Distribution transformer weight and overload prediction method and equipment

Abstract

The invention relates to a weight and overload prediction method for a distribution transformer. The method comprises the following steps: acquiring historical power load data of daily preset interval time in a target area; decomposing the historical power load data according to a preset STL time sequence decomposition algorithm to obtain a plurality of sequence components; performing feature extraction on the sequence component to obtain a sequence feature component; predicting a future sequence value corresponding to the sequence feature component according to a preset prediction model to obtain power load prediction data; performing accuracy evaluation on the power load prediction data according to a preset evaluation model to obtain a sample prediction load; calculating a sample load rate according to the sample plate prediction load; and performing heavy overload judgment on the sample load rate according to a preset heavy overload judgment threshold. According to the technical scheme, the distribution transformer weight and overload prediction method and device have the advantages of being reliable in prediction precision, simple, easy to implement, good in model interpretability and the like, and can be well applied to power load data.

Description

Technical Field

[0001] The present invention relates to the field of electric power, and in particular to a distribution transformer heavy overload prediction method, system, readable storage medium and computer equipment. Background Art

[0002] With the continuous improvement of the consumption level of Chinese residents and the continuous improvement of the national power infrastructure, the national electricity demand continues to grow. On January 18, 2024, the National Energy Administration released data on the total electricity consumption in 2023. In 2023, the total electricity consumption of the whole society was 9224.1 billion kWh, a year-on-year increase of 6.7%, and the growth rate was 3.1 percentage points higher than that in 2022. According to the "2023-2024 National Electricity Supply and Demand Situation Analysis and Forecast Report" of the China Electricity Council, it is estimated that the total electricity consumption of the whole society in 2024 will be 9.8 trillion kWh, an increase of about 6% over 2023; at the same time, it is predicted that the national electricity demand will continue to maintain rigid growth in the next three years.

[0003] Distribution transformers (abbreviated as "distribution transformers") are important nodes for power transmission to users. If the distribution transformer is overloaded for a long time, it will reduce the life of the equipment and may also cause line failures and low voltage at the end of the low-voltage line. The power load of the distribution transformer largely determines the quality and safety of power supply in the region.

[0004] Predicting the load and heavy overload of distribution transformers in the future provides accurate guidance for the regulation of the power supply system, so that the power supply can take measures against heavy overload of distribution transformers in advance, which is an important means to ensure the quality and safety of regional power supply.

[0005] [Current state of the art]

[0006] At present, there are many technical methods for distribution transformer load prediction and heavy overload prediction. These methods can be generally divided into:

[0007] 1) Statistical analysis methods based on time series; 2) Methods based on machine learning; 3) Methods based on neural networks.

[0008] Time series analysis methods mainly include exponential smoothing model method, autoregression method and cumulative autoregression moving average method. For example, paper [1] studies the ARIMA time series model; paper [2] proposes a multi-feature short-term power load forecasting method based on double-layer XGBoost based on machine learning model; paper [3] proposes a distribution transformer daily peak load forecasting method based on BOXGBoost.

[0009] With the popularity of deep learning, more and more studies are using neural network models to predict power load, such as the paper [4] based on the LSTM+Attention model for short-term load prediction of typical distribution stations, and the paper [5] based on the Attention mechanism for CNN-GRU distribution network line overload short-term prediction method. Some researchers have also tried to integrate multiple methods, such as the paper [6] based on SVM-STL-LSTM for regional short-term power load prediction.

[0010] [1] Ai Xin, Zhou Zhiyu, Wei Yanping, et al. Transferable load bidding strategy based on autoregressive integrated moving average model[J]. Automation of Electric Power Systems, 2017, 41(20): 26-31+104;

[0011] [2] Sun Chao, Lu Qi, Zhu Sitong, et al. Ultra-short-term power load forecasting considering multi-feature influence based on double-layer XGBoost algorithm [J]. High Voltage Engineering, 2021, 47(08): 2885-2898;

[0012] [3] Deng Wei, Mei Yujie, Li Yong, et al. Distribution transformer daily peak load prediction and heavy overload warning method based on BOXGBoost[J]. Journal of Electric Power System and Automation, 2024, 36(07): 49-58;

[0013] [4] Huang Chaokai, Lin Honghao, Wang Kecheng, et al. Short-term load forecasting method for typical distribution area based on LSTM+Attention model[J]. Microcomputer Applications, 2024, 40(08): 88-91;

[0014] [5] Yang Xiu, Hu Zhongyu, Tian Yingjie, et al. CNN-GRU distribution network line heavy overload short-term prediction method based on attention mechanism [J]. Journal of Electric Power Science and Technology, 2023, 38(01): 201-209;

[0015] [6] Wang Chen, Li Youxuan, Wu Qiqi, et al. Research on regional short-term power load forecasting based on SVM-STL-LSTM[J]. Hydropower Energy Science, 2024, 42(04): 215-218;

[0016] [Defects of existing technology]

[0017] 1) The statistical analysis method based on time series has high requirements on data stability. At the same time, a simple time series model is difficult to capture complex data characteristics. When the data is complex and there are irregular fluctuations, the prediction accuracy will drop significantly.

[0018] 2) Machine learning-based methods often require a large amount of data and external feature input to obtain good results, which has high data requirements and poor model solvability. At the same time, the collection and maintenance of external data also has high costs for users.

[0019] 3) Neural network-based prediction methods are often complex to implement and require a large amount of data. At the same time, the model training time is long, high computing resources are required (high cost), and the model is poorly interpretable and difficult to debug, which limits its use in daily work. When there is little data, this type of technology will cause a sharp drop in prediction accuracy. Summary of the invention

[0020] The present invention aims to solve at least one of the technical problems existing in the prior art or related art.

[0021] To this end, the purpose of the present invention is to provide a distribution transformer heavy overload prediction method and device with the characteristics of reliable prediction accuracy, simplicity and ease of implementation, good model interpretability, and can be well applied to power load data.

[0022] To achieve the above object, the technical solution of the first aspect of the present invention provides a distribution transformer heavy overload prediction method, comprising:

[0023] Obtain historical power load data at preset daily intervals within the target area;

[0024] Decomposing the historical power load data according to a preset STL time series decomposition algorithm to obtain multiple sequence components;

[0025] Extracting features from the sequence components to obtain sequence feature components;

[0026] Predicting the future sequence value corresponding to the sequence characteristic component according to a preset prediction model to obtain power load prediction data;

[0027] Performing an accuracy evaluation on the power load forecast data according to a preset evaluation model to obtain a sample forecast load;

[0028] Calculating a sample load rate based on the sample predicted load;

[0029] A heavy overload judgment is performed on the sample load rate according to a preset heavy overload judgment threshold.

[0030] In the above technical solution, preferably, the sequence components include trend components, seasonal components and residual components, and the historical power load data is decomposed according to a preset STL time series decomposition algorithm, including:

[0031] Performing initial decomposition on the historical power load data using a local weighted regression scatter point smoothing algorithm to obtain an initial trend;

[0032] According to the initial trend, the seasonal term and the trend term in the historical power load data are respectively smoothed and iterated to obtain the seasonal component and the trend component;

[0033] The residual component is calculated according to the seasonal component and the trend component.

[0034] In the above technical solution, preferably, the historical power load data is initially decomposed using a local weighted regression scatter point smoothing algorithm, including:

[0035] Select a window of length h, for the historical power load data y t ={y1, y2, ..., y n}, when fitting one of the points y i When considering As data points participating in local weighted fitting;

[0036] For each data point in the window, the weight w is calculated using the three-weight function according to the distance between each data point and the center point. j :

[0037] For point y j For the center point y in the window i The weight of , when |y j -y i |≤1,w j =(1-|y j -y i | 3 ) 3 , otherwise w j =0;

[0038] Using the weighted data points in the window, fit the first-order polynomial y = a + bx, by minimizing the weighted least squares objective function ∑ j w j (y j -a-bx j ) 2 To estimate coefficients a and b;

[0039] Move the window along the time series and repeat the above steps to obtain a smoothed initial trend estimate

[0040] In the above technical solution, preferably, the expression of the residual component is: R t =y t -T t -S t , where y t is the historical power load data, T tRepresents the decomposed trend component Trend, S t Represents the decomposed seasonal component Seasonal, R t Residual represents the load residual component after decomposition;

[0041] According to the initial trend, the seasonal item and the trend item in the historical power load data are respectively smoothed and iterated, including:

[0042] The initial trend subtracting from the historical power load data to obtain the historical power load update data after detrending;

[0043] The historical power load update data of each seasonal cycle position are smoothed separately, and LOESS is used to estimate the average value of each seasonal position to form a preliminary seasonal term:

[0044] By formula Calculate the preliminary seasonal terms

[0045] Calculate the deseasonalized series

[0046] During the seasonal cycle, the preliminary seasonal items Perform moving average so that the average value of the seasonal term in one cycle is 0, and obtain the adjusted seasonal term

[0047] Use adjusted seasonal terms Recalculate the deseasonalized series Then, use a window smaller than h to Perform LOESS smoothing operation to obtain the updated trend term T 1

[0048] Repeat the above steps until the difference between the trend term and the seasonal term in two consecutive iterations is less than the preset threshold ∈, and the iteration is terminated to obtain the seasonal component S t and the trend component T t .

[0049] In the above technical solution, preferably, extracting features from the sequence components includes:

[0050] The trend component T t and the residual component R t The t-1 and t-96 feature points are extracted from the data to obtain the trend feature component and the residual feature component.

[0051] In the above technical solution, preferably, predicting the future sequence value corresponding to the sequence characteristic component according to a preset prediction model includes:

[0052] With D = {(x1, y1), (x2, y2), ..., (x i ,y i )} is the training data set, x i is the input feature vector, y i is the corresponding label, i.e. y i is the trend characteristic component or the residual characteristic component;

[0053] Set initial forecast value By iteratively constructing the tree model, at the tth iteration, the decision tree f is obtained t (x), based on the decision tree, the predicted value of the power load is updated, and the update formula is:

[0054] According to the squared error loss function Correct the prediction error of decision trees;

[0055] Using the mean of the seasonal components of the two periods as the future series value corresponding to the seasonal components;

[0056] The future sequence values ​​corresponding to each of the sequence components are superimposed to obtain the power load forecast value.

[0057] In the above technical solution, preferably, historical power load data of the target area at 15-minute intervals per day is obtained; and / or

[0058] The preset evaluation model is: Among them, y i represents the i-th sample load value, represents the load of the i-th sample predicted by the model, and N represents the total number of samples; and / or

[0059] The expression of the sample loading rate is: Among them, y norm is the rated power of the distribution transformer; and / or

[0060] The expression of the preset heavy overload judgment threshold is:

[0061]

[0062] That is, if the load exceeds 100% for 8 consecutive time points, it is judged that the distribution transformer is overloaded, and if the load rate exceeds 80% but does not exceed 100%, it is judged that the distribution transformer is overloaded.

[0063] The technical solution of the second aspect of the present invention provides a distribution transformer heavy overload prediction system, including: an acquisition module, which is configured to obtain historical power load data at preset daily intervals in a target area; a data decomposition module, which is configured to decompose the historical power load data according to a preset STL time series decomposition algorithm to obtain multiple sequence components; a feature extraction module, which is configured to extract features from the sequence components to obtain sequence feature components; a load prediction module, which is configured to predict future sequence values ​​corresponding to the sequence feature components according to a preset prediction model to obtain power load prediction data; a prediction accuracy evaluation module, which is configured to perform accuracy evaluation on the power load prediction data according to a preset evaluation model to obtain a sample predicted load; a load rate calculation module, which is configured to calculate a sample load rate according to the template predicted load; a heavy overload judgment module, which is configured to perform a heavy overload judgment on the sample load rate according to a preset heavy overload judgment threshold.

[0064] The technical solution of the third aspect of the present invention provides a readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the distribution transformer overload prediction method provided by the technical solution of the first aspect are implemented.

[0065] The technical solution of the fourth aspect of the present invention provides a computer device, including a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to implement the steps of the distribution transformer heavy overload prediction method provided by the technical solution of the first aspect above.

[0066] Compared with the prior art, the advantages of the distribution transformer overload prediction method and device provided by the present invention are: 1. The difference from traditional time series and STL technology is that the XGBoost machine learning model is introduced to further mine the residual data, identify data features that are not captured by the time series, and improve the prediction accuracy.

[0067] 2. External characteristic factors can be introduced into the residual component model to make up for the shortcoming that the standard time series cannot introduce external data characteristics.

[0068] 3. Traditional machine learning models need to introduce a large number of external influencing factors for model training, but this method only uses the sequence’s own data features as model feature input, and has low requirements for input data.

[0069] 4. Since the cyclical component is stable and slowly changing, a more complex model is not used in the prediction of the periodic series. Instead, a simple periodic mean is used as an estimate of future periods.

[0070] 5. Compared with the SVM-STL-LSTM combined model proposed in paper [6], its solution uses SVM for initial prediction, which loses the interpretability of the model, and then uses the STL-LSTM combined model to correct the residual data; the present invention first uses STL to obtain different components and utilizes the interpretability of the STL components; and then uses the XGBoost model to correct the residual data.

[0071] 6. Other combined prediction methods mostly use a weighted voting approach to sum up the prediction results of multiple different models according to their weights, which is significantly different from the staged embedded prediction in the present invention.

[0072] 7. In the prediction of power load data, the combined prediction method has higher prediction accuracy than the separate time series algorithm and machine learning algorithm.

[0073] 8. Compared with traditional machine learning, this method has simpler input data features, simpler periodic component prediction methods, and low requirements for computing resources, making it more suitable for use in actual working environments.

[0074] 9. The model of the present invention has better interpretability and provides stronger support for business decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0076] Figure 1 A flowchart of a distribution transformer heavy overload prediction method according to an embodiment of the present invention is shown;

[0077] Figure 2 A flowchart of step S2 according to an embodiment of the present invention is shown;

[0078] Figure 3 A flowchart of step S22 according to an embodiment of the present invention is shown;

[0079] Figure 4 The structure block diagram of the distribution transformer heavy overload prediction system involved in the embodiment of the present invention is shown;

[0080] Figure 5 The figure shows a decomposition diagram of the STL algorithm sequence involved in the embodiment of the present invention. DETAILED DESCRIPTION

[0081] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0082] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0083] like Figure 1 As shown, according to an embodiment of the present invention, a distribution transformer heavy overload prediction method includes:

[0084] S1, obtaining historical power load data at preset daily intervals in a target area;

[0085] In this step, (1) the historical daily 15-minute power load time series data (i.e., historical power load data) is collected and used as t ={y1, y2, ..., y n} represents the time series data of power load, t is the collection time, that is, the data is collected every 15 minutes, and a total of n times; the rated power of the distribution transformer y norm The collection interval of the original data is 15 minutes, that is, one data is collected every 15 minutes from 00:00 to 23:59 every day, so 96 power load data can be collected every day.

[0086] S2, decomposing the historical power load data according to a preset STL time series decomposition algorithm to obtain multiple sequence components;

[0087] In this step, the STL time series decomposition algorithm is used to decompose the load data to obtain three different series components, namely trend term, seasonal term and residual term. The trend term shows the overall trend of future data, the seasonal term shows the repetition pattern of daily data, and the residual term is the difference between the original data after removing the trend term and seasonal term. If the STL algorithm captures all data characteristics, the residual term should be white noise data that obeys the normal distribution.

[0088] The STL algorithm model formula is as follows: t =T t +S t +R t (t=1, 2, ..., n)

[0089] Among them, y t is the load data, T t Represents the decomposed load trend component Trend, S t Represents the seasonal component of the decomposed load, R t Represents the load residual component after decomposition.

[0090] S3, extracting features from the sequence components to obtain sequence feature components;

[0091] S4, predicting the future sequence value corresponding to the sequence characteristic component according to a preset prediction model to obtain power load prediction data;

[0092] S5, evaluating the accuracy of the power load forecast data according to a preset evaluation model to obtain a sample forecast load;

[0093] S6, calculating a sample load rate according to the sample predicted load;

[0094] S7, performing a heavy overload judgment on the sample load rate according to a preset heavy overload judgment threshold.

[0095] like Figure 2 As shown, in the above technical solution, preferably, the sequence components include trend components, seasonal components and residual components, S2, decomposing the historical power load data according to a preset STL time series decomposition algorithm, including:

[0096] S21, performing initial decomposition on the historical power load data using a local weighted regression scatter point smoothing algorithm to obtain an initial trend;

[0097] In this step, the purpose is to smooth the power load time series data and obtain a preliminary estimate of the trend and seasonal components.

[0098] S22, performing smoothing iteration on the seasonal term and the trend term in the historical power load data according to the initial trend, to obtain the seasonal component and the trend component;

[0099] The purpose of this step is to remove the trend T 0 :Subtract the initial trend from the time series to obtain the detrended series; Seasonal subseries decomposition: Smooth the position of each seasonal cycle separately, and use LOESS to estimate the mean value of each seasonal position to form a preliminary seasonal component.

[0100] S23, calculating the residual component according to the seasonal component and the trend component.

[0101] In the above technical solution, preferably, the historical power load data is initially decomposed using a local weighted regression scatter point smoothing algorithm, including:

[0102] Select a window of length h, for the historical power load data y t ={y1, y2, ..., y n}, when fitting one of the points y i When considering As data points participating in local weighted fitting;

[0103] For each data point in the window, the weight w is calculated using the three-weight function according to the distance between each data point and the center point. j :

[0104] For point y j For the center point y in the window i The weight of , when |y j -y i |≤1,w j =(1-|y j -y i | 3 ) 3 , otherwise w j =0;

[0105] Using the weighted data points in the window, fit the first-order polynomial y = a + bx, by minimizing the weighted least squares objective function ∑ j w j (y j -a-bx j ) 2 To estimate coefficients a and b;

[0106] Move the window along the time series and repeat the above steps to obtain a smoothed initial trend estimate

[0107] like Figure 3 As shown, in the above technical solution, preferably, the expression of the residual component is: R t =y t -T t -S t , where y t is the historical power load data, T t Represents the decomposed trend component Trend, S t Represents the decomposed seasonal component Seasonal, R t Residual represents the load residual component after decomposition;

[0108] S22, performing smoothing iteration on the seasonal item and the trend item in the historical power load data according to the initial trend, including:

[0109] S221, the initial trend subtracting from the historical power load data to obtain the historical power load update data after detrending;

[0110] S222, smooth the historical power load update data of each seasonal cycle position respectively, and use LOESS to estimate the average value of each seasonal position to form a preliminary seasonal term:

[0111] By formula Calculate the preliminary seasonal terms

[0112] Calculate the deseasonalized series

[0113] S223, within the seasonal cycle, the preliminary seasonal items Perform moving average so that the average value of the seasonal term in one cycle is 0, and obtain the adjusted seasonal term

[0114] S224, using adjusted seasonal terms Recalculate the deseasonalized series Then, use a window smaller than h to Perform LOESS smoothing operation to obtain the updated trend term T 1

[0115] S225, repeat the above steps until the difference between the trend term and the seasonal term in two adjacent iterations is less than the preset threshold ∈, and the iteration is terminated to obtain the seasonal component S t and the trend component T t .

[0116] In the above technical solution, preferably, extracting features from the sequence components includes:

[0117] The trend component T t and the residual component R t The t-1 and t-96 feature points are extracted from the data to obtain the trend feature component and the residual feature component.

[0118] In the above technical solution, preferably, predicting the future sequence value corresponding to the sequence characteristic component according to a preset prediction model includes:

[0119] With D = {(x1, y1), (x2, y2), ..., (x i ,y i )} is the training data set, x i is the input feature vector, y i is the corresponding label, i.e. y i is the trend characteristic component or the residual characteristic component;

[0120] Set initial forecast value By iteratively constructing the tree model, at the tth iteration, the decision tree f is obtained t (x), based on the decision tree, the predicted value of the power load is updated, and the update formula is:

[0121] According to the squared error loss function Correct the prediction error of decision trees;

[0122] For the prediction of seasonal components, since periodic data has the characteristics of stability and slow change, the mean of the seasonal components of two periods is used as the future series value corresponding to the seasonal components;

[0123] The future sequence values ​​corresponding to each of the sequence components are superimposed to obtain the power load forecast value.

[0124] In this embodiment, the sequence values ​​of the trend component, period component, and residual component sequences obtained by the above decomposition are predicted for the next day. Among them:

[0125] Trend prediction, that is, for the trend component T t , use the XGBoost regression model to train and predict data. XGBoost is a boosting tree algorithm, and its basic idea is to form a strong learner by combining multiple weak learners.

[0126] In the above technical solution, preferably, historical power load data of the target area at 15-minute intervals per day is obtained; and / or

[0127] The preset evaluation model is: Among them, y i represents the i-th sample load value, represents the load of the i-th sample predicted by the model, and N represents the total number of samples; and / or

[0128] The expression of the sample loading rate is: Among them, y norm is the rated power of the distribution transformer; and / or

[0129] The expression of the preset heavy overload judgment threshold is:

[0130]

[0131] That is, if the load at 8 consecutive time points (i.e. 2 hours) exceeds 100% load factor, it is judged that the distribution transformer is overloaded, and if the load factor exceeds 80% but does not exceed 100%, it is judged that the distribution transformer is overloaded.

[0132] like Figure 4 As shown, according to another embodiment of the present invention, a distribution transformer overload prediction system 100 includes:

[0133] An acquisition module 10 is configured to acquire historical power load data at preset daily intervals in a target area;

[0134] The data decomposition module 20 is configured to decompose the historical power load data according to a preset STL time series decomposition algorithm to obtain a plurality of sequence components;

[0135] A feature extraction module 30 is configured to extract features from the sequence components to obtain sequence feature components;

[0136] The load prediction module 40 is configured to predict the future sequence value corresponding to the sequence characteristic component according to a preset prediction model to obtain power load prediction data;

[0137] The prediction accuracy evaluation module 50 is configured to evaluate the accuracy of the power load prediction data according to a preset evaluation model to obtain a sample predicted load;

[0138] A load rate calculation module 60 is configured to calculate a sample load rate according to the sample predicted load;

[0139] The heavy overload judgment module 70 is configured to perform a heavy overload judgment on the sample load rate according to a preset heavy overload judgment threshold. Specific embodiments

[0141] The load data of a distribution transformer in a certain city from March 1, 2023 to March 18, 2024 is collected every 15 minutes, and 96 data points are obtained every day.

[0142] The data format is as follows:

[0143] Distribution transformer number Collection time <![CDATA[Electric load y t > 63***56** 2023-03-01 00:00:00 0.0917 63***56** 2023-03-01 00:15:00 0.097 63***56** 2023-03-01 00:30:00 0.1306 63***56** 2023-03-01 00:45:00 0.0935 63***56** 2023-03-01 01:00:00 0.0963 63***56** 2023-03-01 01:15:00 0.0872 … … … 63***56** 2024-02-08 00:45:00 0.3448 63***56** 2024-05-31 23:00:00 0.3172 63***56** 2024-05-31 23:15:00 0.2386 63***56** 2024-05-31 23:30:00 0.2751 63***56** 2024-05-31 23:45:00 0.2565

[0144] The load data from March 1 to March 15, 2023 is used as the training set to train the model.

[0145] The load data from March 16 to March 18, 2023 is used as a test set to test the accuracy of the load forecasting method proposed in the present invention.

[0146] Take the load data from March 1 to March 15, 2023 as an example, with a total of 96*15 data points. Use the STL algorithm to decompose the data sequence into trend sequence components, seasonal sequence components, and residual sequence components. The decomposed results are as follows: Figure 5 shown.

[0147] Table 1 Comparison of error values ​​of different methods

[0148]

[0149]

[0150] As can be seen from the table, the XGBoost prediction method based on STL decomposition has the smallest MSE and MAE, with an MSE of 0.0049 and a MAE of 0.0511. Compared with the random forest method and the original XGBoost method, the MSE and MAE of the prediction method of the present invention are improved by 14% and 6% respectively, and the prediction of the present invention has a higher accuracy.

[0151] Based on the above Figures 1 to 3 The method shown, accordingly, an embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the distribution transformer overload prediction method of any of the above embodiments are implemented.

[0152] Based on this understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, USB flash drive, mobile hard disk, etc.), including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of the present application.

[0153] Based on the above Figures 1 to 3 The method shown, and Figure 4 The virtual device embodiment shown, in order to achieve the above-mentioned purpose, the embodiment of the present application also provides a computer device, including a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to implement the steps of the distribution transformer overload prediction method of any of the above-mentioned embodiments.

[0154] Optionally, the computer device may further include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a WI-FI module, etc. The user interface may include a display, an input unit such as a keyboard, etc., and the optional user interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a Bluetooth interface, a WI-FI interface), etc.

[0155] Those skilled in the art will appreciate that the computer device structure provided in this embodiment does not limit the computer device, and may include more or fewer components, or a combination of certain components, or different component arrangements.

[0156] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages and saves the hardware and software resources of the computer device, and supports the operation of information processing programs and other software and / or programs. The network communication module is used to realize communication between the components inside the storage medium, and communication with other hardware and software in the physical device.

[0157] In the present invention, the terms "first", "second", and "third" are used for descriptive purposes only and should not be understood as indicating or implying relative importance; the term "plurality" refers to two or more, unless otherwise clearly defined. The terms "installed", "connected", "connected", "fixed", etc. should be understood in a broad sense. For example, "connected" can be a fixed connection, a detachable connection, or an integral connection; "connected" can be a direct connection or an indirect connection through an intermediate medium. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0158] In the description of the present invention, it is necessary to understand that the directions or positional relationships indicated by terms such as “upper”, “lower”, “left”, “right”, “front” and “back” are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or unit referred to must have a specific direction, be constructed and operated in a specific orientation, and therefore, cannot be understood as a limitation on the present invention.

[0159] In the description of this specification, the description of the terms "one embodiment", "some embodiments", "specific embodiments", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0160] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A distribution transformer heavy overload prediction method, characterized in that: include: Obtain historical power load data at preset daily intervals within the target area; Decomposing the historical power load data according to a preset STL time series decomposition algorithm to obtain multiple sequence components; Extracting features from the sequence components to obtain sequence feature components; Predicting the future sequence value corresponding to the sequence characteristic component according to a preset prediction model to obtain power load prediction data; Performing an accuracy evaluation on the power load forecast data according to a preset evaluation model to obtain a sample forecast load; Calculating a sample load rate based on the sample predicted load; A heavy overload judgment is performed on the sample load rate according to a preset heavy overload judgment threshold.

2. The distribution transformer heavy overload prediction method according to claim 1 is characterized in that: The sequence components include trend components, seasonal components and residual components. The historical power load data is decomposed according to a preset STL time series decomposition algorithm, including: Performing initial decomposition on the historical power load data using a local weighted regression scatter point smoothing algorithm to obtain an initial trend; According to the initial trend, the seasonal term and the trend term in the historical power load data are respectively smoothed and iterated to obtain the seasonal component and the trend component; The residual component is calculated according to the seasonal component and the trend component.

3. The distribution transformer overload prediction method according to claim 2 is characterized in that: The historical power load data is initially decomposed using a local weighted regression scatter point smoothing algorithm, including: Select a window of length h, for the historical power load data y t ={y1, y2, ..., y n }, when fitting one of the points y i When considering As data points participating in local weighted fitting; For each data point in the window, the weight w is calculated using the three-weight function according to the distance between each data point and the center point. j : For point y j For the center point y in the window i The weight of j -y i |≤1, w j =(1-|y j -y i | 3 ) 3 , otherwise w j =0; Using the weighted data points in the window, fit the first-order polynomial y = a + bx, and minimize the weighted least squares objective function ∑ j w j (y j -a-bx j ) 2 To estimate coefficients a and b; Move the window along the time series and repeat the above steps to obtain a smoothed initial trend estimate 4. The method for predicting heavy overload of distribution transformer according to claim 3 is characterized in that: The expression of the residual component is: R t =y t -T t -S t , where y t is the historical power load data, T t Represents the decomposed trend component Trend, S t Represents the decomposed seasonal component Seasonal, R t Residual represents the load residual component after decomposition; According to the initial trend, the seasonal item and the trend item in the historical power load data are respectively smoothed and iterated, including: The initial trend subtracting from the historical power load data to obtain the historical power load update data after detrending; The historical power load update data of each seasonal cycle position are smoothed separately, and LOESS is used to estimate the average value of each seasonal position to form a preliminary seasonal term: By formula Calculate the preliminary seasonal terms Calculate the deseasonalized series During the seasonal cycle, the preliminary seasonal items Perform moving average so that the average value of the seasonal term in one cycle is 0, and obtain the adjusted seasonal term Use adjusted seasonal terms Recalculate the deseasonalized series Then, use a window smaller than h to Perform LOESS smoothing operation to obtain the updated trend term T 1 Repeat the above steps until the difference between the trend term and the seasonal term in two consecutive iterations is less than the preset threshold ∈, and the iteration is terminated to obtain the seasonal component S t and the trend component T t .

5. The method for predicting heavy overload of distribution transformer according to claim 4, characterized in that: Extracting features from the sequence components includes: The trend component T t and the residual component R t The t-1 and t-96 feature points are extracted from the data to obtain the trend feature component and the residual feature component.

6. The method for predicting heavy overload of distribution transformer according to claim 5, characterized in that: Predicting the future sequence value corresponding to the sequence characteristic component according to the preset prediction model includes: With D = {(x1, y1), (x2, y2), ..., (x i ,y i )} is the training data set, x i is the input feature vector, y i is the corresponding label, i.e. y i is the trend characteristic component or the residual characteristic component; Set initial forecast value By iteratively constructing the tree model, at the tth iteration, the decision tree f is obtained t (x), based on the decision tree, the predicted value of power load is updated, and the update formula is: According to the squared error loss function Correct the prediction error of the decision tree: Using the mean of the seasonal components of the two periods as the future series value corresponding to the seasonal components; The future sequence values ​​corresponding to each of the sequence components are superimposed to obtain the power load forecast value.

7. The method for predicting heavy overload of distribution transformer according to claim 6, characterized in that: Obtain historical power load data at 15-minute intervals per day in the target area; and / or The preset evaluation model is: Among them, y i represents the i-th sample load value, represents the load of the i-th sample predicted by the model, and N represents the total number of samples; and / or The expression of the sample loading rate is: Among them, y norm is the rated power of the distribution transformer; and / or The expression of the preset heavy overload judgment threshold is: That is, if the load exceeds 100% for 8 consecutive time points, it is judged that the distribution transformer is overloaded, and if the load rate exceeds 80% but does not exceed 100%, it is judged that the distribution transformer is overloaded.

8. A distribution transformer overload prediction system, characterized in that: include: An acquisition module is configured to acquire historical power load data at a preset daily interval in a target area; A data decomposition module is configured to decompose the historical power load data according to a preset STL time series decomposition algorithm to obtain a plurality of sequence components; A feature extraction module is configured to extract features from the sequence components to obtain sequence feature components; A load forecasting module is configured to predict future sequence values ​​corresponding to the sequence characteristic components according to a preset forecasting model to obtain power load forecasting data; A prediction accuracy evaluation module is configured to evaluate the accuracy of the power load prediction data according to a preset evaluation model to obtain a sample predicted load; A load rate calculation module, configured to calculate a sample load rate according to the sample predicted load; The heavy overload judgment module is configured to perform a heavy overload judgment on the sample load rate according to a preset heavy overload judgment threshold.

9. A readable storage medium, characterized in that: A computer program is stored thereon, and when the program is executed by a processor, the steps of the distribution transformer overload prediction method as described in any one of claims 1 to 7 are implemented.

10. A computer device, characterized in that: It comprises a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to implement the steps of the distribution transformer overload prediction method as described in any one of claims 1 to 7.

Images