the historical electricity load at time t, and n represent the number of samples of historical electricity load data.

[0025] Preferably, step 4 specifically includes the following steps:

[0026] Step 4.1: Based on the weights of the known influencing factors obtained from the solution. and random error term The residuals are calculated using equation (4):

[0027] Equation (4)

[0028] in, Represents the residual;

[0029] Step 4.2: Using equation (5) as the optimization objective, the coefficients of the unknown influencing factors are solved by minimizing the error between the residual and the coefficient of the unknown influencing factors multiplied by the coefficient of the unknown influencing factors.

[0030] Equation (5)

[0031] Among them, unknown influencing factors They follow a preset distribution, which includes the normal distribution. Or a Poisson distribution, wherein the preset distribution is based on historical residuals. The actual distribution characteristics are determined, and the coefficients of unknown influencing factors are determined. The standard deviation of the residuals;

[0032] Step 4.3: By making the residual Divide by the coefficient of unknown influencing factors Extracting unknown influencing factors And summarize the unknown influencing factors. The statistical regularities, including the mean, variance, and time autocorrelation of unknown influencing factors, are used to predict the value of unknown influencing factors at the time to be predicted.

[0033] Preferably, the weights of the known influencing factors are adjusted according to the actual load scenario, and the distribution and coefficients of the unknown influencing factors are corrected according to the actual distribution of the unknown influencing factors.

[0034] A second aspect of this application discloses a semi-normalized electricity load regression prediction system, the system comprising:

[0035] Feature module: Explicitly defines and collects known influencing factors affecting electricity load, with each known influencing factor corresponding to a quantifiable feature variable;

[0036] Model building module: Constructs an electricity load prediction model. The electricity load prediction model calculates the predicted electricity load value by weighting and summing each known influencing factor, and combining the self-inductive unknown influencing factors, the coefficients of the unknown influencing factors, and the random error term.

[0037] First training module: Perform the first step of training, based on historical electricity load data and historical known influencing factors, to train the weights of each known influencing factor and the random error term in the electricity load prediction model;

[0038] The second training module: Execute the second training step, extract the residuals based on the results of the first training step, train the coefficients of unknown influencing factors in the electricity load prediction model, and summarize the statistical laws of unknown influencing factors.

[0039] Prediction module: Input the known influencing factors at the time to be predicted into the trained electricity load prediction model, and calculate the predicted value of the electricity load at the time to be predicted.

[0040] The third aspect disclosed in this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the aforementioned semi-normalized electricity load regression prediction method.

[0041] The fourth aspect disclosed in this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the aforementioned semi-normalized electricity load regression prediction method.

[0042] The fifth aspect disclosed in this application provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the aforementioned semi-normalized electricity load regression prediction method.

[0043] The beneficial effects of this invention are as follows:

[0044] 1. Reduced data length requirements: Only 3-6 months of data are needed to train a high-precision model. Traditional methods require 1-3 years of historical data, which is suitable for data missing, new scenarios, or rapid update needs, and can achieve small sample learning.

[0045] 2. Explicitly integrate known patterns: By pre-defining known factors such as week, month, and temperature, the model can actively capture the influence of these patterns, avoiding the problem of insufficient implicit learning in small data scenarios, and achieving a significant improvement in prediction accuracy compared to traditional methods.

[0046] 3. Enhanced flexibility and interpretability: The weights of known influencing factors can be manually adjusted, while unknown influencing factors can be explicitly displayed through residual analysis. The model retains the adaptability of machine learning and the interpretability of statistical models, making it easy to optimize according to actual scenarios. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart of a semi-normalized electricity load regression prediction method.

[0049] Figure 2 This is a diagram showing the overall structure of a semi-normalized electricity load regression prediction system. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Example 1:

[0052] like Figure 1 As shown in the figure, this application provides a semi-normalized electricity load regression prediction method, the method comprising:

[0053] Step 1: Explicitly define and collect known influencing factors affecting electricity load (such as weekly cycles, monthly cycles, temperature effects, etc.), which can be viewed as a set. Each known influencing factor corresponds to a quantifiable characteristic variable (such as "day of the week", "month", "daily high temperature"). These known influencing factors include factors with periodic patterns and external environmental characteristics. These variables can be collected in real time through sensors or databases to ensure that the data is quantifiable and easy to input into the model.

[0054] Step 2: Construct an electricity load forecasting model. This model calculates the predicted electricity load by weighted summation of each known influencing factor and combining the self-inductive unknown influencing factors, their coefficients, and a random error term. This construction method allows the model to handle both known and unknown components simultaneously, improving overall robustness.

[0055] Step 2 specifically includes:

[0056] Constructing an electricity load prediction model of form (1):

[0057] Equation (1)

[0058] in, This represents the weight of the i-th known influencing factor. Let m represent the i-th known influencing factor at time t, and m represent the number of known influencing factors. The coefficient representing the unknown influencing factor. Indicates unknown influencing factors. Represents the random error term. This represents the predicted electricity load at time t. This formula ensures the combination of the linear part of the weighted summation with the unknown part, facilitating subsequent training.

[0059] Step 3: Perform the training from step one. Based on historical electricity load data and known historical influencing factors, train the weights of each known influencing factor and the random error term in the electricity load prediction model. This training focuses on the known parts, ensuring that the model first captures explicitly defined patterns.

[0060] Step 3 specifically includes the following steps:

[0061] Step 3.1: Initialize the weights of each known influencing factor in the electricity load prediction model. For example, the weights can be initially set to zero or an empirically based uniform value to avoid training starting from an unreasonable starting point.

[0062] Step 3.2: Degenerate the electricity load forecasting model into the form of equation (2), that is, ignore the unknown influencing factors and focus on the known factors and random errors:

[0063] Equation (2)

[0064] To temporarily ignore unknown factors (assuming) The technical basis for (=0) is as follows: In small data scenarios (such as 3-6 months of data), the contribution of known influencing factors (weekly cycle, temperature, etc.) to the load exceeds 70%. Prioritizing the fitting of known influencing factors can avoid the interference of unknown influencing factors in the extraction of the patterns of known influencing factors, ensuring... and "stability";

[0065] Step 3.3: Based on historical electricity load data and known historical influencing factors, Equation (3) is used as the optimization objective, and the weights of the known influencing factors are solved using the least squares method. and random error term This minimizes the mean square error between the electricity load calculated by equation (2) and the historical electricity load data.

[0066] Equation (3)

[0067] in, The t represents the historical electricity load at time t, and n represents the number of samples of historical electricity load data. Due to the characteristics of this invention, a small amount of historical electricity load data (such as data from 3-6 months) can be used here.

[0068] Specifically, the training data consists of load data from the past three months (June-August), totaling 8,640 samples (collected every 15 minutes, meeting the requirements for refined load data collection in industrial parks). The samples include timestamps, actual load values, week of the day (f1), month (f2), and daily maximum temperature (f3).

[0069] Step 4: Perform the second training step. Based on the results of the first training step, extract the residuals as proxy variables for unknown factors, train the coefficients of unknown influencing factors in the electricity load prediction model, and summarize the statistical regularities of unknown influencing factors. This training expands the model's ability to handle unknown parts.

[0070] Step 4 specifically includes the following steps:

[0071] Step 4.1: Based on the weights of the known influencing factors obtained from the solution. and random error term The residuals that cannot be explained by known influencing factors in historical electricity load data are calculated using Equation (4):

[0072] Equation (4)

[0073] in, Represents the residual;

[0074] Step 4.2: Assume U(t) is a zero-mean random variable (or has a specific distribution), then Let the standard deviation (or scaling factor) of the unknown factors be used. Equation (5) is adopted as the optimization objective. The coefficients of the unknown factors are solved by minimizing the error between the residual and the coefficient of the unknown factors multiplied by the coefficient of the unknown factors.

[0075] Equation (5)

[0076] Among them, unknown influencing factors They follow a preset distribution, which includes the normal distribution. Or a Poisson distribution, wherein the preset distribution is based on historical residuals. The actual distribution characteristics are determined, and the coefficients of unknown influencing factors are determined. The standard deviation of the residuals;

[0077] Step 4.3: By making the residual Divide by the coefficient of unknown influencing factors Extracting unknown influencing factors And summarize the unknown influencing factors. The statistical regularities (such as mean, variance, and time autocorrelation) are used to predict unknown scenarios. The statistical regularities include the mean, variance, and time autocorrelation of unknown influencing factors, which are used to predict the value of unknown influencing factors at the time to be predicted.

[0078] The weights of the known influencing factors are adjusted according to the actual load scenario, and the distribution and coefficients of the unknown influencing factors are corrected according to the actual distribution of the unknown influencing factors.

[0079] Step 5: Input the known influencing factors at the time to be predicted into the trained electricity load prediction model, and calculate the predicted electricity load value at the time to be predicted.

[0080] Specifically, for the time t to be predicted, input the feature variables corresponding to the known influencing factors at that time. The weights of known influencing factors in the trained electricity load prediction model are determined. Random error term Coefficients of unknown influencing factors Based on the statistical regularity of the unknown influencing factor U(t), the predicted value of the electricity load at the time t to be predicted is calculated using Equation (6):

[0081] Equation (6)

[0082] in These are values ​​of unknown factors predicted based on historical electricity load data patterns (such as the mean or 95% confidence interval).

[0083] The model can be optimized in the following ways:

[0084] 1. Adjust the coefficients of known factors: correct them based on practical experience. (If the impact of temperature is found to be underestimated, manually increase the temperature influence factor.) );

[0085] 2. Correcting the patterns of unknown factors: If U(t) exhibits a non-normal distribution (e.g., a sudden surge in load), the distribution assumption of U(t) can be adjusted (e.g., switching to a Poisson distribution) or corrected. (For example, increasing the standard deviation to cover extreme values).

[0086] In summary, the semi-normalized electricity load regression prediction method provided in this application has the following technical effects:

[0087] 1. Reduced data length requirements: Only 3-6 months of data are needed to train a high-precision model. Traditional methods require 1-3 years of historical data, which is suitable for data missing, new scenarios, or rapid update needs, and can achieve small sample learning.

[0088] 2. Explicitly integrate known patterns: By pre-defining known factors such as week, month, and temperature, the model can actively capture the influence of these patterns, avoiding the problem of insufficient implicit learning in small data scenarios, and achieving a significant improvement in prediction accuracy compared to traditional methods.

[0089] 3. Enhanced flexibility and interpretability: The weights of known influencing factors can be manually adjusted, while unknown influencing factors can be explicitly displayed through residual analysis. The model retains the adaptability of machine learning and the interpretability of statistical models, making it easy to optimize according to actual scenarios.

[0090] Example 2:

[0091] Based on the same inventive concept as the semi-normalized electricity load regression prediction method in Embodiment 1, this application provides a semi-normalized electricity load regression prediction system, such as... Figure 2 As shown, the system includes:

[0092] Feature module: Explicitly defines and collects known influencing factors affecting electricity load, with each known influencing factor corresponding to a quantifiable feature variable;

[0093] Model building module: Constructs an electricity load prediction model. The electricity load prediction model calculates the predicted electricity load value by weighting and summing each known influencing factor, and combining the self-inductive unknown influencing factors, the coefficients of the unknown influencing factors, and the random error term.

[0094] First training module: Perform the first step of training, based on historical electricity load data and historical known influencing factors, to train the weights of each known influencing factor and the random error term in the electricity load prediction model;

[0095] The second training module: Execute the second training step, extract the residuals based on the results of the first training step, train the coefficients of unknown influencing factors in the electricity load prediction model, and summarize the statistical laws of unknown influencing factors.

[0096] Prediction module: Input the known influencing factors at the time to be predicted into the trained electricity load prediction model, and calculate the predicted value of the electricity load at the time to be predicted.

[0097] Through the foregoing detailed description of a semi-normalized electricity load regression prediction method, those skilled in the art can clearly understand that this embodiment provides a semi-normalized electricity load regression prediction system. Since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant details can be found in the method section.

[0098] Example 3:

[0099] In Embodiment 3, a computer device is provided, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the semi-normalized electricity load regression prediction method described above.

[0100] Example 4:

[0101] In Embodiment 4, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the above-described semi-normalized electricity load regression prediction method.

[0102] Example 5:

[0103] In Embodiment 5, a computer program product is provided, including a computer program or instructions, which, when executed by a processor, implement the steps of the aforementioned semi-normalized electricity load regression prediction method.

[0104] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0105] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A semi-normalized electrical load regression prediction method, characterized by, The method comprises the following steps: Step 1: explicitly defining and collecting known influencing factors affecting power load, each known influencing factor corresponding to a quantifiable characteristic variable, the known influencing factors including factors with periodic regularity and external environmental characteristic factors; Step 2: constructing a power load prediction model, the power load prediction model calculating a power load prediction value by weighted summation of each known influencing factor, combined with self-inductive unknown influencing factors, coefficients of unknown influencing factors and random error terms; Step 3: performing first-step training, training weights of each known influencing factor and random error terms in the power load prediction model based on historical power load data and historical known influencing factors; Step 4: performing second-step training, extracting residuals based on the results of the first-step training, training coefficients of unknown influencing factors in the power load prediction model, and inducting statistical rules of unknown influencing factors; Step 5: inputting known influencing factors at a to-be-predicted time into the trained power load prediction model, and calculating a power load prediction value at the to-be-predicted time; The step 2 specifically comprises: constructing a power load prediction model in the form of formula (1): Formula (1) wherein, represents the weight of the i-th known influencing factor, represents the i-th known influencing factor at time t, and m represents the number of known influencing factors, represents the coefficient of the unknown influencing factor, represents the unknown influencing factor, represents a random error term, represents the predicted value of the power load at time t; The step 3 specifically comprises the following steps: Step 3.1: initializing weights of each known influencing factor in the power load prediction model; Step 3.2: degrading the power load prediction model into the form of formula (2): Formula (2) Step 3.3: Based on the historical electricity load data and the historical known influencing factors, use formula (3) as the optimization objective, and solve the weights of the known influencing factors by the least square method and a random error term to minimize the mean square error between the electricity load calculated by formula (2) and the historical electricity load data. Formula (3) wherein, represents the historical electricity consumption load at time t, and n represents the number of samples of the historical electricity consumption load data. The step 4 specifically comprises the following steps: Step 4.1: Calculate the weight of the known influencing factors based on the solved weights and a random error term , the residual is calculated in the manner of equation (4): Formula (4) wherein represents a residual error; Step 4.2: using formula (5) as an optimization objective, solving coefficients of unknown influencing factors by minimizing errors between residuals and products of unknown influencing factors and the coefficients: Formula (5) Among them, unknown influencing factors They follow a preset distribution, which includes the normal distribution. Or a Poisson distribution, wherein the preset distribution is based on historical residuals. The actual distribution characteristics are determined, and the coefficients of unknown influencing factors are determined. The standard deviation of the residuals; Step 4.3: Extracting the unknown influencing factor from the residual error by dividing the coefficient of the unknown influencing factor Step 4.4: Extracting the unknown influencing factor Step 4.5: Inducing the statistical law of the unknown influencing factor including the mean, variance and time autocorrelation of the unknown influencing factor, for predicting the value of the unknown influencing factor at the to-be-predicted time.

2. The semi-normalized electrical load regression forecasting method of claim 1, wherein, The weights of the known influencing factors are adjusted according to actual load scenarios, and the distribution of the unknown influencing factors and the coefficients of the unknown influencing factors are corrected according to actual distributions of unknown influencing factors.

3. A semi-normalized electrical load regression forecasting system, comprising: The system comprises: a characteristic module that explicitly defines and collects known influencing factors affecting power load, each known influencing factor corresponding to a quantifiable characteristic variable; a model construction module that constructs a power load prediction model, the power load prediction model calculating a power load prediction value by weighted summation of each known influencing factor, combined with self-inductive unknown influencing factors, coefficients of unknown influencing factors and random error terms; a first training module that performs first-step training, training weights of each known influencing factor and random error terms in the power load prediction model based on historical power load data and historical known influencing factors; a second training module that performs second-step training, extracting residuals based on the results of the first-step training, training coefficients of unknown influencing factors in the power load prediction model, and inducting statistical rules of unknown influencing factors; a prediction module that inputs known influencing factors at a to-be-predicted time into the trained power load prediction model, and calculates a power load prediction value at the to-be-predicted time; The step 2 specifically comprises: constructing a power load prediction model in the form of formula (1): Formula (1) wherein, represents the weight of the i-th known influencing factor, represents the i-th known influencing factor at time t, and m represents the number of known influencing factors, represents the coefficient of the unknown influencing factor, represents the unknown influencing factor, represents a random error term, represents the predicted value of the power load at time t; The step 3 specifically comprises the following steps: Step 3.1: initializing weights of each known influencing factor in the power load prediction model; Step 3.2: Degenerate the electricity load forecasting model into the form of equation (2): formula (2) Step 3.3: Based on the historical electricity load data and the historical known influencing factors, use formula (3) as the optimization objective, and solve the weights of the known influencing factors by the least square method and a random error term to minimize the mean square error between the electricity load calculated by formula (2) and the historical electricity load data: Formula (3) wherein, represents the historical power consumption load at time t, and n represents the number of samples of the historical power consumption load data. The step 4, specifically comprising the following steps: Step 4.1: Calculate the weight of the known influencing factors based on the solved weights and a random error term , the residual is calculated in the manner of equation (4): formula (4) wherein represents a residual error; Step 4.2: Solve the coefficient of unknown influencing factors by minimizing the error between the residual and the product of the coefficient of unknown influencing factors and unknown influencing factors using equation (5) as the optimization objective: Formula (5) Among them, unknown influencing factors They follow a preset distribution, which includes the normal distribution. Or a Poisson distribution, wherein the preset distribution is based on historical residuals. The actual distribution characteristics are determined, and the coefficients of unknown influencing factors are determined. The standard deviation of the residuals; Step 4.3: By making the residual Divide by the coefficient of unknown influencing factors Extracting unknown influencing factors And summarize the unknown influencing factors. The statistical regularities, including the mean, variance, and time autocorrelation of unknown influencing factors, are used to predict the value of unknown influencing factors at the time to be predicted.

4. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to implement the steps of the semi-normalized electricity load regression forecasting method of any one of claims 1 to 2.

5. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the semi-normalized electricity load regression forecasting method of any one of claims 1 to 2.

6. A computer program product comprising computer programs or instructions, characterized in that, The computer program or instructions are executed by the processor to implement the steps of the semi-normalized electricity load regression forecasting method of any one of claims 1 to 2.

Images