Electric quantity monitoring service system based on data analysis

Abstract

The invention discloses an electric quantity monitoring service system based on data analysis, and relates to the technical field of electric quantity service. The electric quantity monitoring service system based on data analysis comprises the following system processes: S1, carrying out analysis and application service on timeliness data by a daily rolling average line; s2, a statistical method of curve decomposition and an application analysis service; s3, performing concavity and convexity judgment service on a prediction result cost function by the data offset; s4, data classification is carried out according to a mathematical statistical algorithm, and when a new user is accessed, data classification can be carried out timely according to recent historical data; s5, customer electric quantity change multi-angle factor analysis service; s6, data governance analysis and application service; s7, performing data preprocessing and electricity price analysis according to an evolutionary programming algorithm; s8, performing power prediction service by using a particle swarm algorithm; and S9, carrying out load aggregation, disassembling the power transaction data into a plurality of components such as trend, period and seasonality, and predicting the change trend of future electric quantity.

Description

Technical Field

[0001] The present invention relates to the technical field of electric quantity service, and in particular to an electric quantity monitoring service system based on data analysis. Background Art

[0002] In the current spot power trading system environment, real-time data processing is required from power plant power generation, inter-provincial power transmission and transformation channels, to customer power consumption. With the help of power historical data and real-time market data analysis technology, potential modes and laws can be found from the massive data of the power system, which is a comprehensive application to provide decision support for traders. Through data management, data planning, and data forecasting for medium- and long-term transactions, spot transactions, and green power inter-provincial channel transactions, scientific methods can be normalized and deeply integrated with various transaction types to provide support capabilities. The existing power detection service system has unreasonable problems in power planning, which can easily lead to supply and demand contradictions and there is a certain deviation in the accuracy of the data during the same period. Summary of the invention

[0003] The purpose of the present invention is to solve at least one of the technical problems existing in the prior art and to provide an electric quantity monitoring service system based on data analysis, which can solve the above-mentioned problem.

[0004] To achieve the above purpose, the present invention provides the following technical solutions: an electric power monitoring service system based on data analysis, including the following system processes, S1 daily rolling average analysis and application services for timeliness data, S1 is divided into 5+2 averages and holiday averages;

[0005] The statistical method and application analysis service of S2 curve decomposition uses the moving average method and Fourier analysis method for curve decomposition statistical analysis. These methods can decompose a time series data into multiple components to better understand and analyze its characteristics. The moving average method is a commonly used curve decomposition method. It smoothes the data by calculating the average value of the data in a certain time window before and after each time point, which can remove noise and instantaneous fluctuations and obtain the trend component of the data. Fourier analysis is a method that decomposes a time series data into multiple frequency components. It uses Fourier transform to convert time domain signals into frequency domain signals, and can obtain components of different frequencies in the data, which is very helpful for analyzing periodic changes;

[0006] S3 data deviation determines the convexity of the prediction result cost function;

[0007] S4 classifies data based on mathematical statistical algorithms. When a new user is connected, data can be classified in a timely manner based on recent historical data.

[0008] S5 multi-angle factor analysis service for customer power consumption changes;

[0009] S6 data governance analysis and application services;

[0010] The data input of the power system usually leads to data input errors or data loss due to network transmission delays, sensor failures or power outages, resulting in abnormalities in certain input points. Data optimization is required for these abnormal data points. If the power data shows a linear trend or changes smoothly, a linear interpolation method is needed to fill in the missing values. If the power data has a nonlinear trend or changes more complexly, it is necessary to consider using the spline interpolation method. Spline interpolation optimizes the data into a series of smooth curve segments, and each curve segment is optimized between adjacent data points. If the power data has time correlation and seasonal changes, it is necessary to use time series analysis methods for optimization and prediction. Time series analysis includes model selection, parameter estimation, model diagnosis and prediction steps;

[0011] S7 performs data preprocessing and electricity price analysis based on evolutionary programming algorithms;

[0012] S8 uses particle swarm algorithm for power forecasting services;

[0013] S9 load aggregation.

[0014] Preferably, how to quickly screen the versatility of customers when there are many customers and the customer data types are complex is a big problem. In this process, the periodicity of customers can be quickly screened out and the recent electricity consumption trend of customers can be determined. Based on this demand, the 5+2 moving average can be introduced. The 5+2 moving average needs to process the customer's working day first to determine whether the customer has periodicity, and then determine Saturdays and Sundays. In this process, not only can it be determined whether there is periodicity, but also the user's electricity consumption trend can be quickly determined based on the electricity consumption on working days, Saturdays and Sundays;

[0015] Among them, the holiday moving average, in the medium and long term and spot trading, holidays are different from working days. When calculating the average line, it is necessary to exclude these non-working day data and form independent holiday data to obtain the electricity information during the holiday period. The electricity during all holidays is calculated to obtain the value of the holiday moving average. The values ​​of these holiday moving averages are plotted into a chart to observe the trend and changes of electricity during the holidays. At the same time, the holiday moving average is compared with the moving averages of other time periods to understand the market activity during the holidays relative to other periods.

[0016] Preferably, the application analysis of curve decomposition mainly includes the following aspects:

[0017] Trend analysis: By decomposing the trend components, we can study the long-term trend of the data and its changing patterns, which is very helpful for trend analysis and prediction in the fields of economy and market.

[0018] Periodicity analysis: By decomposing the frequency components, we can study the periodic changes of data, which is very helpful for the study and prediction of periodic changes such as weather and economic seasonality;

[0019] Noise filtering: By decomposing the detailed components, the noise and instantaneous fluctuations in the data can be removed to obtain a cleaner data signal;

[0020] Feature extraction: The local features of the data can be extracted by decomposing the scale components, which can be used in signal processing, image processing and other fields.

[0021] Preferably, S3 evaluates the performance and optimization effect of the prediction model by analyzing whether the cost function of the prediction result is concave or convex due to the deviation of the data. First, the data needs to be normalized and the original data needs to be processed to have a certain symmetry and linear relationship. Then, the deviation of the data is evaluated by calculating the deviation index of the data, such as variance, skewness, kurtosis, etc. The deviation reflects the deviation degree and shape characteristics of the data distribution. A smaller deviation indicates that the data distribution is relatively uniform and symmetrical, and a larger deviation indicates that the data distribution has a larger asymmetry or fluctuation.

[0022] The cost function can be constructed according to the difference between the predicted results and the actual observed values ​​to measure the performance of the prediction model. The cost function can be mean square error (Mean Squared Error), cross entropy (Cross Entropy), etc. Then, the optimization effect of the model is determined by analyzing the convexity of the cost function. The convex cost function has a global optimal value, which can make it easier to find the optimal solution. For convex functions, the optimization algorithm can guarantee convergence to the global optimal solution, while for non-convex functions, it may converge to the local optimal solution.

[0023] Preferably, the S4 stability classification

[0024] In view of the different electricity consumption of users, the stability of customers is classified using the outlier judgment method. Standard deviation is a commonly used indicator in statistics to measure the degree of dispersion of a data set. It measures the degree of deviation between the observed value and the average value of the data set. They are divided into stable users, stable users, and unstable users. The prediction of stable users has basically achieved the expected results. A deeper data analysis and adjustment will be carried out for stable and unstable users.

[0025] Among them, (1) according to the user's 24-hour time-sharing electricity consumption, a single user's monthly overall, monthly time-sharing, single-hour monthly summary and other related data are established for comprehensive comparison, and judgment and classification are made based on the results

[0026] (2) Determine whether each user is a stable user based on the standard deviation ratio and the maximum deviation ratio observed for each user.

[0027] Standard deviation calculation formula:

[0028] σ=sqrt(((x1-x)^2+(x2-x)^2+...(xn-x)^2) / n+α+β+θ)+θ

[0029] Standard Deviation Ratio Calculation Formula:

[0030] S=sqrt(σ^2 / (x1+x2+...+xn)*θ)

[0031] The maximum difference calculation formula is:

[0032] c=sqrt((xmax-xmin) / (x1+x2+...+xn) / (n+β)+θ)+θ

[0033] The formula for calculating the maximum difference ratio is:

[0034] S=sqrt(c^2 / (x1+x2+...+xn)*θ)

[0035] Coefficient calculation formula:

[0036] α=(Σ(xn-xmin)(ymax-ymin)) / (Σ(xmax-xmin)^2+θ)

[0037] β=ymax-α*xmax

[0038] Periodic classification

[0039] Based on the periodic classification of users' electricity consumption patterns, users are divided into periodic users and non-periodic users through mathematical algorithm models.

[0040] Periodic users: For periodic users, that is, users who use electricity according to periodic demand, their electricity consumption patterns can be found out according to their specific periodic demand. It is worth noting that different types of periodic users have different electricity consumption patterns, such as industrial users, agricultural users, etc. The specific electricity consumption patterns need to be analyzed and described in combination with the actual situation and user characteristics. In this case, we will conduct contemporaneous data forecasts for users to ensure the accuracy of the data.

[0041] Non-periodic users: The electricity consumption of non-periodic users may vary, and the specific electricity consumption patterns need to be analyzed and studied in combination with actual conditions and user characteristics.

[0042] Preferably, the S5 is subdivided into temperature factors

[0043] Temperature changes have a significant impact on the power system, mainly in the following aspects:

[0044] Cooling load: Rising temperatures will increase people's demand for air conditioning, especially in the hot summer. Extensive use of air conditioning will lead to an increase in electricity consumption, placing higher demands on the power supply capacity of the power grid. Therefore, accurate prediction of temperature changes can reasonably allocate electricity in the power system to meet people's demand for electricity.

[0045] Heating load: In cold seasons, the drop in temperature will increase the demand for heating equipment, especially in northern regions. The increase in heating load will lead to an increase in electricity consumption, so accurately predicting the change in electricity consumption caused by temperature changes can help power companies better dispatch and plan.

[0046] Holiday Factors

[0047] Holidays have a certain impact on power consumption forecast. The specific influencing factors are as follows:

[0048] Residential electricity consumption: Holidays are a time when people tend to travel, visit relatives and friends, etc., which results in relatively low household electricity consumption. Compared with normal working days, household electricity consumption may decrease during holidays.

[0049] Industrial and commercial electricity consumption: Holidays have a dual impact on industrial and commercial electricity consumption. On the one hand, some industries or enterprises and institutions will stop working or take holidays during holidays, resulting in a decrease in industrial and commercial electricity consumption; on the other hand, some commercial areas or tourist attractions will have a large number of people during holidays, and commercial electricity consumption may increase.

[0050] Holidays have a certain impact on power forecasts, but the specific impact varies depending on factors such as holiday type, region, and industry. The power forecast during holidays is adjusted based on historical data and real-time demand to ensure the accuracy of the predicted power.

[0051] Work plan factors

[0052] The work plan has a certain impact on the power forecast. The specific influencing factors are as follows:

[0053] Working days and non-working days: Generally speaking, the power consumption on working days is higher because enterprises, institutions and office buildings are more active during this period. In contrast, the power consumption on non-working days (such as weekends or holidays) is lower because most workplaces are in a resting state.

[0054] The work plan has a certain impact on the power consumption forecast. The power consumption on different working days and non-working days is predicted and adjusted to ensure the stability and reliability of the forecast.

[0055] In view of the above factors, we focus on analyzing the changes in customer electricity consumption and build corresponding algorithm models to solve such problems, thereby improving the accuracy of electricity consumption forecasts.

[0056] Preferably, the S7 programming algorithm is an evolving process, which gradually selects better solutions from the initial set of candidate solutions. The algorithm uses a set of candidate solutions (also called individuals), generates new individuals generation by generation by performing operations such as mutation, crossover and selection on individuals, and selects individuals with higher fitness according to certain evaluation criteria, thereby gradually improving the quality of the solution.

[0057] Mutation is the only operation that brings changes to the EP population, so it is very important to consider the trade-off between exploration and exploitation capabilities in the design of mutation operators. The process of mutation requires sufficient exploration capabilities in the early stage to ensure that the search space can be covered as much as possible; after the initial exploration, individuals should use relevant search information to fine-tune the solution.

[0058] Individual phenotype X: X = (x1, x2, ┉┉, xn, σ1, σ2, ┉┉σn), where x1┉┉xn is the component of individual performance, and σ1┉┉σn is the variation parameter of the individual performance component;

[0059] Select N individuals from 2N individuals: The method uses the random q-competition method. The selection strategy of the evolutionary programming algorithm uses the q-competition mechanism, which is also the biggest difference from the evolutionary strategy algorithm. In simple terms, q-competition is to select high-quality solutions while accepting poor solutions with a certain random probability. Most of them are good solutions, and a few are relatively poor solutions. Together, they form a group of parents to prepare for the next evolution. Some formulas involved (such as Figure 2 );

[0060] When using evolutionary programming algorithms for data preprocessing and electricity price analysis, it is usually necessary to set a fitness function to evaluate the quality of the preprocessing results, select appropriate evolutionary operations, set parameters such as the number of iterations and population size, and optimize the algorithm to improve the effect and efficiency of preprocessing. It should be noted that different data preprocessing tasks may require the selection of different evolutionary programming algorithms and corresponding parameter settings. In addition, the performance of the algorithm may be affected by the complexity of the characteristic problems of the data set. Therefore, when applying evolutionary programming algorithms for data preprocessing and electricity price analysis, adjustments and optimizations should be made according to specific problems to achieve the best results.

[0061] During this process, three developers, two data processing personnel and two people responsible for adjusting the coefficient of variation are required.

[0062] Preferably, the S8 particle swarm algorithm can effectively find out which part of the customer data appears more frequently, and the range of customer data can be adjusted based on this frequency. In this algorithm process, all particles have a fitness value determined by the optimized function, and each particle also has a speed that determines the direction and distance they fly. Then the particles follow the current optimal particle to search in the solution space, and then find the optimal solution through iteration. In each iteration, the particles update themselves by tracking two "extreme values". In the process of updating themselves, the most accurate result of the prediction can be determined. At this time, 2 developers, 2 data processing personnel and 3 personnel are needed to track the particle situation and adjust the parameters in time.

[0063] The calculation model is (see Figure 3 )

[0064] y=x(1)^2+x(2)^2-10*cos(2*pi*x(1))-10*cos(2*pi*x(2))+20.

[0065] Preferably, S9 is subdivided into exploratory data analysis

[0066] Exploratory data analysis refers to a method of discovering the structure and regularity of data through descriptive statistics, visualization, feature calculation, equation fitting, etc., without prior assumptions or with very few assumptions. In traditional statistical analysis, we always make a priori judgment on the distribution characteristics of the data, and then estimate the characteristics of the population based on some characteristics of the sample. However, in daily life, the distribution of data often does not perfectly meet the requirements of statistics, which will bring certain deviations to our application.

[0067] Therefore, the data analysis process is divided into two stages: the exploration stage and the verification stage. In the exploration stage, we focus on discovering the models or rules hidden behind the data; in the verification stage, we focus on verifying whether the model we found in the data exploration stage is correct.

[0068] Exploratory data analysis can discover some characteristics hidden behind the data, so as to better model the data;

[0069] Application of prior and posterior under conjugate distribution

[0070] Conjugate distribution is an important concept in probability theory and statistics. In Bayesian statistics, if the posterior distribution and the prior distribution belong to the same distribution family, and the product of the prior distribution and the likelihood function also belongs to the distribution family, then this distribution family is called the conjugate prior distribution of the likelihood function, or conjugate distribution for short.

[0071] The importance of conjugate distribution lies in that it simplifies the calculation process of Bayesian inference. In this process, we can analyze the data and then complete algebraic operations to verify the predicted data. When the data deviation is too large, we need to re-analyze the impact of the data weight. The greater the impact, the more adjustments are needed to the prior data. To be precise, both prior data and posterior data are to assist in determining whether the results of data prediction meet the expected goals, such as Figure 4 As shown;

[0072] In Bayesian analysis, with priors and likelihoods, we can Figure 5 As shown;

[0073] Load application under differential evolution algorithm

[0074] Differential evolution algorithm is a random search algorithm used to solve global optimization problems. Compared with genetic algorithm, particle swarm optimization algorithm, ant colony algorithm, artificial bee colony algorithm and other intelligent algorithms, DE algorithm has fewer parameters and stronger optimization ability. Its purpose is to process abnormal load data while being able to determine the use range for valid data, and then bring these valid data into the algorithm for load analysis and prediction, and then bring the results into the conjugate distribution for data verification. Therefore, whether it is data judgment, differential prediction or conjugate data verification, it is to be able to screen accurate data while achieving the best prediction effect. In this process, 3 developers, 2 data analysts and 2 data verifiers are required;

[0075] Each individual in the algorithm's initialization population is a solution within the search range. The individuals in the population need to be distributed as much as possible in all areas within the search range so that the algorithm can perform large-scale optimization and increase the probability of finding the optimal value (such as Figure 6 shown).

[0076] Compared with the prior art, the present invention has the following beneficial effects:

[0077] (1) The power monitoring service system based on data analysis also adopts the curve decomposition method mentioned above. The curve decomposition for real-time transactions mainly analyzes the 96-point power curve of the day. The power is adjusted and allocated through the curve of the day, and the power transaction data is decomposed into multiple components such as trend, cycle, seasonality, etc., and the change trend of future power is predicted. This helps to rationally plan and dispatch power resources and reduce the contradiction between supply and demand.

[0078] (2) The power monitoring service system based on data analysis determines the optimization effect of the model by analyzing the convexity of the cost function. The convex cost function has a global optimal value, which makes it easier to find the optimal solution. For convex functions, the optimization algorithm can guarantee convergence to the global optimal solution, while for non-convex functions, it may converge to the local optimal solution.

[0079] (3) The power monitoring service system based on data analysis can decompose the power transaction data into multiple components such as trend, cycle, seasonality, etc., and predict the future trend of power consumption. This helps to rationally plan and dispatch power resources and reduce the contradiction between supply and demand.

[0080] (4) The electricity consumption monitoring service system based on data analysis has different types of periodic users with different electricity consumption patterns, such as industrial users, agricultural users, etc. The specific electricity consumption patterns need to be analyzed and described in combination with the actual situation and user characteristics. In this case, we will conduct data forecasts for the users in the same period to ensure the accuracy of the data.

[0081] (5) In the power monitoring service system based on data analysis, each individual in the algorithm initialization population is a solution within the search range. The individuals in the population need to be distributed as much as possible in all areas within the search range so that the algorithm can perform large-scale optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] The present invention is further described below in conjunction with the accompanying drawings and embodiments:

[0083] Figure 1 It is the double grey model flow chart of the present invention;

[0084] Figure 2 It is a schematic diagram of the calculation formula of the present invention;

[0085] Figure 3 It is a schematic diagram of the calculation model of the present invention;

[0086] Figure 4 It is a schematic diagram of the calculation formula of the present invention;

[0087] Figure 5 It is a schematic diagram of the calculation formula of the present invention;

[0088] Figure 6 It is a schematic diagram of the calculation formula of the present invention. DETAILED DESCRIPTION

[0089] This section will describe in detail the specific embodiments of the present invention. The preferred embodiments of the present invention are shown in the accompanying drawings. The purpose of the accompanying drawings is to supplement the description of the text part of the specification with graphics, so that people can intuitively and vividly understand each technical feature and the overall technical solution of the present invention, but it cannot be understood as a limitation on the scope of protection of the present invention.

[0090] In the description of the present invention, it should be understood that descriptions involving orientations, such as up, down, front, back, left, right, etc., and orientations or positional relationships indicated are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present invention.

[0091] In the description of the present invention, "greater than", "less than", "exceed" etc. are understood as not including the number itself, and "above", "below", "within" etc. are understood as including the number itself. If there is a description of "first" or "second", it is only used for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0092] In the description of the present invention, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention based on the specific content of the technical solution.

[0093] The present invention provides a technical solution: a power monitoring service system based on data analysis, including the following system processes: S1 daily rolling average analysis and application service for timeliness data

[0094] S1 is divided into 5+2 moving average and holiday moving average

[0095] How to quickly screen the commonality of customers when there are many customers and their data types are complex is a big problem. In this process, we can first quickly screen out the periodicity of customers and determine their recent electricity consumption trends. Based on this demand, we can introduce the 5+2 moving average. The 5+2 moving average needs to process the customer's working day first to determine whether the customer has periodicity, and then determine Saturdays and Sundays. In this process, we can not only determine whether there is periodicity, but also quickly determine the user's electricity consumption trend based on the electricity consumption on working days, Saturdays and Sundays.

[0096] Among them, the holiday moving average, in the medium and long term and spot trading, holidays will be different from working days. When calculating the average line, it is necessary to exclude these non-working day data and form independent holiday data to obtain the electricity information during the holiday period. Calculate the electricity during all holidays to get the value of the holiday moving average. Plot the values ​​of these holiday moving averages into a chart to observe the trend and changes of electricity during the holidays. At the same time, compare the holiday moving average with the moving average of other time periods to understand the market activity during the holidays relative to other periods;

[0097] Statistical methods and application analysis services for S2 curve decomposition

[0098] Moving average method and Fourier analysis method are used for curve decomposition statistical analysis. These methods can decompose a time series data into multiple components in order to better understand and analyze its characteristics. Moving average method is a commonly used curve decomposition method. It smoothes the data by calculating the average value of the data in a certain time window before and after each time point, which can remove noise and instantaneous fluctuations and obtain the trend component of the data. Fourier analysis is a method that decomposes a time series data into multiple frequency components. It uses Fourier transform to convert time domain signals into frequency domain signals, and can obtain components of different frequencies in the data, which is very helpful for analyzing periodic changes.

[0099] Among them, the application analysis of curve decomposition mainly includes the following aspects:

[0100] Trend analysis: By decomposing the trend components, we can study the long-term trend of the data and its changing patterns, which is very helpful for trend analysis and prediction in the fields of economy and market.

[0101] Periodicity analysis: By decomposing the frequency components, we can study the periodic changes of data, which is very helpful for the study and prediction of periodic changes such as weather and economic seasonality;

[0102] Noise filtering: By decomposing the detailed components, the noise and instantaneous fluctuations in the data can be removed to obtain a cleaner data signal;

[0103] Feature extraction: The local features of the data can be extracted by decomposing the scale components, which can be used in signal processing, image processing and other fields;

[0104] In summary, curve decomposition can break down power trading data into multiple components such as trends, cycles, and seasonality, and predict future trends in power consumption. This helps to rationally plan and dispatch power resources and reduce supply-demand contradictions.

[0105] S3 data deviation service for judging the convexity of prediction result cost function

[0106] The performance and optimization effect of the prediction model can be evaluated by analyzing whether the data deviation affects the cost function of the prediction result. First, the data needs to be normalized and the original data needs to be processed to have a certain symmetry and linear relationship. Then, the data deviation is evaluated by calculating the deviation indicators of the data, such as variance, skewness, kurtosis, etc. The deviation reflects the deviation degree and shape characteristics of the data distribution. A smaller deviation indicates that the data distribution is relatively uniform and symmetrical, and a larger deviation indicates that the data distribution has a larger asymmetry or fluctuation.

[0107] The cost function can be constructed according to the difference between the predicted results and the actual observed values ​​to measure the performance of the prediction model. The cost function can be mean square error (Mean Squared Error), cross entropy (Cross Entropy), etc. Then, the optimization effect of the model is determined by analyzing the convexity of the cost function. The convex cost function has a global optimal value, which can make it easier to find the optimal solution. For convex functions, the optimization algorithm can guarantee convergence to the global optimal solution, while for non-convex functions, it may converge to the local optimal solution.

[0108] S4 classifies data based on mathematical statistical algorithms. When a new user is connected, data can be classified in a timely manner based on recent historical data.

[0109] Stability classification

[0110] In view of the different electricity consumption of users, the stability of customers is classified using the outlier judgment method. Standard deviation is a commonly used indicator in statistics to measure the degree of dispersion of a data set. It measures the degree of deviation between the observed value and the average value of the data set. They are divided into stable users, stable users, and unstable users. The prediction of stable users has basically achieved the expected results. A deeper data analysis and adjustment will be carried out for stable and unstable users.

[0111] Among them, (1) according to the user's 24-hour time-sharing electricity consumption, a single user's monthly overall, monthly time-sharing, single-hour monthly summary and other related data are established for comprehensive comparison, and judgment and classification are made based on the results

[0112] (2) Determine whether each user is a stable user based on the standard deviation ratio and the maximum deviation ratio observed for each user.

[0113] Standard deviation calculation formula:

[0114] σ=sqrt(((x1-x)^2+(x2-x)^2+...(xn-x)^2) / n+α+β+θ)+θ

[0115] Standard Deviation Ratio Calculation Formula:

[0116] S=sqrt(σ^2 / (x1+x2+...+xn)*θ)

[0117] The maximum difference calculation formula is:

[0118] c=sqrt((xmax-xmin) / (x1+x2+...+xn) / (n+β)+θ)+θ

[0119] The formula for calculating the maximum difference ratio is:

[0120] S=sqrt(c^2 / (x1+x2+...+xn)*θ)

[0121] Coefficient calculation formula:

[0122] α=(Σ(xn-xmin)(ymax-ymin)) / (Σ(xmax-xmin)^2+θ)

[0123] β=ymax-α*xmax

[0124] Periodic classification

[0125] Based on the periodic classification of users' electricity consumption patterns, users are divided into periodic users and non-periodic users through mathematical algorithm models.

[0126] Periodic users: For periodic users, that is, users who use electricity according to periodic demand, their electricity consumption patterns can be found out according to their specific periodic demand. It is worth noting that different types of periodic users have different electricity consumption patterns, such as industrial users, agricultural users, etc. The specific electricity consumption patterns need to be analyzed and described in combination with the actual situation and user characteristics. In this case, we will conduct contemporaneous data forecasts for users to ensure the accuracy of the data.

[0127] Non-periodic users: The electricity consumption of non-periodic users may vary, and the specific electricity consumption patterns need to be analyzed and studied in combination with actual conditions and user characteristics.

[0128] S5 Customer power consumption change multi-angle factor analysis service

[0129] Temperature Factor

[0130] Temperature changes have a significant impact on the power system, mainly in the following aspects:

[0131] Cooling load: Rising temperatures will increase people's demand for air conditioning, especially in the hot summer. Extensive use of air conditioning will lead to an increase in electricity consumption, placing higher demands on the power supply capacity of the power grid. Therefore, accurate prediction of temperature changes can reasonably allocate electricity in the power system to meet people's demand for electricity.

[0132] Heating load: In cold seasons, the drop in temperature will increase the demand for heating equipment, especially in northern regions. The increase in heating load will lead to an increase in electricity consumption, so accurately predicting the change in electricity consumption caused by temperature changes can help power companies better dispatch and plan.

[0133] Holiday Factors

[0134] Holidays have a certain impact on power consumption forecast. The specific influencing factors are as follows:

[0135] Residential electricity consumption: Holidays are a time when people tend to travel, visit relatives and friends, etc., which results in relatively low household electricity consumption. Compared with normal working days, household electricity consumption may decrease during holidays.

[0136] Industrial and commercial electricity consumption: Holidays have a dual impact on industrial and commercial electricity consumption. On the one hand, some industries or enterprises and institutions will stop working or take holidays during holidays, resulting in a decrease in industrial and commercial electricity consumption; on the other hand, some commercial areas or tourist attractions will have a large number of people during holidays, and commercial electricity consumption may increase.

[0137] Holidays have a certain impact on power forecasts, but the specific impact varies depending on factors such as holiday type, region, and industry. The power forecast during holidays is adjusted based on historical data and real-time demand to ensure the accuracy of the predicted power.

[0138] Work plan factors

[0139] The work plan has a certain impact on the power forecast. The specific influencing factors are as follows:

[0140] Working days and non-working days: Generally speaking, the power consumption on working days is higher because enterprises, institutions and office buildings are more active during this period. In contrast, the power consumption on non-working days (such as weekends or holidays) is lower because most workplaces are in a resting state.

[0141] The work plan has a certain impact on the power consumption forecast. The power consumption on different working days and non-working days is predicted and adjusted to ensure the stability and reliability of the forecast.

[0142] In view of the above factors, we focus on analyzing the changes in customer electricity consumption and build corresponding algorithm models to solve such problems, thereby improving the accuracy of electricity consumption forecasts.

[0143] S6 Data Governance Analysis and Application Services

[0144] The data input of the power system usually leads to data input errors or data loss due to network transmission delays, sensor failures or power outages, resulting in abnormalities in some points of input. Data optimization is required for these abnormal data points. If the power data shows a linear trend or changes smoothly, a linear interpolation method is needed to fill in the missing values. If the power data has a nonlinear trend or changes more complexly, it is necessary to consider using the spline interpolation method. Spline interpolation optimizes the data into a series of smooth curve segments, and each curve segment is optimized between adjacent data points. If the power data has time correlation and seasonal changes, it is necessary to use time series analysis methods for optimization and prediction. Time series analysis includes model selection, parameter estimation, model diagnosis and prediction.

[0145] S7 performs data preprocessing and electricity price analysis based on evolutionary programming algorithm

[0146] See also Figure 2 The optimization planning algorithm is an evolving process that gradually selects better solutions from the initial set of candidate solutions. The algorithm uses a set of candidate solutions (also called individuals) to generate new individuals generation by generation by performing operations such as mutation, crossover and selection on individuals, and selects individuals with higher fitness according to certain evaluation criteria, thereby gradually improving the quality of the solution.

[0147] Mutation is the only operation that brings changes to the EP population, so it is very important to consider the trade-off between exploration and exploitation capabilities in the design of mutation operators. The process of mutation requires sufficient exploration capabilities in the early stage to ensure that the search space can be covered as much as possible; after the initial exploration, individuals should use relevant search information to fine-tune the solution.

[0148] Individual phenotype X: X = (x1, x2, ┉┉, xn, σ1, σ2, ┉┉σn), where x1┉┉xn is the component of individual performance, and σ1┉┉σn is the variation parameter of the individual performance component;

[0149] Select N individuals from 2N individuals: The method uses the random q-competition method. The selection strategy of the evolutionary programming algorithm uses the q-competition mechanism, which is also the biggest difference from the evolutionary strategy algorithm. In simple terms, q-competition is to select high-quality solutions while accepting poor solutions with a certain random probability. Most of them are good solutions, and a few are relatively poor solutions. Together, they form a group of parents to prepare for the next evolution. Some formulas involved (such as Figure 2 );

[0150] When using evolutionary programming algorithms for data preprocessing and electricity price analysis, it is usually necessary to set a fitness function to evaluate the quality of the preprocessing results, select appropriate evolutionary operations, set parameters such as the number of iterations and population size, and optimize the algorithm to improve the effect and efficiency of preprocessing. It should be noted that different data preprocessing tasks may require the selection of different evolutionary programming algorithms and corresponding parameter settings. In addition, the performance of the algorithm may be affected by the complexity of the characteristic problems of the data set. Therefore, when applying evolutionary programming algorithms for data preprocessing and electricity price analysis, adjustments and optimizations should be made according to specific problems to achieve the best results.

[0151] During this process, three developers, two data processing personnel and two people responsible for adjusting the coefficient of variation are required.

[0152] S8 uses particle swarm algorithm for power forecasting service

[0153] The particle swarm algorithm can effectively find out which part of the customer data appears more frequently, and based on this frequency, the range of customer data can be adjusted. In this algorithm process, all particles have a fitness value determined by the optimized function, and each particle also has a speed that determines the direction and distance they fly. Then the particles follow the current optimal particle to search in the solution space, and then find the optimal solution through iteration. In each iteration, the particles update themselves by tracking two "extreme values". In the process of updating themselves, the most accurate prediction result can be achieved. At this time, 2 developers, 2 data processing personnel, and 3 people who track the particle situation and adjust the parameters in time are needed.

[0154] The calculation model is (see Figure 3 )

[0155] y=x(1)^2+x(2)^2-10*cos(2*pi*x(1))-10*cos(2*pi*x(2))+20

[0156] S9 Load Aggregation (See Figure 4-6 )

[0157] Exploratory Data Analysis

[0158] Exploratory data analysis refers to a method of discovering the structure and regularity of data through descriptive statistics, visualization, feature calculation, equation fitting, etc., without prior assumptions or with very few assumptions. In traditional statistical analysis, we always make a priori judgments on the distribution characteristics of the data, and then estimate the characteristics of the population based on some characteristics of the sample. However, in daily life, the distribution of data often does not perfectly meet the requirements of statistics. This will bring certain deviations to our application.

[0159] Therefore, the data analysis process is divided into two stages: the exploration stage and the verification stage. In the exploration stage, the focus is on discovering the models or rules hidden behind the data; in the verification stage, the focus is on verifying whether the model we found in the data exploration stage is correct.

[0160] Exploratory data analysis can discover some characteristics hidden behind the data, so as to better model the data.

[0161] Application of prior and posterior under conjugate distribution

[0162] Conjugate distribution is an important concept in probability theory and statistics. In Bayesian statistics, if the posterior distribution and the prior distribution belong to the same distribution family, and the product of the prior distribution and the likelihood function also belongs to this distribution family, then this distribution family is called the conjugate prior distribution of the likelihood function, or simply the conjugate distribution;

[0163] The importance of conjugate distribution is that it simplifies the calculation process of Bayesian inference. In this process, we can analyze the data and then complete the algebraic operation to verify the predicted data. When the data deviation is too large, we need to re-analyze the impact of the data weight. The greater the impact, the more adjustments are needed to the prior data. To be precise, both prior data and posterior data are to assist in whether the results of data prediction meet the expected goals, such as Figure 4 As shown;

[0164] In Bayesian analysis, with priors and likelihoods, we can Figure 5 As shown;

[0165] Load application under differential evolution algorithm

[0166] Differential evolution algorithm is a random search algorithm used to solve global optimization problems.

[0167] Compared with intelligent algorithms such as particle swarm optimization algorithm, ant colony algorithm, artificial bee colony algorithm, etc., DE algorithm has fewer parameters and stronger optimization ability. The purpose is to be able to determine the use range for valid data while processing abnormal load data, and then bring these valid data into the algorithm for load analysis and prediction, and then bring the results into the conjugate distribution for data verification. Therefore, whether it is data judgment, differential prediction or conjugate data verification, it is to be able to screen accurate data while achieving the best prediction effect. In this process, 3 developers, 2 data analysts and 2 data verifiers are required;

[0168] Each individual in the algorithm's initialization population is a solution within the search range. The individuals in the population need to be distributed as much as possible in all areas within the search range so that the algorithm can perform large-scale optimization and increase the probability of finding the optimal value (such as Figure 6 shown).

[0169] See also Figure 1 , double grey prediction algorithm

[0170] Grey prediction identifies the difference in the development trends between data factors, that is, conducts correlation analysis, and performs grey generation on the original data to find the law of data changes, generate a data sequence with strong regularity, and then establish a corresponding differential equation model to predict the future development trend of things. Since the law is found through data, it has a wide range of applications. A general prediction model can be established to first find out customers who are sensitive to periodicity or holidays and industries. On the basis of establishing a general model, considering that some customers cannot be applied to the general model due to fluctuations, at this time, the customer can be modeled separately, the customer's trend can be analyzed, and the customer's hidden classification can be found from the data. This process requires not only 3 developers, 2 data analysts, but also 2 parameter adjustment personnel.

[0171] Growth rate prediction algorithm

[0172] In the customer electricity model, there are not only periodic customers, but also non-periodic customers. For example, the customer's electricity consumption continues to increase or decrease over a period of time. At this time, due to the latency of the data, it is impossible to timely judge and adjust the trend of the customer's future electricity consumption, so a growth rate correction is needed to quickly determine the algorithm of the customer's continuous changes. The advantage of the algorithm is that it can well judge the electricity consumption trend for applicable customers and can greatly reduce the prediction deviation rate. It also has disadvantages. It can be quickly applied to periodic customers, but for non-periodic and non-continuously changing customers, it will cause the prediction trend to be unable to determine whether it is an upward trend or a downward trend, but it is still an applicable algorithm. Therefore, in this process, 2 developers, 2 data analysts, and 2 parameter adjustment and verification personnel are first required.

[0173] The embodiments of the present invention are described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Various changes can be made within the knowledge scope of ordinary technicians in the technical field without departing from the purpose of the present invention.

Claims

1. A power monitoring service system based on data analysis, characterized in that: Including the following system processes: S1 daily rolling average analysis and application services for time-sensitive data, S1 is divided into 5+2 moving average and holiday moving average; The statistical method and application analysis service of S2 curve decomposition uses the moving average method and Fourier analysis method for curve decomposition statistical analysis. These methods can decompose a time series data into multiple components to better understand and analyze its characteristics. The moving average method is a commonly used curve decomposition method. It smoothes the data by calculating the average value of the data in a certain time window before and after each time point, which can remove noise and instantaneous fluctuations and obtain the trend component of the data. Fourier analysis is a method that decomposes a time series data into multiple frequency components. It uses Fourier transform to convert time domain signals into frequency domain signals, and can obtain components of different frequencies in the data, which is very helpful for analyzing periodic changes; S3 data deviation determines the convexity of the prediction result cost function; S4 classifies data based on mathematical statistical algorithms. When a new user is connected, data can be classified in a timely manner based on recent historical data. S5 multi-angle factor analysis service for customer power consumption changes; S6 data governance analysis and application services; The data input in the power system usually leads to data input errors or data loss due to network transmission delays, sensor failures or power outages, resulting in abnormal situations at certain input points. Data optimization is required for these abnormal data points. If the power data presents a linear trend or changes smoothly, a linear interpolation method is needed to fill in the missing values. If the power data has a nonlinear trend or complex changes, it is necessary to consider using the spline interpolation method. Spline interpolation optimizes the data into a series of smooth curve segments, and each curve segment is optimized between adjacent data points. If power data has time correlation and seasonal changes, time series analysis methods are needed for optimization and prediction. Time series analysis includes model selection, parameter estimation, model diagnosis and prediction. S7 performs data preprocessing and electricity price analysis based on evolutionary programming algorithms; S8 uses particle swarm algorithm for power forecasting services; S9 load aggregation.

2. The power monitoring service system based on data analysis according to claim 1 is characterized in that: How to quickly screen the commonality of customers when there are many customers and their data types are complex is a big problem. In this process, we can first quickly screen out the periodicity of customers and determine their recent electricity consumption trends. Based on this demand, we can introduce the 5+2 moving average. The 5+2 moving average needs to process the customer's working day first to determine whether the customer has periodicity, and then determine Saturdays and Sundays. In this process, we can not only determine whether there is periodicity, but also quickly determine the user's electricity consumption trend based on the electricity consumption on working days, Saturdays and Sundays. Among them, the holiday moving average, in the medium and long term and spot trading, holidays are different from working days. When calculating the average line, it is necessary to exclude these non-working day data and form independent holiday data to obtain the electricity information during the holiday period. The electricity during all holidays is calculated to obtain the value of the holiday moving average. The values ​​of these holiday moving averages are plotted into a chart to observe the trend and changes of electricity during the holidays. At the same time, the holiday moving average is compared with the moving averages of other time periods to understand the market activity during the holidays relative to other periods.

3. The power monitoring service system based on data analysis according to claim 1 is characterized in that: The application analysis of curve decomposition mainly includes the following aspects: Trend analysis: By decomposing the trend components, we can study the long-term trend of the data and its changing patterns, which is very helpful for trend analysis and prediction in the fields of economy and market. Periodicity analysis: By decomposing the frequency components, we can study the periodic changes of data, which is very helpful for the study and prediction of periodic changes such as weather and economic seasonality; Noise filtering: By decomposing the detailed components, the noise and instantaneous fluctuations in the data can be removed to obtain a cleaner data signal; Feature extraction: The local features of the data can be extracted by decomposing the scale components, which can be used in signal processing, image processing and other fields.

4. The power monitoring service system based on data analysis according to claim 1 is characterized in that: S3 evaluates the performance and optimization effect of the prediction model by analyzing whether the cost function of the prediction result is concave or convex due to the deviation of the data. First, the data needs to be normalized and the original data needs to be processed to have a certain symmetry and linear relationship. Then, the deviation of the data is evaluated by calculating the deviation indicators of the data, such as variance, skewness, kurtosis, etc. The deviation reflects the deviation degree and shape characteristics of the data distribution. A smaller deviation indicates that the data distribution is relatively uniform and symmetrical, and a larger deviation indicates that the data distribution has a larger asymmetry or fluctuation. The cost function can be constructed according to the difference between the predicted results and the actual observed values ​​to measure the performance of the prediction model. The cost function can be mean square error (Mean Squared Error), cross entropy (CrossEntropy), etc. Then, the optimization effect of the model is determined by analyzing the convexity of the cost function. The convex cost function has a global optimal value, which can make it easier to find the optimal solution. For convex functions, the optimization algorithm can guarantee convergence to the global optimal solution, while for non-convex functions, it may converge to the local optimal solution.

5. The power monitoring service system based on data analysis according to claim 1 is characterized in that: The S4 stability classification In view of the different electricity consumption of users, the stability of customers is classified using the outlier judgment method. Standard deviation is a commonly used indicator in statistics to measure the degree of dispersion of a data set. It measures the degree of deviation between the observed value and the average value of the data set. They are divided into stable users, stable users, and unstable users. The prediction of stable users has basically achieved the expected results. A deeper data analysis and adjustment will be carried out for stable and unstable users. Among them, (1) according to the user's 24-hour time-sharing electricity consumption, a single user's monthly overall, monthly time-sharing, single-hour monthly summary and other related data are established for comprehensive comparison, and judgment and classification are made based on the results (2) Determine whether a user is stable based on the standard deviation ratio and the maximum deviation ratio of each user. Standard deviation calculation formula: σ=sqrt(((x1-x)^2+(x2-x)^2+...(xn-x)^2) / n+α+β+θ)+θ Standard Deviation Ratio Calculation Formula: S=sqrt(σ^2 / (x1+x2+...+xn)*θ) The maximum difference calculation formula is: c=sqrt((xmax-xmin) / (x1+x2+...+xn) / (n+β)+θ)+θ The formula for calculating the maximum difference ratio is: S=sqrt(c^2 / (x1+x2+...+xn)*θ) Coefficient calculation formula: α=(Σ(xn-xmin)(ymax-ymin)) / (Σ(xmax-xmin)^2+θ) β=ymax-α*xmax Periodic classification Based on the periodic classification of users' electricity consumption patterns, users are divided into periodic users and non-periodic users through mathematical algorithm models. Periodic users: For periodic users, that is, users who use electricity according to periodic demand, their electricity consumption patterns can be found out according to their specific periodic demand. It is worth noting that different types of periodic users have different electricity consumption patterns, such as industrial users, agricultural users, etc. The specific electricity consumption patterns need to be analyzed and described in combination with the actual situation and user characteristics. In this case, we will conduct contemporaneous data forecasts for users to ensure the accuracy of the data. Non-periodic users: The electricity consumption of non-periodic users may vary, and the specific electricity consumption patterns need to be analyzed and studied in combination with actual conditions and user characteristics.

6. The power monitoring service system based on data analysis according to claim 1 is characterized in that: The S5 is subdivided into temperature factors Temperature changes have a significant impact on the power system, mainly in the following aspects: Cooling load: Rising temperatures will increase people's demand for air conditioning, especially in the hot summer. Extensive use of air conditioning will lead to an increase in electricity consumption, placing higher demands on the power supply capacity of the power grid. Therefore, accurate prediction of temperature changes can reasonably allocate electricity in the power system to meet people's demand for electricity. Heating load: In cold seasons, the drop in temperature will increase the demand for heating equipment, especially in northern regions. The increase in heating load will lead to an increase in electricity consumption, so accurately predicting the change in electricity consumption caused by temperature changes can help power companies better dispatch and plan. Holiday Factors Holidays have a certain impact on power consumption forecast. The specific influencing factors are as follows: Residential electricity consumption: Holidays are a time when people tend to travel, visit relatives and friends, etc., which results in relatively low household electricity consumption. Compared with normal working days, household electricity consumption may decrease during holidays. Industrial and commercial electricity consumption: Holidays have a dual impact on industrial and commercial electricity consumption. On the one hand, some industries or enterprises and institutions will stop working or take holidays during holidays, resulting in a decrease in industrial and commercial electricity consumption; on the other hand, some commercial areas or tourist attractions will have a large number of people during holidays, and commercial electricity consumption may increase. Holidays have a certain impact on power forecasts, but the specific impact varies depending on factors such as holiday type, region, and industry. The power forecast during holidays is adjusted based on historical data and real-time demand to ensure the accuracy of the predicted power. Work plan factors The work plan has a certain impact on the power forecast. The specific influencing factors are as follows: Working days and non-working days: Generally speaking, the power consumption on working days is higher because enterprises, institutions and office buildings are more active during this period. In contrast, the power consumption on non-working days (such as weekends or holidays) is lower because most workplaces are in a resting state. The work plan has a certain impact on the power consumption forecast. The power consumption on different working days and non-working days is predicted and adjusted to ensure the stability and reliability of the forecast. In view of the above factors, we focus on analyzing the changes in customer electricity consumption and build corresponding algorithm models to solve such problems, thereby improving the accuracy of electricity consumption forecasts.

7. The power monitoring service system based on data analysis according to claim 1 is characterized in that: The S7 programming algorithm is an evolving process that gradually selects better solutions from the initial set of candidate solutions. The algorithm uses a set of candidate solutions (also called individuals) to generate new individuals generation by generation by performing operations such as mutation, crossover and selection on individuals, and selects individuals with higher fitness according to certain evaluation criteria, thereby gradually improving the quality of the solution. Mutation is the only operation that brings changes to the EP population, so it is very important to consider the trade-off between exploration and exploitation capabilities in the design of mutation operators. The changing process requires sufficient exploration capability in the early stages to ensure that the largest possible search space can be covered; after the initial exploration, individuals should use relevant search information to fine-tune the solution. Individual phenotype X: X = (x1, x2, ┉┉, xn, σ1, σ2, ┉┉σn), where x1┉┉xn is the component of individual performance, and σ1┉┉σn is the variation parameter of the individual performance component; Select N individuals from 2N individuals: The method uses the random q-competition method. The selection strategy of the evolutionary programming algorithm uses the q-competition mechanism, which is also the biggest difference from the evolutionary strategy algorithm. In simple terms, q-competition is to select high-quality solutions while accepting poor solutions with a certain random probability. Most of them are excellent solutions, and a few are relatively poor solutions. Together, they form a group of parents to prepare for the next evolution. Some formulas are involved (as shown in Figure 2); When using evolutionary programming algorithms for data preprocessing and electricity price analysis, it is usually necessary to set a fitness function to evaluate the quality of the preprocessing results, select appropriate evolutionary operations, set parameters such as the number of iterations and population size, and optimize the algorithm to improve the effect and efficiency of preprocessing. It should be noted that different data preprocessing tasks may require the selection of different evolutionary programming algorithms and corresponding parameter settings. In addition, the performance of the algorithm may be affected by the complexity of the characteristic problems of the data set. Therefore, when applying evolutionary programming algorithms for data preprocessing and electricity price analysis, adjustments and optimizations should be made according to specific problems to achieve the best results. During this process, three developers, two data processing personnel and two people responsible for adjusting the coefficient of variation are required.

8. The power monitoring service system based on data analysis according to claim 1 is characterized in that: The S8 particle swarm algorithm can effectively find out which part of the customer data appears more frequently, and the range of customer data can be adjusted based on this frequency. In this algorithm process, all particles have a fitness value determined by the optimized function, and each particle also has a speed that determines the direction and distance they fly. Then the particles follow the current optimal particle to search in the solution space, and then find the optimal solution through iteration. In each iteration, the particles update themselves by tracking two "extreme values". In the process of updating themselves, the most accurate prediction result can be achieved. At this time, 2 developers, 2 data processing personnel and 3 personnel are needed to track the particle situation and adjust the parameters in time. The calculation model is (see Figure 3) y=x(1)^2+x(2)^2-10*cos(2*pi*x(1))-10*cos(2*pi*x(2))+20.

9. The power monitoring service system based on data analysis according to claim 1, characterized in that: S9 is subdivided into exploratory data analysis Exploratory data analysis refers to a method of discovering the structure and regularity of data through descriptive statistics, visualization, feature calculation, equation fitting, etc., without prior assumptions or with very few assumptions. In traditional statistical analysis, we always make a priori judgment on the distribution characteristics of the data, and then estimate the characteristics of the population based on some characteristics of the sample. However, in daily life, the distribution of data often does not perfectly meet the requirements of statistics, which will bring certain deviations to our application. Therefore, the data analysis process is divided into two stages: the exploration stage and the verification stage. In the exploration stage, we focus on discovering the models or rules hidden behind the data; in the verification stage, we focus on verifying whether the model we found in the data exploration stage is correct. Exploratory data analysis can discover some characteristics hidden behind the data, so as to better model the data; Application of prior and posterior under conjugate distribution Conjugate distribution is an important concept in probability theory and statistics. In Bayesian statistics, if the posterior distribution and the prior distribution belong to the same distribution family, and the product of the prior distribution and the likelihood function also belongs to the distribution family, then this distribution family is called the conjugate prior distribution of the likelihood function, or conjugate distribution for short. The importance of conjugate distribution lies in that it simplifies the calculation process of Bayesian inference. In this process, we can analyze the data and then complete algebraic operations to verify the predicted data. When the data deviation is too large, we need to re-analyze the impact of the data weight. The greater the impact, the more adjustments are needed to the prior data. To be precise, both prior data and posterior data are to assist in whether the results of data prediction meet the expected goals, as shown in Figure 4; In Bayesian analysis, with priors and likelihoods, we have the following as shown in Figure 5; Load application under differential evolution algorithm Differential evolution algorithm is a random search algorithm used to solve global optimization problems. Compared with genetic algorithm, particle swarm optimization algorithm, ant colony algorithm, artificial bee colony algorithm and other intelligent algorithms, DE algorithm has fewer parameters and stronger optimization ability. Its purpose is to process abnormal load data while being able to determine the use range for valid data, and then bring these valid data into the algorithm for load analysis and prediction, and then bring the results into the conjugate distribution for data verification. Therefore, whether it is data judgment, differential prediction or conjugate data verification, it is to be able to screen accurate data while achieving the best prediction effect. In this process, 3 developers, 2 data analysts and 2 data verifiers are required; Each individual in the algorithm's initialized population is a solution within the search range. The individuals in the population need to be distributed as much as possible in all areas within the search range so that the algorithm can perform large-scale optimization and increase the probability of finding the optimal value (as shown in Figure 6).

Images