A load early warning method based on seasonal load hierarchical analysis

Abstract

The present invention relates to a load warning method based on seasonal load hierarchical analysis, which belongs to the field of load warning technology and solves the problem of relatively low load warning accuracy caused by the lack of load prediction and warning methods for different main transformer access scenarios in the prior art. The method includes: obtaining the scenario of main transformer access load characteristics, and for the scenario where the main transformer access load characteristics are fuzzy, using a seasonal comprehensive hierarchical load forecasting method to perform load forecasting to obtain the load forecast results for the next year; for the scenario where the main transformer access load characteristics are clear and the load data can be obtained independently, using a seasonal multivariate load forecasting method to perform load forecasting on each category of loads respectively to obtain the load forecast results of various categories of loads in the next year; performing cluster analysis on the load forecast results for the next year to obtain cluster analysis results; and making a main transformer extreme load warning judgment based on the cluster analysis results to obtain a warning result.

Description

Technical Field

[0001] The present invention relates to the technical field of load early warning, and in particular to a load early warning method based on seasonal load hierarchical analysis. Background Art

[0002] In modern society, with the rapid development of industrialization and urbanization, the continuous increase in power load has led to unstable voltage, unsafe grid operation, and insufficient power supply capacity. Severe consequences will cause large-scale power outages in the power grid and even cause the grid to collapse. Therefore, load forecasting and load warning have become important tasks in power system management.

[0003] Traditional methods of monitoring power grid loads often only consider the statistical characteristics of historical data, ignoring the diversification and dynamics of loads. In actual applications, the scenarios of main transformer access load characteristics include situations where the main transformer access load characteristics are vague or clear, and the data that can be obtained in different scenarios are different, which increases the difficulty of implementing load forecasting and load warning. Therefore, how to better deal with extreme load conditions that may occur in the distribution network and ensure the accuracy of load forecasting and warning is a technical problem that needs to be solved urgently. Summary of the invention

[0004] In view of the above analysis, an embodiment of the present invention aims to provide a load warning method based on seasonal load hierarchical analysis to solve the problem of relatively low load warning accuracy caused by the lack of load prediction and warning methods for different main transformer access scenarios in the prior art.

[0005] The present invention discloses a load early warning method based on seasonal load hierarchical analysis, the method comprising:

[0006] Obtain the load characteristics of the main transformer access. For scenarios where the load characteristics of the main transformer access are fuzzy, use the seasonal comprehensive hierarchical load forecasting method to perform load forecasting and obtain the load forecast results for the next year. For scenarios where the load characteristics of the main transformer access are clear and the load data can be obtained independently, use the seasonal multivariate load forecasting method to perform load forecasting for each category of load, and obtain the load forecast results for each category of load in the next year.

[0007] Performing cluster analysis on the load forecast result for the next year to obtain cluster analysis results;

[0008] According to the cluster analysis results, the extreme load early warning judgment of the main transformer is carried out to obtain the early warning result.

[0009] On the basis of the above scheme, the present invention also makes the following improvements:

[0010] Furthermore, the seasonal comprehensive hierarchical load forecasting method is used to perform load forecasting, and the following steps are performed:

[0011] Collect the historical load data of the previous year when the main transformer is connected, and decompose the historical load data of the previous year into historical benchmark load data and historical fluctuating load data according to the fluctuation of historical load data in different seasons;

[0012] Using a time series method to perform load forecasting on the historical benchmark load data to obtain forecast benchmark load data for the next year;

[0013] Using a multi-factor load forecasting method and combining the historical meteorological data of the previous year, the load forecasting of the historical fluctuating load data is performed to obtain the predicted fluctuating load data of the next year;

[0014] The predicted baseline load data and predicted fluctuating load data for the same day and the same period of the next year are superimposed to obtain the load forecast results for the corresponding period of the corresponding day of the next year, and finally the load forecast results for the next year are obtained.

[0015] Furthermore, the historical load data of the previous year is decomposed into historical benchmark load data and historical fluctuating load data, and the following steps are performed:

[0016] Calculate the average values ​​of typical daily load curves in spring and autumn in the previous year's historical load data in each period as the load data for the corresponding period in the historical benchmark load data;

[0017] The maximum difference between the summer and winter load curves in the previous year's historical load data and the historical benchmark load data in the corresponding time period is calculated respectively as the load data for the corresponding time period in the historical fluctuating load data.

[0018] Furthermore, the seasonal multivariate load forecasting method is used to perform load forecasting, and the following steps are performed:

[0019] According to the load characteristics of the main transformer, determine the type of multiple loads, and establish forecast models that take seasonality into consideration for each type of load;

[0020] By using the forecasting model that takes seasonality into consideration for various types of loads, the load forecast results for the corresponding types of loads in the next year are obtained.

[0021] Furthermore, the multi-type load is divided into output type load, input type load, and input-output type load; wherein,

[0022] Output loads are distributed power sources, including photovoltaic power generation, hydropower generation, wind power generation and biomass power generation;

[0023] Input loads include pure load users such as charging stations, residents, businesses, and industries;

[0024] The input and output load is energy storage.

[0025] Further, for the scenario where the load characteristics of the main transformer are fuzzy, a cluster analysis is performed on the load forecast results for the next year, and the cluster analysis results are obtained, and the following is executed:

[0026] Divide the load forecast results for the next year by season to obtain the load forecast results for four seasons;

[0027] For each season, the load forecast results of each natural day in the load forecast results of the season are clustered as samples to form several clusters, and the 24-hour load forecast maximum value curve of each cluster is established, and the occurrence probability of each cluster is calculated; from the 24-hour load forecast maximum value curve of all clusters, the load forecast maximum value corresponding to each time period and its corresponding cluster are selected respectively; the load forecast maximum value corresponding to each time period is connected to obtain the maximum load curve of the corresponding season;

[0028] The maximum load curves of the four seasons are summarized as the cluster analysis results.

[0029] Furthermore, for scenarios where the load characteristics of the main transformer are fuzzy, the extreme load warning of the main transformer is judged based on the cluster analysis results, and the following are executed:

[0030] The cluster analysis results are compared with the main transformer judgment threshold. If the cluster analysis results do not contain the maximum load forecast value that exceeds the main transformer judgment threshold, no warning is issued.

[0031] Otherwise, for each season, each period in the maximum load curve that exceeds the main transformer judgment threshold is used as an early warning investigation period, and the clusters corresponding to each early warning investigation period are obtained; samples that exceed the main transformer judgment threshold in the early warning investigation period are screened out from the clusters corresponding to each early warning investigation period, and the load forecast results of the samples in the early warning investigation period are used as early warning data, the date of the screened samples and the early warning investigation period are used as the early warning time, and the occurrence probability of the cluster is used as the early warning probability; the early warning data, warning time and warning probability are output as early warning results.

[0032] Furthermore, for the scenario where the load characteristics of the main transformer access are clear and the load data can be independently obtained, a cluster analysis is performed on the load forecast results for the next year to obtain the cluster analysis results, and the following is executed:

[0033] Divide the load forecast results for the next year by season to obtain the load forecast results for four seasons;

[0034] For each category of load in each season, the load forecast results of each natural day in the load forecast results of the category of load in the season are clustered as samples to form several clusters, and the 24-hour load forecast maximum value curve of each cluster is established, and the occurrence probability of each cluster is calculated; the load forecast maximum value corresponding to each time period and its corresponding cluster are selected respectively; the load forecast maximum value corresponding to each time period is connected to obtain the maximum load curve of the corresponding category of load in the corresponding season;

[0035] The maximum load curves of all types of loads in the corresponding season are superimposed. In each period, the occurrence probabilities of the clusters corresponding to the maximum load forecast values ​​of various types of loads are multiplied and the maximum load forecast values ​​are added to obtain the maximum load curve of the main transformer in the corresponding season.

[0036] The maximum value curves of multivariate loads of main transformers in four seasons are summarized as the cluster analysis results.

[0037] Furthermore, for scenarios where the load characteristics of the main transformer are clear and the load data can be obtained independently, the extreme load warning of the main transformer is judged based on the cluster analysis results, and the following are executed:

[0038] Compare the cluster analysis results with the main transformer judgment threshold. If the cluster analysis results do not contain the maximum value of the main transformer multi-load that exceeds the main transformer judgment threshold, no warning will be issued.

[0039] Otherwise, for each season, each period in the maximum value curve of the main transformer multivariate load that exceeds the main transformer judgment threshold is used as an early warning investigation period, and the combination of clusters of various types of loads corresponding to each early warning investigation period is obtained;

[0040] The load forecast values ​​of the samples of different categories of loads in each intersection date involved in the combination of clusters of each category of loads corresponding to each early warning investigation period are added together, and the load forecast result whose sum of the load forecast values ​​exceeds the main transformer judgment threshold is used as the early warning data, the corresponding intersection date and early warning investigation period are used as the early warning time, and the probability of occurrence of the cluster combination is used as the early warning probability; the early warning data, early warning time and early warning probability are output as the early warning result.

[0041] Furthermore, the K-means clustering method is used to perform cluster analysis on the load forecast results for the next year.

[0042] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0043] The load early warning method based on seasonal load hierarchical analysis provided by the present invention can provide corresponding load forecasting methods according to the access load characteristics in different scenarios, thereby obtaining more accurate load forecasting results, and determine the time period when the maximum load occurs through cluster analysis, and make subsequent early warning judgments based on this, and finally obtain load early warning results. Therefore, the present invention proposes corresponding load forecasting and early warning methods for different main transformer access scenarios, so that the accuracy of load early warning is effectively improved, which is of great significance to ensuring the safe and stable operation of the distribution network.

[0044] In the present invention, the above-mentioned technical solutions can also be combined with each other to achieve more preferred combination solutions. Other features and advantages of the present invention will be described in the subsequent description, and some advantages can become obvious from the description, or can be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings are only used for the purpose of illustrating specific embodiments and are not to be considered as limiting the present invention. In the entire drawings, the same reference symbols represent the same components;

[0046] Figure 1 A flow chart of a load early warning method based on seasonal load hierarchical analysis provided by an embodiment of the present invention;

[0047] Figure 2 The cluster result curve provided by the embodiment of the present invention;

[0048] Figure 3 The maximum value curve of multi-element load of the main transformer provided by the embodiment of the present invention;

[0049] Figure 4 An extreme load alarm judgment diagram provided by an embodiment of the present invention;

[0050] Figure 5 The 2022 benchmark current load curve provided by the embodiment of the present invention;

[0051] Figure 6 The 2023 benchmark load forecast curve provided by the embodiment of the present invention;

[0052] Figure 7 A schematic diagram of a decomposed fluctuating load provided by an embodiment of the present invention;

[0053] Figure 8 A load fluctuation prediction curve of the electric load provided by an embodiment of the present invention;

[0054] Fig. 9A total load forecast curve for seasonal comprehensive hierarchical load forecasting provided by an embodiment of the present invention;

[0055] Fig.10 A result graph of cluster generation for seasonal comprehensive hierarchical load forecasting provided by an embodiment of the present invention;

[0056] Fig.11 A variety of cluster typical curves for seasonal comprehensive hierarchical load forecasting provided by an embodiment of the present invention;

[0057] Fig.12 A result graph of clustering generation of industrial loads provided by an embodiment of the present invention;

[0058] Fig.13 Typical curves of various clusters of industrial loads provided by embodiments of the present invention;

[0059] Fig.14 A result graph of clustering generation of biomass load provided by an embodiment of the present invention;

[0060] Fig.15 Typical curves of various clusters of biomass loading provided for embodiments of the present invention;

[0061] Fig.16 The industrial load maximum value curve provided by the embodiment of the present invention;

[0062] Fig.17 The maximum biomass load curve provided by the embodiment of the present invention;

[0063] Fig.18 This is a maximum value curve of the multi-element load of the main transformer provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0065] In a specific embodiment of the present invention, considering that the loads in various regions are often affected by factors such as seasonality and weather conditions, a load early warning method based on hierarchical analysis of seasonal loads is proposed. The flow chart is as follows: Figure 1 The specific steps are described as follows.

[0066] Step S1: Obtain the scenario of the load characteristics of the main transformer access. For the scenario where the load characteristics of the main transformer access are fuzzy, use the seasonal comprehensive hierarchical load forecasting method to perform load forecasting and obtain the load forecast results for the next year; For the scenario where the load characteristics of the main transformer access are clear and the load data can be obtained independently, use the seasonal multivariate load forecasting method to perform load forecasting on each category of loads and obtain the load forecast results;

[0067] Step S2: performing cluster analysis on the load forecast result for the next year to obtain cluster analysis results;

[0068] Step S3: Perform a warning judgment on the extreme load of the main transformer according to the cluster analysis result to obtain a warning result.

[0069] In step S1, specifically perform:

[0070] Step S11: For the scenario where the load characteristics of the main transformer are fuzzy, a seasonal comprehensive hierarchical load forecasting method is used to perform load forecasting to obtain a load forecasting result.

[0071] Below, the specific implementation process of the seasonal comprehensive stratified load forecasting method is described in detail.

[0072] Step S111: Collect the historical load data of the previous year when the main transformer is connected, and decompose the historical load data of the previous year into historical benchmark load data and historical fluctuating load data according to the fluctuation of the historical load data in different seasons.

[0073] Step S1111: Calculate the average values ​​of typical daily load curves in spring and autumn in the historical load data of the previous year in each time period as the load data of the corresponding time period in the historical benchmark load data.

[0074] 1) Select representative months for spring and autumn.

[0075] 2) Assume that the load curve for each day in the representative month of spring is ;here Indicates the day, Indicates hours .

[0076] 3) Select the load curve for a typical day in a representative month in spring, and use Indicates that here Indicates a typical day.

[0077] 4) Use the average of multiple typical daily load curves to represent the typical load curve in spring , the specific expression is as follows:

[0078] (1)

[0079] In the formula, Represents the number of typical days.

[0080] 5) Similarly, the typical daily load curve in autumn can be obtained: .

[0081] 6) The average value of the spring and autumn load curves is used as the load curve for non-fluctuating load electricity consumption in the current year (i.e. historical benchmark load data) to eliminate the impact of natural load growth.

[0082] (2)

[0083] Step S1112: Calculate the maximum difference between the summer and winter load curves in the historical load data of the previous year and the historical benchmark load data in the corresponding time period as the load data for the corresponding time period in the historical fluctuating load data.

[0084] That is, the difference between the load curve of each day in summer and winter and the reference curve is the fluctuation load curve of the day, and the maximum value of the fluctuation load curve is the maximum fluctuation load of the day. The specific process is as follows:

[0085] 7) Summer and winter loads , The difference between the load curve without load fluctuation and the load curve without load fluctuation is the load curve affected by temperature.

[0086] (3)

[0087] (4)

[0088] 8) Obtain the maximum fluctuating load in each time period.

[0089] 9) Take the largest value among the values ​​obtained in the previous step, which is the maximum fluctuating load in summer and winter of the year.

[0090] Step S112: Use the time series method to perform load forecasting on the historical benchmark load data to obtain forecast benchmark load data for the next year.

[0091] This embodiment uses the autoregressive moving average method (ARMA) in time series for load forecasting. The autoregressive moving average model consists of two parts: the autoregressive part and the moving average part, so it contains two orders and can be expressed as , is the autoregressive order, is the moving average order, and the regression equation is expressed as:

[0092] (5)

[0093] is the white noise of the time series, are the indicator values ​​recorded at different time points. is the autoregression coefficient.

[0094] In the specific implementation process, the autoregressive moving average model is used to model and predict the benchmark load forecast. The steps are as follows: receive the input historical benchmark load data, and use the autocorrelation and partial correlation functions, and combine the AIC or BIC criteria to perform pattern recognition and order determination on the established model; after completing the model recognition and order determination, enter the parameter estimation stage of the model; after completing the parameter estimation, perform an adaptability test on the model. If the fitting model passes the test, the prediction stage begins. If the model test fails, the model identification and test are performed again, and the model is reselected. Finally, a highly adaptable fitting model is used to predict the future trend of the sequence.

[0095] The autoregressive moving average model is an important method for studying time series. Since the model is a mixture of the autoregressive model and the moving average model, it has the characteristics of a wide range of applications and small prediction errors.

[0096] Step S113: adopting a multi-factor load forecasting method and combining the historical meteorological data of the previous year, load forecasting is performed on the historical fluctuating load data to obtain the predicted fluctuating load data for the next year.

[0097] In this embodiment, considering that the load is affected by seasonality, especially weather conditions, temperature and other factors, a load forecasting method that considers multiple factors is established. The model can highlight the impact of key influencing factors on the load. The core of the multi-factor load forecasting method is to analyze all influencing factors of the load, then obtain the key influencing factors through correlation analysis, and finally use multiple regression to establish a model. In this embodiment, meteorological factors and time factors and other load forecasting related factors are considered.

[0098] 1) Meteorological factors

[0099] Meteorological factors include precipitation, absolute humidity at two meters, relative humidity at two meters, surface pressure, temperature level at 2 meters, surface temperature, freezing point at 2 meters, maximum temperature at 2 meters, average temperature at 2 meters, wind level at 50 meters, wind level at 10 meters, minimum wind speed at 50 meters, minimum wind speed at 10 meters, maximum wind speed at 50 meters, maximum wind speed at 10 meters, wind direction at 50 meters, wind direction at 10 meters, wind speed at 50 meters and wind speed at 10 meters.

[0100] 2) Time Factor

[0101] Time factors include current date, current time, working days, weekends, and holidays.

[0102] In this embodiment, the multi-factor load forecasting method preferably uses the multivariate linear regression forecasting method to construct a multivariate load forecasting model. The modeling process is as follows: Data cleaning processing: Establish the relationship trend between time and fluctuating load, analyze the distribution of the maximum differentiated electricity demand, and find the distribution and abnormal data. Data processing: Non-dimensionalize the load data and its key influencing factor data. Multivariate regression modeling: Using the statistical method of multivariate linear regression, taking the fluctuating load as the dependent variable and other multidimensional related factors as independent variables, a multivariate linear regression model and a multivariate nonlinear regression model are established. Optimal load model selection: Select the determination coefficient R as the basis for selecting the optimal model, and select the model with the largest R square in the linear regression model and the multivariate nonlinear regression model as the load forecasting model.

[0103] The multiple regression analysis prediction method refers to a method of predicting by establishing a prediction model through correlation analysis of two or more independent variables and one dependent variable. It includes multiple linear regression and multiple nonlinear regression.

[0104] Step S14: superimpose the predicted baseline load data and the predicted fluctuating load data for the same day and the same period of the next year to obtain the load forecast result for the corresponding period of the corresponding day of the next year, and finally obtain the load forecast result for the next year.

[0105] In step S1, the following steps are also performed:

[0106] Step S12: For the scenario where the load characteristics of the main transformer access are clear and the load data can be obtained independently, a seasonal multivariate load forecasting method is used to perform load forecasting to obtain the load forecast result for the next year.

[0107] In this embodiment, the type of multiple loads is determined according to the load characteristics of the main transformer access, and prediction models considering seasonality are established for each type of load, and the prediction models considering seasonality for each type of load are used to realize the load prediction of the corresponding category of loads. Preferably, the multiple loads are divided into output loads, input loads, and input-output loads; wherein the output loads are distributed power sources, including photovoltaic power generation, hydropower generation, wind power generation and biomass power generation; the input loads include pure load users such as charging stations, residents, commerce, and industry; and the input-output loads are energy storage. Exemplarily, load prediction models for output loads (photovoltaic, biomass, charging stations) and input loads (industry, commerce, and residents) considering seasonality are established respectively.

[0108] (1) Photovoltaic output forecast

[0109] Photovoltaic output prediction adopts an output prediction method based on light intensity and photovoltaic properties.

[0110] Solar photovoltaic power generation refers to photovoltaic cells using the photovoltaic effect of semiconductor materials to convert solar energy into electrical energy. The power output of photovoltaic cells is mainly related to the intensity of solar radiation, the area of ​​the photovoltaic array, and the photoelectric conversion efficiency. The total power output of photovoltaic power generation is approximately:

[0111] (6)

[0112] In the formula, is the solar amplitude intensity; is the total area of ​​the PV array; is the overall conversion efficiency.

[0113] The output of photovoltaic power generation system is greatly affected by the light intensity. Therefore, the probability density function of light intensity in a certain period of time can be expressed as follows.

[0114] (7)

[0115] In the formula, , ; is the maximum radiation intensity; is the Gamma function.

[0116] use The probability distribution function of the light intensity is also used to obtain the probability density function of photovoltaic power output:

[0117] (8)

[0118] Then the active power of the entire photovoltaic system is:

[0119] (9)

[0120] In the formula, is the photovoltaic array area; For efficiency; is the maximum radiation intensity.

[0121] Similar to the treatment of wind power generation, photovoltaic power generation can also be regarded as a PQ node. In addition, since the direct output of photovoltaic power generation is in the form of direct current, in some cases it can be considered that only active power is input to the grid, and reactive power is not considered.

[0122] (2) Biomass output prediction

[0123] The biomass output prediction adopts the output prediction method based on the power generation mechanism.

[0124] Biomass energy refers to a form of energy that uses organic matter from plants, animals and other organisms as energy. The basic principle of biomass power generation is to convert the heat energy generated by biomass combustion into power, and then into electrical energy. When biomass burns, its chemical energy is released to generate heat, which can be used to heat water or steam to drive steam turbines or generators to generate electricity. This time, based on the basic method of calculating the efficiency of existing coal-fired power plants and combining the empirical data obtained from the performance test of biomass direct-fired power plants, a power generation output model was obtained. The basic principles are:

[0125] 1) Use the positive balance method (input-output heat method) as a reference;

[0126] 2) The air temperature at the blower inlet is used as the reference temperature for boiler thermal efficiency calculation, that is, the atmospheric temperature near the blower;

[0127] 3) Since the main purpose of this implementation is to calculate the boiler thermal efficiency under actual operating conditions, no correction is made.

[0128] Since most biomass power plants currently have no external heat source to heat the air and fuel mist steam, in order to simplify the calculation, the sensible heat of the furnace fuel is not included, and the basic low calorific value of the fuel is taken as the input heat, that is:

[0129] (10)

[0130] In the formula, It is the basic low calorific value of the fuel, kJ / kg.

[0131] Heat output:

[0132] The positive balance thermal efficiency calculation formula is:

[0133] (11)

[0134] In the formula, is the boiler thermal efficiency, %; is the input heat, kJ; is the heat output, kJ.

[0135] (12)

[0136] In the formula, is the fuel consumption, kg; is the main steam flow rate of the boiler, kg / h; is the outlet enthalpy of the main steam of the boiler, kJ / kg; is the boiler feed water enthalpy, kJ / kg; is the amount of wastewater discharged from the boiler, %; is the enthalpy of boiler wastewater, kJ / kg. Since the continuous and periodic wastewater discharge is very small, generally about 2% of the main steam flow, the boiler wastewater discharge is not considered to simplify the calculation. The steam and feed water enthalpy values ​​are realized by programming the general calculation model of water and steam thermodynamic properties IAPWS-IF97.

[0137] According to the thermal equivalent of thermal power generation in thermal power plants, 1kg of standard coal = 7000 kcal = 29301 kj. That is, the standard coal consumed by the heat conversion generated by biomass power generation is:

[0138] (13)

[0139] Where M is the standard coal consumed, kg.

[0140] The prediction formula of biomass power generation output is:

[0141] (14)

[0142] In the formula, The conversion efficiency of coal-fired power generation is generally between 35% and 38%.

[0143] (3) Charging station charging prediction

[0144] Charging stations use a charging prediction method based on the Monte Carlo method. Charging station model analysis refers to modeling and analyzing the operation and management of electric vehicle charging stations. The Monte Carlo method is used to sample the charging start time, battery charging capacity, and average charging power of an electric vehicle. Then, by superimposing the charging load of each electric vehicle in the area at a certain moment, the charging load prediction model of the electric vehicle can be obtained.

[0145] 1) Simplified charging behavior of electric vehicles

[0146] In order to quantitatively analyze the weekday charging load of large-scale electric vehicles, the charging behavior of electric vehicles is simplified as follows:

[0147] (a) The capacity of electric vehicle batteries is C = 25-35 kWh;

[0148] (b) The charging power Pcd of electric vehicles is evenly distributed within 3-5kW;

[0149] (c) Considering the loss of electric vehicle batteries and the gradual change of related parameters, it is assumed that the state of charge of the electric vehicle rechargeable battery is 10%-90%;

[0150] (d) Consider the distribution of charging load from 0:00 to 24:00 in a day, and the expected number of charging times per electric vehicle per day is 1;

[0151] (e) The peak electricity price is higher during the time-of-use electricity price period, and electric car users will try to choose to charge during the off-peak electricity price period.

[0152] 3) Disordered charging model

[0153] Without the implementation of the electricity price guidance policy, electric car users will go to work in the morning on weekdays, choose to charge at work after arriving at the company, and recharge the electric car, or charge at home after getting home from get off work in the afternoon. In the disorderly charging model of private car users, 20% of users have a charging start time that meets the , 50% of users start charging at the time that meets , and 30% of the users’ starting time meets U (0, 24). The probability distribution density of the starting charging time of electric vehicles with disordered charging is:

[0154] (15)

[0155] (16)

[0156] (17)

[0157] In the formula, .

[0158] Assuming that the electric car starts charging after the driving is finished, the end time of the car's driving is the start time of the electric car's charging, and the charging behavior is subject to the simplification made above. Based on the assumption of the travel habits of private car owners, the duration of battery charging can be obtained The calculation formula is:

[0159] (18)

[0160] In the formula, is the capacity of a randomly selected battery. ; is the randomly selected charging power kW; The charging duration of an electric vehicle, .

[0161] 4) Orderly charging model

[0162] In order to lower the cost of purchasing electricity, electric car users choose to respond to the time-of-use electricity price control strategy and start charging near the start time of the valley electricity price. The load of private car users who set the charging start time of the response time-of-use electricity price roughly follows Distribution, the probability distribution density of the starting charging time of electric vehicles in an orderly manner is:

[0163] (19)

[0164] In the formula, , .

[0165] Private car users who do not respond to time-of-use electricity prices still follow the disorderly charging model of electric vehicles. Suppose the proportion of electric vehicles that respond to time-of-use electricity prices to the total number of electric vehicles is ,but The distribution of the starting time of electric vehicle charging follows the orderly charging model under time-of-use electricity prices. The distribution of charging start time of electric vehicles follows the disordered charging model on weekdays, and the total charging load is:

[0166] (20)

[0167] Where: Represents the number of private cars that charge in an orderly manner under the time-of-use electricity price in the area in one day; Represents the extraction time t The charging power of one of the electric vehicles; Represents the number of private cars that are charged in an unordered manner during the working day in the area; Represents the extraction time t The charging power of one of the electric vehicles.

[0168] According to the established electric vehicle charging model, the Monte Carlo method is used to extract the electric vehicle starting charging time and charging power, calculate the charging duration, simulate the charging model, and obtain the electric vehicle charging load prediction curve.

[0169] (4) Large user load forecasting

[0170] The load types of large users in each region mainly include commercial load, industrial load and residential load. Among them, the load forecasting method of time series method is adopted for industrial large users, and the multivariate regression forecasting method considering temperature, GDP and time is adopted for commercial and residential large users.

[0171] 1) Load forecasting methods for large commercial and residential users

[0172] Commercial and residential users are greatly affected by seasonal fluctuations, so a multi-factor load forecasting method suitable for fluctuating loads is used to establish a forecasting model. According to the actual situation in a certain area, the correlation of load influencing factors is calculated, and the calculation results of the main load influencing factors are as follows:

[0173] Table 1 Comparison of the influence of load factors on large users in a certain area

[0174]

[0175] As shown in Table 1, the key influencing factors of commercial load are time, GDP, and temperature; the key influencing factors of residential load are temperature, GDP, and time. Then, the multivariate regression method is used to establish the commercial and residential large user load prediction model.

[0176] 2) Load forecasting method for large industrial users

[0177] The load of large industrial users is relatively stable. Its load is mainly affected by its production plan and has little seasonal impact. Therefore, the time series method suitable for stable load is used for load forecasting.

[0178] In step S2, a K-means clustering method is used to perform cluster analysis on the load forecasting results to obtain cluster analysis results.

[0179] For different scenarios, the implementation methods of step S2 and step S3 are different, as described below:

[0180] Specifically, for scenarios where the load characteristics of the main transformer are fuzzy:

[0181] Perform cluster analysis on the load forecast results for the next year and obtain the cluster analysis results. Execute:

[0182] (1) Divide the load forecast results for the next year by season to obtain the load forecast results for four seasons.

[0183] (2) For each season, the load forecast results of each natural day in the load forecast results of that season are clustered as samples into several clusters, and the 24-hour load forecast maximum value curve of each cluster is established, and the occurrence probability of each cluster is calculated; from the 24-hour load forecast maximum value curve of all clusters, the load forecast maximum value corresponding to each time period and its corresponding cluster are selected respectively; the load forecast maximum value corresponding to each time period is connected to obtain the maximum load curve of the corresponding season.

[0184] Preferably, in this embodiment, the 24-hour load forecast maximum value curve of each cluster is established in the following manner: for each time period in 24 hours, the load forecast maximum value of all samples in the cluster in that time period is calculated respectively, and the load forecast maximum values ​​of each time period in the cluster are connected in sequence to obtain the 24-hour load forecast maximum value curve of the cluster.

[0185] (3) The maximum load curves of the four seasons are summarized as the cluster analysis results.

[0186] According to the cluster analysis results, the extreme load early warning of the main transformer is judged and executed:

[0187] The cluster analysis results are compared with the main transformer judgment threshold. If the cluster analysis results do not contain the maximum load forecast value that exceeds the main transformer judgment threshold, no warning is issued.

[0188] Otherwise, for each season, each period in the maximum load curve that exceeds the main transformer judgment threshold is used as an early warning investigation period, and the clusters corresponding to each early warning investigation period are obtained; samples that exceed the main transformer judgment threshold in the early warning investigation period are screened out from the clusters corresponding to each early warning investigation period, and the load forecast results of the samples in the early warning investigation period are used as early warning data, the date of the screened samples and the early warning investigation period are used as the early warning time, and the occurrence probability of the cluster is used as the early warning probability; the early warning data, warning time and warning probability are output as early warning results.

[0189] For scenarios where the load characteristics of the main transformer are clear and the load data can be obtained independently:

[0190] Perform cluster analysis on the load forecast results for the next year and obtain the cluster analysis results. Execute:

[0191] (1) Divide the load forecast results for the next year by season to obtain the load forecast results for four seasons.

[0192] (2) For each category of load in each season, the load forecast results of each natural day in the load forecast results of the category of load in the season are clustered into several clusters, and the 24-hour load forecast maximum value curve of each cluster is established, and the occurrence probability of each cluster is calculated; the load forecast maximum value corresponding to each time period and its corresponding cluster are selected respectively; the load forecast maximum value corresponding to each time period is connected to obtain the maximum load curve of the corresponding category of load in the corresponding season.

[0193] (3) Superimpose the maximum load curves of all categories of loads in the corresponding season. In each period, multiply the occurrence probabilities of the clusters corresponding to the maximum load forecast values ​​of various categories of loads and add the maximum load forecast values ​​to obtain the maximum load curve of the main transformer in the corresponding season.

[0194] (4) The maximum value curves of the multivariate loads of the main transformers in the four seasons are summarized as the cluster analysis results.

[0195] According to the cluster analysis results, the extreme load early warning of the main transformer is judged and executed:

[0196] Compare the cluster analysis results with the main transformer judgment threshold. If the cluster analysis results do not contain the maximum value of the main transformer multi-load that exceeds the main transformer judgment threshold, no warning will be issued.

[0197] Otherwise, for each season, each period in the maximum value curve of the main transformer multivariate load that exceeds the main transformer judgment threshold is used as an early warning investigation period, and the combination of clusters of various types of loads corresponding to each early warning investigation period is obtained;

[0198] The load forecast values ​​of the samples of different categories of loads in each intersection date involved in the combination of clusters of each category of loads corresponding to each early warning investigation period are added together, and the load forecast result whose sum of the load forecast values ​​exceeds the main transformer judgment threshold is used as the early warning data, the corresponding intersection date and early warning investigation period are used as the early warning time, and the probability of occurrence of the cluster combination is used as the early warning probability; the early warning data, early warning time and early warning probability are output as the early warning result.

[0199] It can be understood that firstly, according to the early warning and troubleshooting period, the cluster corresponding to the early warning and troubleshooting period can be determined in the maximum load curve of each load, and each load corresponds to a cluster, thereby forming a combination of clusters of various categories of loads corresponding to the early warning and troubleshooting period.

[0200] It should be clarified that the crossover date involved in the combination of clusters of loads of various categories corresponding to the early warning and troubleshooting period specifically refers to: the cluster of each category of loads corresponding to the early warning and troubleshooting period contains samples of several dates, and the overlapping sample dates in the clusters of loads of various categories corresponding to the early warning and troubleshooting period are used as crossover dates. Since the clusters of loads of various categories are determined based on the maximum load forecast corresponding to each period, the load forecast values ​​of different categories of loads in the early warning and troubleshooting period of the crossover date are generally large, so the load forecast values ​​of the samples of different categories of loads in the crossover date can be added to determine whether the current early warning period of the crossover date needs an early warning.

[0201] Taking the scenario where the load characteristics of the main transformer are clear and the load data can be obtained independently as an example, the clustering and warning process is explained as follows:

[0202] According to the cluster analysis results, the occurrence probability of each cluster of multi-load is calculated, and the main transformer alarm threshold is set with the main transformer load rate as the constraint condition, and an extreme warning is issued for excessive load.

[0203] Assume that the loads connected to a main transformer are charging stations, large residential users, and large industrial users. The loads are clustered into N categories based on seasons. The clusters are: For example, if N=3 categories, the cluster distribution of 3 categories is , including the days are The clustering result curve is as follows Figure 2 As shown. Figure 2 In the example, the maximum load corresponding to time t1 is in the cluster The maximum load corresponding to time t2 is in the cluster The maximum load corresponding to time t3 is in the cluster , connect the maximum values ​​of each load into a line to obtain the maximum load curve. The corresponding multivariate load occurrence probabilities are The maximum load curve of multiple loads is superimposed to obtain the total maximum load curve of the main transformer. The maximum load curve of multiple loads of the main transformer is as follows: Figure 3 As shown, the total probability of occurrence is the product of the probability of occurrence of multivariate loads, and the intersection probability is the maximum probability of the two types of clusters. The calculation formula is as follows:

[0204] (twenty one)

[0205] In the formula, Expressed as The probability of occurrence of a load at time t.

[0206] The total load of the main transformer at this time is the sum of the loads at each moment of the multi-load , the calculation formula is as follows:

[0207] (twenty two)

[0208] In the formula, For the The maximum load value of the class load.

[0209] Based on the total maximum load curve obtained, the main transformer is set to operate without overload (80%). The main transformer capacity) is used as the judgment threshold, and the excessive part is judged as extreme load warning. The extreme load alarm judgment diagram is shown in Figure 4 As shown. Figure 4 It can be seen that the curve above the threshold is the extreme load that requires an alarm, and the alarm date and time period are found from the corresponding cluster; the time corresponding to the curve below the threshold does not require an alarm. Based on the probability of extreme load, the load rate is calculated. Based on the "Q∕GDW 10738-2020 Technical Guidelines for Distribution Network Planning and Design" and considering factors such as operation requirements, the main transformer load rate is selected as the warning indicator, and the extreme load warning level in the pilot area is divided into 3 levels, and the indicator threshold range for each level is established as follows: No alarm is required: When 0%<load rate≤80%, it is considered to be normal operation. Yellow warning: When 80%<load rate≤100%, it is considered to be overloaded. Red warning: When the load rate>100%, it is considered to be overloaded. Extreme load warnings are issued for overload and overload conditions.

[0210] The specific embodiment of the present invention also provides a specific application example of the above method to carry out specific example analysis. The load nature of the No. 1 main transformer 1 is difficult to determine, and a seasonal comprehensive hierarchical load forecasting method is used for forecasting; the load nature of the No. 2 main transformer is biomass and industry, and a seasonal multivariate load forecasting method is used for forecasting.

[0211] (1) Seasonal comprehensive stratified load forecasting and extreme warning

[0212] The No. 1 main transformer is 63MVA, and the nature of the load is difficult to judge. The seasonal comprehensive stratified load forecasting method is used to decompose it into fluctuating components and stable components. The fluctuating component adopts the multi-factor prediction method, and the stable component adopts the time series method. The two prediction curves are superimposed to obtain the overall prediction result, and then the extreme load warning is issued with the load rate as a constraint.

[0213] 1) Baseline load forecast

[0214] The maximum load comparison method and the benchmark load comparison method are used to check each other, and the benchmark current load curve for 2022 after the load data is decomposed is shown in Figure 5. Figure 5 It can be seen that the benchmark current load is mainly concentrated between 23.5-26.5MW, and the load curve is relatively stable. Using the time series method, the load at 8760 benchmark point in 2023 is predicted. The benchmark load prediction curve for 2023 is as follows: Figure 6 As shown in the figure. Figure 6 It can be seen that the fluctuation range of the benchmark load curve is relatively stable, concentrated between 25-29MW.

[0215] 2) Fluctuating load forecasting

[0216] The maximum load comparison method and the benchmark load comparison method are used to check each other. The schematic diagram of the decomposed fluctuating load is as follows: Figure 7 As shown. Figure 7 It can be seen that the load fluctuations in summer and winter are large, which is also consistent with the current situation in Songjiang where load fluctuations are greatly affected by air conditioning. Using the multi-factor prediction method, the key influencing factors include temperature, working days, and holidays. Based on the historical load data, the load fluctuation prediction curve of the power load is obtained. Figure 8 As shown. Figure 8 It can be seen that the overall predicted load situation is consistent with the current load characteristics, that is, the load fluctuates greatly in summer and winter.

[0217] 3) Total load forecast results

[0218] Through the above load prediction curve, the total load prediction curve is obtained as follows: Fig. 9 As shown. Fig. 9 It can be seen that through the comprehensive prediction of the above curves, it can be seen that the load curve fluctuates greatly in summer and winter, and the load fluctuates less in spring and autumn.

[0219] 4) Load warning

[0220] According to the known load data, three types of clustering results are selected as the basis for load division. Since the load characteristics of spring, summer, autumn and winter are obviously different, it is not suitable to cluster the data throughout the year, so cluster analysis is required separately. This example selects the summer with the highest load throughout the year as an example for extreme warning analysis.

[0221] Select the summer time period T = 92 days, extract the average value (AVERAGE), maximum value (MAX), and minimum value (MIN) of the load data, and use the K-means clustering method to cluster them into three categories. The clustering result is shown in the figure below. Fig.10 shown.

[0222] According to the clustering results, cluster 1 includes 42 days, cluster 2 includes 30 days, and cluster 3 includes 20 days. The specific time periods are shown in Table 1.

[0223] Table 1 Classification of load days in 2023

[0224]

[0225] Establish a typical curve for each cluster Fig.11 As shown. Fig.11 It can be seen that cluster 1 reaches its peak load around 1 to 4 a.m., cluster 2 reaches its peak load at 4 p.m., and cluster 3 reaches its peak load at 12 noon. The load occurrence probability and maximum load value of the three clusters are calculated, as shown in Table 2 below.

[0226] Table 2 Clustering results

[0227]

[0228] The capacity of the No. 1 main transformer is 63MVA, so the basic threshold of transformer load warning is 50.4MW (i.e. ), it can be found that the maximum loads of the clustering results of category 1, category 2, and category 3 do not exceed the threshold of the load of main transformer No. 1, so there is no need for load warning.

[0229] (2) Use seasonal multi-load forecasting and extreme warning

[0230] The capacity of No. 2 main transformer is 70MVA, and the connected loads are biomass and industry respectively. The multi-element load forecasting method is used to carry out load forecasting. By superimposing two forecasting curves, the overall forecast result is obtained, and then an early warning is issued with the load rate as a constraint.

[0231] Next, we will focus on the load warning part.

[0232] Since the No. 2 main transformer is connected to biomass and large industrial users, cluster analysis is performed on them according to the known load data, and the summer and spring seasons with the highest load throughout the year are selected as examples for extreme warning analysis.

[0233] Based on the industrial load forecast results, the maximum (MAX), average (AVERAGE), and range (XN) characteristic values ​​of the load are extracted, and the K-means clustering method is used to perform three-category clustering division. The result of the division is shown in the figure below: Fig.12 shown.

[0234] Summer includes T=92 days. According to the clustering results, cluster 1 includes 30 days, cluster 2 includes 29 days, and cluster 3 includes 33 days. The specific time periods are shown in Table 3:

[0235] Table 3 Classification of industrial load days in 2023

[0236]

[0237] Three cluster analysis curves of industrial load, and typical curves of each cluster are established. Fig.13 As shown. Fig.13 It can be seen that there are multiple load peaks in clusters 1, 2, and 3, so the probability of calculating each cluster is , , .

[0238] Based on the biomass load prediction results, the average value (AVERAGE), average difference (AVEDEY), and range (XN) characteristic values ​​of the load are extracted, and the K-means clustering method is used to cluster the load into three categories. The clustering result generated by the division is shown in Figure 14.

[0239] Summer includes T=92 days. According to the clustering results, cluster 1 includes 35 days, cluster 2 includes 27 days, and cluster 3 includes 30 days. The specific time periods are shown in Table 4 below.

[0240] Table 4 Classification of bioburden days in 2023

[0241]

[0242] Three cluster analysis curves of biomass load are established, and typical curves of each cluster are established. Fig.15 shown.

[0243] Depend on Fig.15 It can be seen that cluster 1 has a load peak at 4 am, cluster 2 has a load peak at 14 pm, and cluster 3 has a load peak at 21 pm. The probability of cluster 3 appearing is calculated as follows: , , .

[0244] In summary, the probability of the combined superposition of No. 2 main transformer being connected to industry and biomass is shown in Table 5.

[0245] Table 5 Total probability of No. 2 main transformer in 2022

[0246]

[0247] The maximum values ​​of the three types of clustered loads among the multiple loads connected to the No. 2 main transformer are shown in Table 6.

[0248] Table 6 Maximum load results

[0249]

[0250] Extract the maximum load curve: The maximum industrial load and the maximum biomass load curves are as follows: Fig.16 , Fig.17 The maximum value curve of the main transformer multi-load is shown in Fig.18 The capacity of the No. 2 main transformer is 70MVA, so the basic threshold of the transformer load warning is 56MW. Due to the existence of two dimensions of non-intersecting time and intersecting time in different clusters in biomass and industry, in order to ensure the consistency of basic conditions and more accurately evaluate the two sets of data, load warning needs to be carried out in the same time dimension. Fig.18 It can be seen that the first and third loads exceeding the alarm threshold line are corresponding to cluster 2. Therefore, extreme load warnings will be issued from 0:00 to 1:00 in the morning and from 15:40 to 16:20 in the afternoon on June 19-22 and July 21-31, 2023, with an alarm probability of 9.25%; the second load exceeding the alarm threshold line corresponds to the superposition of clusters 1 and 3. Therefore, warnings will be issued from 3:35 to 4:30 in the morning on June 1-9, July 8-10, and August 4-5, 2023, with an alarm probability of 10.84%.

[0251] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0252] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A load early warning method based on seasonal load hierarchical analysis, characterized in that: The method comprises: Obtain the scenario of the load characteristics of the main transformer access; for the scenario where the load characteristics of the main transformer access are fuzzy, use the seasonal comprehensive hierarchical load forecasting method to perform load forecasting and obtain the load forecast results for the next year; perform cluster analysis on the load forecast results for the next year to obtain cluster analysis results; for the scenario where the load characteristics of the main transformer access are clear and the load data can be obtained independently, use the seasonal multivariate load forecasting method to perform load forecasting on each category of loads respectively, and obtain the load forecast results of various categories of loads in the next year; perform cluster analysis on the load forecast results of various categories of loads in the next year to obtain cluster analysis results; According to the cluster analysis results, the extreme load early warning judgment of the main transformer is carried out to obtain the early warning result; For the scenario where the load characteristics of the main transformer are fuzzy, cluster analysis is performed on the load forecast results for the next year to obtain the cluster analysis results and execute: Divide the load forecast results for the next year by season to obtain the load forecast results for four seasons; For each season, the load forecast results of each natural day in the load forecast results of the season are clustered as samples to form several clusters, and the 24-hour load forecast maximum value curve of each cluster is established, and the occurrence probability of each cluster is calculated; from the 24-hour load forecast maximum value curve of all clusters, the load forecast maximum value corresponding to each time period and its corresponding cluster are selected respectively; the load forecast maximum value corresponding to each time period is connected to obtain the maximum load curve of the corresponding season; The maximum load curves of the four seasons were summarized as cluster analysis results; For scenarios where the load characteristics of the main transformer access are clear and the load data can be obtained independently, cluster analysis is performed on the load forecast results of various types of loads in the next year to obtain the cluster analysis results and execute: Divide the load forecast results for the next year by season to obtain the load forecast results for four seasons; For each category of load in each season, the load forecast results of each natural day in the load forecast results of the category of load in the season are clustered as samples to form several clusters, and the 24-hour load forecast maximum value curve of each cluster is established, and the occurrence probability of each cluster is calculated; the load forecast maximum value corresponding to each time period and its corresponding cluster are selected respectively; the load forecast maximum value corresponding to each time period is connected to obtain the maximum load curve of the corresponding category of load in the corresponding season; The maximum load curves of all types of loads in the corresponding season are superimposed. In each period, the occurrence probabilities of the clusters corresponding to the maximum load forecast values ​​of various types of loads are multiplied and the maximum load forecast values ​​are added to obtain the maximum load curve of the main transformer in the corresponding season. The maximum value curves of multivariate loads of main transformers in four seasons are summarized as the cluster analysis results.

2. The load early warning method based on seasonal load hierarchical analysis according to claim 1 is characterized in that: The seasonal comprehensive hierarchical load forecasting method is used to perform load forecasting, and the following steps are performed: Collect the historical load data of the previous year when the main transformer is connected, and decompose the historical load data of the previous year into historical benchmark load data and historical fluctuating load data according to the fluctuation of historical load data in different seasons; Using a time series method to perform load forecasting on the historical benchmark load data to obtain forecast benchmark load data for the next year; Using a multi-factor load forecasting method and combining the historical meteorological data of the previous year, the load forecasting of the historical fluctuating load data is performed to obtain the predicted fluctuating load data of the next year; The predicted baseline load data and predicted fluctuating load data for the same day and the same period of the next year are superimposed to obtain the load forecast results for the corresponding period of the corresponding day of the next year, and finally the load forecast results for the next year are obtained.

3. The load early warning method based on seasonal load hierarchical analysis according to claim 2 is characterized in that: The above method decomposes the historical load data of the previous year into historical benchmark load data and historical fluctuating load data, and executes: Calculate the average values ​​of typical daily load curves in spring and autumn in the historical load data of the previous year in each period respectively, and use them as the load data of the corresponding period in the historical benchmark load data; The maximum difference between the summer and winter load curves in the previous year's historical load data and the historical benchmark load data in the corresponding time period is calculated respectively as the load data for the corresponding time period in the historical fluctuating load data.

4. The load early warning method based on seasonal load hierarchical analysis according to claim 1 is characterized in that: The seasonal multivariate load forecasting method is used to perform load forecasting, and the following steps are performed: According to the load characteristics of the main transformer, the types of multiple loads are determined, and seasonal prediction models are established for each type of load; By using the forecasting model that takes seasonality into consideration for various types of loads, the load forecast results for the corresponding types of loads in the next year are obtained.

5. The load early warning method based on seasonal load hierarchical analysis according to claim 4 is characterized in that: The multi-type loads are divided into output type loads, input type loads, and input-output type loads; among which, The output load is distributed power source, including photovoltaic power generation, hydropower generation, wind power generation and biomass power generation; Input loads include pure load users such as charging stations, residents, businesses, and industries; The input and output load is energy storage.

6. The load early warning method based on seasonal load hierarchical analysis according to any one of claims 1 to 3, characterized in that: For scenarios where the load characteristics of the main transformer are fuzzy, the extreme load warning of the main transformer is judged based on the cluster analysis results, and the following are executed: Compare the cluster analysis results with the main transformer judgment threshold. If the cluster analysis results do not contain the maximum load forecast value that exceeds the main transformer judgment threshold, no warning will be issued. Otherwise, for each season, each period in the maximum load curve that exceeds the main transformer judgment threshold is used as an early warning investigation period, and the cluster corresponding to each early warning investigation period is obtained; Samples that exceed the main transformer judgment threshold in the early warning investigation period are selected from the clusters corresponding to each early warning investigation period, and the load forecast results of the samples in the early warning investigation period are used as early warning data, the date of the selected samples and the early warning investigation period are used as the early warning time, and the occurrence probability of the cluster is used as the early warning probability; The warning data, warning time and warning probability are output as warning results.

7. The load early warning method based on seasonal load hierarchical analysis according to any one of claims 1, 4 or 5, characterized in that: For scenarios where the load characteristics of the main transformer are clear and the load data can be obtained independently, the extreme load warning of the main transformer is judged based on the cluster analysis results and the following are executed: Compare the cluster analysis results with the main transformer judgment threshold. If the cluster analysis results do not contain the maximum value of the main transformer multi-load that exceeds the main transformer judgment threshold, no warning will be issued. Otherwise, for each season, each period in the maximum value curve of the main transformer multivariate load that exceeds the main transformer judgment threshold is used as an early warning investigation period, and the combination of clusters of various types of loads corresponding to each early warning investigation period is obtained; The load forecast values ​​of the samples of different types of loads in each cross date involved in the combination of the clusters of each type of load corresponding to each early warning investigation period are added together, and the load forecast result whose sum of the load forecast values ​​exceeds the main transformer judgment threshold is used as the early warning data, the corresponding cross date and early warning investigation period are used as the early warning time, and the occurrence probability of the cluster combination is used as the early warning probability; The warning data, warning time and warning probability are output as warning results.

8. The load early warning method based on seasonal load hierarchical analysis according to claim 7 is characterized in that: The K-means clustering method is used to perform cluster analysis on the load forecast results for the next year.

Images