Method for accurately predicting weight and overload of distribution transformer

Abstract

The invention discloses a method for accurately predicting the heavy overload of a distribution transformer, and the method carries out the load prediction and heavy overload judgment through reconstructing a data structure and a data algorithm for the collected data of the distribution transformer. The method comprises the following steps: preprocessing historical load data, converting the historical load data into a load rate per 15 minutes, calculating a maximum load rate, a minimum load rate and an average load rate, constructing a clustering analysis algorithm to analyze the preprocessed data, selecting a proper clustering number according to different conditions of the distribution transformer to determine a data algorithm, and determining a final clustering scheme. And optimizing and constructing a load prediction model for each determined cluster and algorithm, and finally obtaining a prediction value of the model for the future load of the distribution transformer.

Description

Technical Field

[0001] The present invention relates to the technical field of load prediction, and in particular to a method for accurately predicting heavy overload of a distribution transformer. Background Art

[0002] With the rapid development of modern social economy, the demand for electricity continues to grow, and the scale and complexity of the power system continue to increase. As a key device connecting the high-voltage transmission network and the low-voltage user terminal in the power system, the operating status of the distribution transformer is directly related to the quality, reliability and stability of the power supply. Under the background of the widespread application of smart grids and distributed energy, the distribution transformer faces a more diversified and dynamically changing operating environment. The power supply structure in the distribution network has changed significantly. The intermittent and volatile power supply has brought new challenges to the load forecasting and operation management of the distribution transformer. At the same time, users' requirements for power quality and power supply reliability are also increasing. Any heavy overload problem of the distribution transformer may cause local power outages, equipment damage, and even affect the safe and stable operation of the entire power system. Therefore, accurately predicting the heavy overload of the distribution transformer is of vital importance to ensure the reliable operation of the power system. How to effectively integrate the real-time data of distributed energy and incorporate it into the load forecasting model of the distribution transformer is one of the key problems to improve the accuracy of the forecast.

[0003] However, the existing distribution transformer overload prediction still has many technical defects, including at least the following prominent aspects: 1. The traditional distribution transformer load prediction and overload determination methods have difficulties in processing massive, high-dimensional, and heterogeneous distribution transformer operation data, and cannot fully mine the effective information in the data. In terms of feature extraction, there is a lack of effective means to fuse multiple related data into meaningful features, resulting in the input features of the model being not comprehensive and accurate enough, and unable to accurately reflect the operating status of the distribution transformer. 2. The traditional prediction model is based on a simple linear assumption, which is difficult to adapt to the nonlinear and non-stationary characteristics of the distribution transformer load. Moreover, the traditional model lacks pertinence when dealing with different types of distribution transformers, and cannot fully consider the differences in load capacity, operating characteristics, etc. of different types of transformers. 3. In some traditional methods, although cluster analysis may be used, the selection of clustering algorithms and the determination of the number of clusters often lack scientific basis. The number of clusters is simply selected based on experience without fully considering the actual distribution of the data, resulting in the clustering results cannot accurately reflect the differences in overload-related features of different types of transformers, and cannot provide effective classification information for subsequent prediction models, affecting the accuracy and practicality of the entire prediction method. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method for accurately predicting heavy overload of a distribution transformer in view of various deficiencies in the prior art.

[0005] To achieve the above purpose, the technical solution adopted by the present invention is as follows: A method for accurately predicting heavy overload of distribution transformers, characterized in that load prediction and heavy overload determination are performed by reconstructing data structure and data algorithm for collected distribution transformer data.

[0006] As a preferred technical solution of the present invention, historical data of the distribution transformer is collected every 15 minutes, including but not limited to historical load data, actual temperature data, weather forecast temperature data, distribution transformer ledger data, historical heavy overload data, and data analysis and cleaning are performed.

[0007] As a preferred technical solution of the present invention, the historical load data is converted into a load rate every 15 minutes, and the maximum, minimum and average load rates are calculated. The load rate of each 15-minute group is calculated as follows: Where BL is the transformer ratio, BYQRL is the transformer capacity, and P is the load at the corresponding point.

[0008] Taking the average load value of 96 points on day T as the base, the calculation formula is: As a preferred technical solution of the present invention, a cluster analysis algorithm is constructed to analyze the preprocessed data. The specific steps are: based on the preprocessed load data, feature construction is performed, and the actual weather temperature data, weather forecast temperature data, date and holiday data, distribution transformer ledger data, etc. are combined with the constructed features to jointly form a feature vector; the number of clusters is determined and cluster analysis is performed; cluster evaluation and adjustment.

[0009] As a preferred technical solution of the present invention, the feature construction step in the cluster analysis algorithm is set with a series of historical load values ​​of the distribution transformer arranged in chronological order, the number of data points is N, and for each frequency component k, the Fourier coefficient X at the frequency is calculated. k , calculate X k When , each data point in the entire load data series is operated, and each data point is multiplied by a complex factor related to frequency k and time point n. , and then add these products together; Determine the number m of low-frequency Fourier coefficients to be selected, and select the first m Fourier coefficients with smaller k values X [ 0 ] , X [ 1 ] , ⋯ , X [ m − 1 ] as features; the mean, variance, maximum value and minimum value of m low-frequency coefficients are used as statistical features to further enrich the feature information; the feature data including but not limited to temperature, weather, holidays and transformer ledger data are converted into feature vectors after numerical conversion, and combined with the feature information obtained previously to construct a complete feature space; Calculate the mean and standard deviation of each feature in the feature vector. For each eigenvalue of each data point, use the standardized formula to calculate, subtract the mean from the eigenvalue, and then divide it by the standard deviation to get the standardized eigenvalue. All features are converted to a new scale with 0 as the mean and 1 as the standard deviation, to obtain a feature vector that is more suitable for model training.

[0010] As a preferred technical solution of the present invention, the cluster number determination step is to calculate different The clustering error under the value is determined by adding the square of the distance from each data point in each cluster to the center of the cluster to which it belongs. As increases, the clustering error value will gradually decrease. When is less than the actual number of clusters, the clustering error value decreases rapidly; when When the number of clusters reaches or exceeds the true number, the clustering error value decreases gradually. The value is the appropriate number of clusters, that is, the distribution transformer data is divided into categories.

[0011] As a preferred technical solution of the present invention, the step of determining the number of clusters can also be implemented by the following data algorithm: Determine a series of values ​​for the number of clusters into which the distribution transformer data is divided. For each selected number of clusters The clustering algorithm is used to cluster the entire distribution transformer data set to obtain Different clustering results are marked as , for any data point in the dataset ,calculate and Except The sum of the distances between all other data points except middle The number of data points other than For data points The number of the cluster where the data point is obtained to the cluster to which it belongs The average distance of all other data points in ; For each other cluster ,calculate arrive The average and minimum distances of all data points in the data point. This minimum value is the data point To other clusters The average distance between the data points ;use and The difference, divided by m a x { a [ i ] , b [ i ] } Calculate data points The silhouette coefficient ; The silhouette coefficient of all data points Add and divide by the total number of data points Calculate the number of clusters for the entire data set The average silhouette coefficient , will be different The average silhouette coefficient corresponding to the value To compare, The largest The value is the optimal number of clusters determined, that is, the number of categories that is most suitable for dividing the distribution transformer data into.

[0012] As a preferred technical solution of the present invention, the step of determining the number of clusters can also be implemented by the following data algorithm: Determine a set of values ​​for the number of clusters into which the distribution transformer data is divided , calculate the original data set At the current number of clusters The degree of cluster dispersion under , for each cluster , each eigenvalue of all data points in the cluster is added separately, and then divided by the number of data points to obtain the vector of its central mean vector , calculate each data point arrive The distance is squared and then added to get the original data set The degree of cluster dispersion; Setting Group reference uniform distribution dataset , the number of samples in these datasets and the original dataset is the same, and the data points are randomly generated within the range of the original data. For each set of reference uniformly distributed data sets , using the same clustering algorithm and the current The values ​​are clustered to obtain the clustering results. ; For each reference uniformly distributed data set The clustering results of , calculate its cluster dispersion ; By all Group Add and divide Get the average cluster dispersion by comparing different The average cluster dispersion corresponding to the value is The maximum value of the logarithm of The value is the optimal number of clusters initially determined; for The rationality of the value selection is calculated by calculating the standard error of this maximum value. , determine whether the maximum value is greater than The corresponding average cluster dispersion is The logarithm and standard error of If this condition is met, the current The value is the number of categories into which the distribution transformer data is most appropriately divided.

[0013] As a preferred technical solution of the present invention, the step of determining the number of clusters also samples the data multiple times, runs a clustering algorithm on the data sampled each time, observes the stability of the clustering results, and determines the final clustering solution.

[0014] As a preferred technical solution of the present invention, for each determined cluster, a load forecasting model is constructed by the Light GBM algorithm, so that the model can learn the relationship between the characteristics of the load data of the distribution transformer in the cluster and the load change law; the feature vector of the distribution transformer to be predicted after standardized processing is input into the trained prediction model of the corresponding cluster to obtain the model's predicted value for the future load of the distribution transformer.

[0015] The beneficial effect of adopting the above technical scheme is that: the present invention constructs features based on Fourier coefficients, so that the constructed feature vector can more accurately display the characteristics of the distribution transformer, and by constructing different cluster quantity data algorithms, the most suitable number of clusters is selected for different situations of the distribution transformer, so that the prediction model can make full use of the clustering results, and the model has stronger adaptability and pertinence for different types of distribution transformers. The model can better learn the load change law of the transformer in the cluster, thereby improving the accuracy of the prediction.

[0016] The following embodiments describe in detail the technical advantages and beneficial effects of various technical details of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0018] In the attached picture: Figure 1 is a flow chart of the present invention; Figure 2 The data flow chart of the Light GBM tree model algorithm of the present invention; Figure 3 The load rate trends of various distribution transformers that are the prediction targets of the present invention Figure 4 This is an example diagram of the correlation between temperature and load of the present invention; Figure 5 It is a load rate diagram of 14 points of different distribution transformers of the present invention; Figure 6 This is a comparison diagram of the model effects of two prediction methods for the traditional distribution transformer of the present invention; Figure 7 This is a comparison diagram of the effects of two prediction methods including distributed access models of the present invention; DETAILED DESCRIPTION

[0019] The following examples describe the present invention in detail. In the description of the following examples, specific details such as specific system structures, technologies, etc. are proposed for illustration rather than for limitation, so as to thoroughly understand the present application embodiments. However, it should be clear to those skilled in the art that the present application can also be implemented in other embodiments without these specific details. In other cases, the detailed description of well-known systems, devices, circuits and methods is omitted to prevent unnecessary details from hindering the description of the present application. It should be understood that when used in the present application specification and the attached claims, the term "comprising" indicates the presence of the described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their collections. It should also be understood that the term "and / or" used in the present application specification and the attached claims refers to any combination of one or more of the items listed in association and all possible combinations, and includes these combinations.

[0020] As used in the specification of the present application and the appended claims, the term "if" can be interpreted as "when..." or "once" or "in response to determination" or "in response to detection" according to the context. Similarly, the phrase "if it is determined" or "if (the described condition or event) is detected" can be interpreted as meaning "once it is determined" or "in response to determination" or "once (the described condition or event) is detected" or "in response to detection (the described condition or event)" according to the context. In addition, in the description of the specification of the present application and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance. The reference "one embodiment" or "some embodiments" described in the specification of the present application means that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Thus, the sentences "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in the specification are not necessarily all referring to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "include", "comprising", "having" and variations thereof all mean "including but not limited to", unless specifically emphasized otherwise. Example 1

[0021] For the historical load values ​​of distribution transformers arranged in chronological order and with N data points, for each frequency component k, the Fourier coefficient is calculated by multiplying each data point by a complex factor related to the time point and summing them up, determining the number of low-frequency Fourier coefficients to be selected, selecting the first one with a smaller value as the feature, and calculating the mean, variance, maximum and minimum of this low-frequency coefficient as statistical features. The temperature, weather, holidays, transformer ledger and other data are converted into feature vectors through numerical conversion, and combined with the previous Fourier coefficient features and statistical features to construct a complete feature space. The mean and standard deviation of each feature in the feature space are calculated, and the feature value of each data point is subtracted from its mean and then divided by the standard deviation using the standardized formula, and all features are converted to a new scale with a mean and a standard deviation, thereby obtaining a feature vector that is more suitable for model training.

[0022] The frequency components represented by the low-frequency Fourier coefficients correspond to the slowly changing parts of the load data, such as long-term electricity consumption trends and seasonal electricity consumption patterns. Because heavy overload is often related to long-term high load demand or seasonal electricity peaks, choosing low-frequency Fourier coefficients can more accurately predict heavy overload conditions of transformers.

[0023] In order to more comprehensively describe the operating status of the distribution transformer and improve the accuracy of heavy overload prediction, the Fourier coefficient features are combined with other relevant data features. The load of the distribution transformer is affected by many factors. In addition to its own historical load change law, environmental factors, time factors and the transformer's own properties have an important impact on the load. Combining these features from different sources can construct a more complete feature vector, so that the prediction model can comprehensively consider various factors and thus more accurately predict heavy overload conditions. Example 2

[0024] Collect the historical 15-minute load data of distribution transformers, actual weather temperature data, weather forecast temperature data, distribution transformer ledger data, historical distribution transformer heavy overload data, and perform data analysis and cleaning, and organize date and holiday data.

[0025] The data format of the 96-point load data of the distribution transformer load is as follows: Field Name Field Type Notes DEV_ID VARCHAR2(100) Distribution Transformer ID V_DATETIME DATETIME Date + Time V_LOAD NUMBER load The weather temperature data format is as follows: Field Name Field Type Notes ID VARCHAR2(100) area V_DATETIME DATETIME Date + Time (Hourly Level) WD NUMBER Temperature (two decimal places) The weather forecast temperature data format is as follows: Field Name Field Type Notes ID VARCHAR2(100) area PRE_DATE VARCHAR2(100) Start time V_DATETIME_PRE DATETIME Forecast temperature period (24 hours) WD_PRE NUMBER Forecast temperature (two decimal places) The format of holiday information data is as follows: Field Name Field Type Notes V_DATE DATETIME date V_YEAR NUMBER years V_MONTH NUMBER month V_WEEK VARCHAR2(20) Week HOLI VARCHAR2(20) Festival Name Distribution transformer ledger data: Field Name Field Type Notes ID VARCHAR2(100) area DEV_ID VARCHAR2(100) Distribution Transformer ID BYQRL NUMBER Transformer capacity TATV NUMBER Transformer ratio Historical distribution transformer overload data: Field Name Field Type Notes V_DATE DATETIME date DEV_ID NUMBER Distribution transformer ID IS_OVER NUMBER Overload IS_HEAVE NUMBER Overload Preprocess the historical load data and convert it into a load rate every 15 minutes. Calculate the maximum, minimum and average loads including: The calculation method of load rate for every 15 groups is: Where BL is the transformer ratio, BYQRL is the transformer capacity, and P is the load at the corresponding point.

[0026] Taking the average load value of 96 points on day T as the base, the calculation formula is: By means of clustering analysis based on specific distance measurement and iterative optimization mechanism on multi-source and massive data including historical operation data of distribution transformers, real-time monitoring data and related environmental data obtained after a series of complex and targeted data cleaning, conversion, standardization and other preprocessing operations, the distribution transformers with different power capacities, different load characteristics, different operating years and different geographical distributions in the huge and complex network structure of the power grid are divided into groups or categories showing different internal laws and external manifestations in terms of electrical parameter characteristics, load change trends, potential fault risks and other aspects, and then a category grouping feature table with rigorous structure and complete data is created, which can provide rich and key category grouping information such as the proportion of distribution transformers in each category, statistical values ​​of typical electrical parameters of distribution transformers in each category, and correlation data between distribution transformers in each category and heavy overload events for the subsequent construction of prediction models with strong data fitting and generalization capabilities represented by deep neural networks and support vector machines. Each category Cm corresponds to a feature table, which contains the following fields: Distribution transformer ID (DEV_ID): a unique identifier for each distribution transformer; Category number (C_ID): the category number to which the distribution transformer belongs; Overload times in the past 30 days (OVER_CNT_M): The number of times the distribution transformer was overloaded in the past 30 days; HEAVY_CNT_M: The number of times the distribution transformer has been overloaded in the past 30 days. Overload times in the past 7 days (OVER_CNT_W): The number of times the distribution transformer was overloaded in the past 7 days; HEAVY_CNT_W: The number of times the distribution transformer has been overloaded in the past 7 days. Maximum load factor (MAX_FZL): The maximum load factor recorded by the distribution transformer in the past 30 days; Minimum load factor (MIN_FZL): The minimum load factor recorded by the distribution transformer in the past 30 days; Average load factor (AVG_FZL): the average load factor recorded by the distribution transformer in the past 30 days; Time since last heavy overload (ZGZ_LAST_DAY): The length of time since the last heavy overload occurred.

[0027] Generate a load ratio coefficient and restore the generated load ratio coefficient to a predicted load, including: multiplying the 96-point loads of each distribution transformer on day T by the corresponding ratio and dividing by the capacity as the load rate, multiplying the 96-point loads of each distribution transformer on day T+1 by the corresponding ratio and dividing by the capacity to obtain the load rate as the prediction target, using the Light GBM algorithm to generate a model, predicting the load rate on day T+1, and using the heavy overload judgment method to determine the prediction results and confirm the heavy overload situation.

[0028] The load rate of the distribution transformer evaluates the operation of the distribution transformer. The operation risks of the distribution transformer are mainly divided into three situations: normal, heavy load and overload.

[0029] Among them, heavy load refers to the distribution transformer in actual production where the load reaches 80% of the rated capacity of the distribution transformer and lasts for two hours or more. This state is recorded as heavy load.

[0030] Overload refers to a distribution transformer in which the load reaches 100% of the rated capacity of the distribution transformer and lasts for two hours or more in actual production. This state is recorded as overload.

[0031] Among them, normal means that in actual production, a distribution transformer with a load rate lower than 80% is normal. Example 3

[0032] a) Load rate characteristics analysis: Depend on Figure 3 It can be seen that due to the access of distributed photovoltaics and other energy sources such as hydropower, there are many different load trends. If all distribution transformers are modeled uniformly and predicted using the same set of models, there will be large errors in predicting the three models. This patent uses kmeans to group different distribution transformers to improve the model's learning ability for specific trend curves. b) Correlation analysis between load rate and temperature: The load factor and temperature of the distribution transformer are given by Figure 4 It can be seen that by analyzing the correlation between the load rate of the distribution transformer and the temperature, it is concluded that the temperature and the load rate of the distribution transformer show a second-order linear correlation.

[0033] c) Taking the load at 14:00 every day as an example, the corresponding load rate is calculated for the current load, and it can be concluded that the load rates of different distribution transformers at 14:00 are different. Because 14:00 is roughly the period of high temperature every day and the load is large, for traditional loads, the load rate of distribution transformers is also large, and it is easy to be overloaded, so it is taken as a data example.

[0034] Depend on Figure 6The traditional unified modeling method can well handle the prediction of distribution transformers without distributed access, and the prediction results of the traditional modeling method and the modeling method of this patent are not much different. Figure 7 It can be seen that the traditional unified modeling model predicts that when distributed photovoltaics are connected, the load rate curve will be greatly distorted, which is significantly different from the actual load rate. The curve distortion phenomenon is smaller when the modeling method of this patent is used.

[0035]

[0036] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0037] In various embodiments, the hardware implementation of the technology can directly adopt existing intelligent devices, including but not limited to industrial computers, PCs, smart phones, handheld stand-alone computers, floor-standing stand-alone computers, etc. Its input device preferably adopts a screen keyboard, its data storage and calculation module adopts existing memory, calculator, controller, its internal communication module adopts existing communication ports and protocols, and its remote communication adopts existing GPRS network, World Wide Internet, etc.

[0038] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0039] In the embodiments provided by the present invention, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are only schematic, for example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0040] Each functional unit in each embodiment of the present invention may be integrated into a processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit. If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it may be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and may also instruct the relevant hardware to complete it through a computer program. The computer program may be stored in a computer-readable storage medium, and the computer program may be executed by the processor to implement the steps of the above-mentioned method embodiments. Among them, the computer program includes computer program code, and the computer program code may be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media does not include electrical carrier signals and telecommunications signals.

[0041] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for accurately predicting heavy overload of a distribution transformer, characterized in that: By reconstructing the data structure and data algorithm of the collected distribution transformer data, load forecasting and heavy overload judgment are carried out.

2. The method for accurately predicting heavy overload of distribution transformer according to claim 1 is characterized in that: Collect historical data of distribution transformers every 15 minutes, including but not limited to historical load data, actual temperature data, weather forecast temperature data, distribution transformer ledger data, historical heavy overload data, and perform data analysis and cleaning.

3. The method for accurately predicting heavy overload of distribution transformer according to claim 1 is characterized in that: Convert the historical load data into a load rate every 15 minutes, and calculate the maximum, minimum, and average load rates. The load rate for each 15-minute group is calculated as follows: ； Where BL is the transformer ratio, BYQRL is the transformer capacity, and P is the load at the corresponding point; Taking the average load value of 96 points on day T as the base, the calculation formula is:

4. The method for accurately predicting heavy overload of distribution transformer according to claim 1, characterized in that: A cluster analysis algorithm is constructed to analyze the preprocessed data. The specific steps are as follows: based on the preprocessed load data, feature construction is performed, and the actual weather temperature data, weather forecast temperature data, date and holiday data, and distribution transformer ledger data are combined with the constructed features to form a feature vector; the number of clusters is determined and cluster analysis is performed; cluster evaluation and adjustment are performed.

5. The method for accurately predicting heavy overload of distribution transformer according to claim 4 is characterized in that: The feature construction step in the cluster analysis algorithm is set with a series of historical load values ​​of the distribution transformer arranged in chronological order, the number of data points is N, and for each frequency component k, the Fourier coefficient X at the frequency is calculated. k , calculate X k When , each data point in the entire load data series is operated, and each data point is multiplied by a complex factor related to frequency k and time point n. , and then add these products together; Determine the number m of low-frequency Fourier coefficients to be selected, and select the first m Fourier coefficients with smaller k values as features; the mean, variance, maximum value and minimum value of m low-frequency coefficients are used as statistical features to further enrich the feature information; the feature data including but not limited to temperature, weather, holidays and transformer ledger data are converted into feature vectors after numerical conversion, and combined with the feature information obtained previously to construct a complete feature space; Calculate the mean and standard deviation of each feature in the feature vector. For each eigenvalue of each data point, use the standardized formula to calculate, subtract the mean from the eigenvalue, and then divide it by the standard deviation to get the standardized eigenvalue. All features are converted to a new scale with 0 as the mean and 1 as the standard deviation, to obtain a feature vector that is more suitable for model training.

6. The method for accurately predicting heavy overload of distribution transformer according to claim 4, characterized in that: The step of determining the number of clusters is to calculate different The clustering error under the value is determined by adding the square of the distance from each data point in each cluster to the center of the cluster to which it belongs. As increases, the clustering error value will gradually decrease. When is less than the actual number of clusters, the clustering error value decreases rapidly; when When the number of clusters reaches or exceeds the true number, the clustering error value decreases gradually. The value is the appropriate number of clusters, that is, the distribution transformer data is divided into categories.

7. The method for accurately predicting heavy overload of distribution transformer according to claim 4, characterized in that: The step of determining the number of clusters can also be implemented by the following data algorithm: Determine a series of values ​​for the number of clusters into which the distribution transformer data is divided. For each selected number of clusters The clustering algorithm is used to cluster the entire distribution transformer data set to obtain Different clustering results are marked as , for any data point in the dataset ,calculate and Except The sum of the distances between all other data points except middle The number of data points other than For data points The number of the cluster where the data point is obtained to the cluster to which it belongs The average distance of all other data points in ; For each other cluster ,calculate arrive The average and minimum distances of all data points in the data point. This minimum value is the data point To other clusters The average distance between the data points ;use and The difference, divided by Calculate data points The silhouette coefficient ; The silhouette coefficient of all data points Add and divide by the total number of data points Calculate the number of clusters for the entire data set The average silhouette coefficient , will be different The average silhouette coefficient corresponding to the value To compare, The largest The value is the optimal number of clusters determined, that is, the number of categories that is most suitable for dividing the distribution transformer data into.

8. The method for accurately predicting heavy overload of distribution transformer according to claim 4, characterized in that: The step of determining the number of clusters can also be implemented by the following data algorithm: Determine a set of values ​​for the number of clusters into which the distribution transformer data is divided , calculate the original data set At the current number of clusters The degree of cluster dispersion under , for each cluster , each eigenvalue of all data points in the cluster is added separately, and then divided by the number of data points to obtain the vector of its central mean vector , calculate each data point arrive The distance is squared and then added to get the original data set The degree of cluster dispersion; Setting Group reference uniform distribution dataset , the number of samples in these datasets and the original dataset is the same, and the data points are randomly generated within the range of the original data. For each set of reference uniformly distributed data sets , using the same clustering algorithm and the current The values ​​are clustered to obtain the clustering results. ; For each reference uniformly distributed data set The clustering results of , calculate its cluster dispersion ; By all Group Add and divide Get the average cluster dispersion by comparing different The average cluster dispersion corresponding to the value is The maximum value of the logarithm of The value is the optimal number of clusters initially determined; for The rationality of the value selection is calculated by calculating the standard error of this maximum value. , determine whether the maximum value is greater than The corresponding average cluster dispersion is The logarithm and standard error of If this condition is met, the current The value is the number of categories into which the distribution transformer data is most appropriately divided.

9. The method for accurately predicting heavy overload of distribution transformer according to claim 4, characterized in that: The step of determining the number of clusters is to sample the data multiple times, run the clustering algorithm on the sampled data each time, observe the stability of the clustering results, and determine the final clustering solution.

10. The method for accurately predicting heavy overload of distribution transformer according to claim 1, characterized in that: For each determined cluster, a load forecasting model is constructed using the Light GBM algorithm to enable the model to learn the relationship between the characteristics of the distribution transformer load data in the cluster and the load change law; the standardized feature vector of the distribution transformer to be predicted is input into the trained prediction model of the corresponding cluster to obtain the model's predicted value for the future load of the distribution transformer.

Images