Consumer relation identification method based on probability prediction and energy conservation verification

Abstract

The invention relates to a platform user relation identification method based on probability prediction and energy conservation verification. The method comprises the following steps: carrying out feature extraction on a power utilization behavior and a geographic position of a user, and dividing a platform area to which the user belongs based on a machine learning model of a support vector machine; converting into probability output on the basis of the divided classification, and calculating a prediction probability; calculating the sum of electricity consumption of the users, verifying a prediction result in combination with an energy conservation principle, and identifying and correcting existing prediction errors; calculating a deviation value between the power supply quantity of the transformer and the total power consumption of the users, setting deviation tolerance, and prompting adjustment and correction if the deviation exceeds the range; and after each adjustment, recalculating the sum of the power consumption of the users and comparing the sum with the power supply quantity of the transformer. According to the method, the accuracy and the reliability of the user relationship identification are improved through user area division and probability prediction in combination with the physical law of energy conservation; meanwhile, the management efficiency and the operation maintenance level of the power distribution system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of station-user relationship identification, and in particular to a station-user relationship identification method based on probability prediction and energy conservation verification. Background Art

[0002] The low-voltage distribution system is an important part of the power system, responsible for transmitting electrical energy from distribution transformers to end users. Due to the complexity and widespread distribution of low-voltage distribution networks, traditional management methods have difficulty accurately identifying the relationship between substations and users, resulting in low management efficiency and increased difficulty in fault diagnosis and maintenance. With the development of smart grids, data mining technology has gradually been applied to distribution networks. By analyzing massive electricity consumption data and combining physical laws, it can effectively obtain the topological structure information of low-voltage substations, improve the status monitoring capabilities and management quality of distribution systems, and achieve efficient operation and maintenance and fault diagnosis.

[0003] The invention patent with publication number CN110389267A discloses a method for identifying the relationship between low-voltage substations and households based on data collected by smart energy meters, including: collecting and obtaining the original voltage data of a preset number of gateway smart energy meters and user smart energy meters in the low-voltage substation to be identified for a preset number of days and cleaning them to obtain the valid data of each smart energy meter; according to the valid data of each smart energy meter obtained, obtaining the daily average voltage and standard deviation of the smart energy meter; performing correlation analysis on the average value series and standard deviation series of the user smart energy meter and the gateway smart energy meter respectively; if the analysis results show that the correlation coefficients of the average value and standard deviation of a certain user smart energy meter and a certain gateway smart energy meter are both the maximum, it is determined to be a match. The household matching of this invention lacks effective physical law restrictions.

[0004] Therefore, providing a reliable identification method with effective physical law constraints is an issue that needs to be addressed urgently. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a method for identifying household relationships based on probability prediction and energy conservation verification.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] According to one aspect of the present invention, a method for identifying a user relationship based on probability prediction and energy conservation verification is provided, the method comprising the following steps:

[0008] S1. Extract features of user electricity usage behavior and geographic location;

[0009] S2, dividing the area to which the user belongs through a machine learning model based on a support vector machine according to the extracted features;

[0010] S3. Based on the classification of the area to which the user belongs, convert the classification result into a probability output and calculate the predicted probability;

[0011] S4. Calculate the total power consumption of users based on the predicted probability and verify the predicted results in combination with the principle of energy conservation to identify and correct the existing prediction errors;

[0012] S5. Calculate the deviation between the transformer's power supply and the user's total power consumption, and set the deviation tolerance with reference to the line loss in the power supply area. If the deviation exceeds the range, prompt for adjustment and correction;

[0013] S6. After each adjustment, recalculate the total power consumption of users and compare it with the power supply of the transformer.

[0014] As a preferred technical solution, the user's electricity consumption behavior characteristics are extracted based on the user's electricity load data.

[0015] As a preferred technical solution, the power load data is subjected to time series analysis and power consumption time characteristics are extracted, wherein the power consumption time characteristics include peak load, peak power consumption period, average daily load, nighttime power consumption ratio and power consumption volatility.

[0016] As a preferred technical solution, the power load data is analyzed in the frequency domain to extract high-frequency and low-frequency components, thereby capturing the periodic changes in the user's power consumption.

[0017] As a preferred technical solution, the geographical location includes the user installation address and the electrical equipment installation address.

[0018] As a preferred technical solution, in the said area division, the support vector machine model uses a radial basis kernel function to map the user's feature data from the original space to a high-dimensional space.

[0019] As a preferred technical solution, the probability prediction is performed by fitting the decision function value of the support vector machine using logistic regression to generate the probability of belonging to the positive class. Problems that require probability prediction include binary classification problems and multi-classification problems.

[0020] As an optimal technical solution, the binary classification problem directly outputs the probability value belonging to the positive class; the multi-classification problem adopts an extension method based on a one-to-one strategy, treating the classification task between each category pair as a binary classification problem, performing probability calibration on each binary classification model separately, and integrating all calibrated results into a multi-classification probability distribution.

[0021] As an optimal technical solution, the power supply of the transformer is calculated by retrieving the historical operation data of the transformer in the substation, wherein the historical operation data includes the load curve, maximum power supply power, maximum output power, average load value and average power output during daily operation of the transformer; the upper limit of the total power supply load of the substation is calculated based on the maximum output power and the average load value.

[0022] As a preferred technical solution, the adjustment of the predicted probability is comprehensively corrected according to the power deviation, the power deviation is used as a feedback signal and the user is reallocated to other possible areas, and the user's area affiliation is adjusted to reduce the power deviation.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] 1. The present invention calculates the power supply and user power consumption through substation division and prediction probability, and combines the physical law of energy conservation to determine whether the power deviation is within the range, thereby improving the accuracy and reliability of substation-user relationship identification; at the same time, it improves the management efficiency and operation and maintenance level of the distribution system.

[0025] 2. The present invention sets the deviation tolerance by referring to the line loss in the power supply area, determines whether the deviation of the total power supply and power consumption is within the range, and determines the accuracy of the station-user relationship identification.

[0026] 3. The present invention further improves the accuracy of household relationship identification by combining probability prediction and energy conservation verification.

[0027] 4. The present invention uses logistic regression to fit the decision function value of the support vector machine to generate the probability of belonging to the positive class, thereby improving the accuracy of household relationship recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a schematic diagram of the process flow of identifying the relationship between a platform and a user of the present invention;

[0029] Figure 2 It is a schematic diagram of the probability prediction and energy conservation verification process of the present invention. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0031] The low-voltage distribution system is an important part of the power system, responsible for transmitting electrical energy from distribution transformers to end users. Due to the complexity and widespread distribution of low-voltage distribution networks, traditional management methods have difficulty accurately identifying the relationship between substations and users, resulting in low management efficiency and increased difficulty in fault diagnosis and maintenance. With the development of smart grids, data mining technology has gradually been applied to distribution networks. By analyzing massive electricity consumption data and combining physical laws, it can effectively obtain the topological structure information of low-voltage substations, improve the status monitoring capabilities and management quality of distribution systems, and achieve efficient operation and maintenance and fault diagnosis.

[0032] The present invention relates to low-voltage power distribution system data analysis and identification of user relationships in substations, specifically to obtaining low-voltage power distribution substation topology information by combining data mining methods and physical laws to improve the management quality and operation efficiency of the power distribution network. The aim is to accurately identify user relationships in low-voltage power distribution substations through the combination of data mining and physical laws, and improve the management efficiency and operation and maintenance level of the power distribution network.

[0033] Therefore, the present invention provides a method for identifying the relationship between transformers and users based on probability prediction and energy conservation verification. The present invention calculates the power supply and user power consumption through substation division and prediction probability, and combines the physical law of energy conservation to determine whether the power deviation is within the range, thereby improving the accuracy and reliability of transformer-user relationship identification; at the same time, it improves the management efficiency and operation and maintenance level of the power distribution system. The present invention sets the deviation tolerance by referring to the line loss in the power supply area, determines whether the deviation of the total power supply and power consumption is within the range, and determines the accuracy of transformer-user relationship identification. The present invention further improves the accuracy of transformer-user relationship identification by combining probability prediction and energy conservation verification. The present invention uses logistic regression to fit the decision function value of the support vector machine to generate the probability of belonging to the positive class, thereby improving the accuracy of transformer-user relationship identification. Example 1

[0034] like Figure 1 , Figure 2 As shown, a method for identifying the relationship between a station and a user based on probability prediction and energy conservation verification is characterized in that the method comprises the following steps:

[0035] S1. Extract features of user electricity usage behavior and geographic location;

[0036] S2, dividing the area to which the user belongs through a machine learning model based on a support vector machine according to the extracted features;

[0037] S3. Based on the classification of the area to which the user belongs, convert the classification result into a probability output and calculate the predicted probability;

[0038] S4. Calculate the total power consumption of users based on the predicted probability and verify the predicted results in combination with the principle of energy conservation to identify and correct the existing prediction errors;

[0039] S5. Calculate the deviation between the transformer's power supply and the user's total power consumption, and set the deviation tolerance with reference to the line loss in the power supply area. If the deviation exceeds the range, prompt for adjustment and correction;

[0040] S6. After each adjustment, recalculate the total power consumption of users and compare it with the power supply of the transformer.

[0041] The power load data is subjected to time series analysis and power consumption time characteristics are extracted, wherein the power consumption time characteristics include peak load, peak power consumption period, daily average load, nighttime power consumption ratio and power consumption volatility.

[0042] The power load data is analyzed in the frequency domain to extract high-frequency and low-frequency components, thereby capturing the periodic changes in the user's power consumption.

[0043] The geographical location includes the user installation address and the electrical equipment installation address.

[0044] In this embodiment, based on the low-voltage power distribution network data, features that can effectively reflect the structure of the substation and user behavior are extracted. Among them, the power load curve reflects the daily power consumption rules and behavior patterns of users, and the possibility that users of the same type in the substation have similar power consumption patterns. Time series analysis is performed on the user's power load data to extract the time characteristics of power consumption, including peak load, peak power consumption period, average daily load, nighttime power consumption ratio, power consumption volatility and other characteristics, in order to identify the user's peak power consumption period, basic power demand, power consumption stability, etc. At the same time, high-frequency and low-frequency components are extracted through frequency domain analysis to capture the periodic changes in user power consumption and enrich the feature space of power consumption data.

[0045] Extract user geographic information as the judgment feature of its substation area, considering that users in the substation area usually present certain clustering characteristics in space. Extract the precise geographic coordinates of each user through the user installation address and power equipment installation address provided by the power company or relevant departments. Further calculate the spatial distance between each user and the surrounding transformer or substation center. Through distance analysis, the user can be associated with the nearest substations to establish a preliminary correspondence.

[0046] The support vector machine model in the area division uses the radial basis kernel function to map the user's feature data from the original space to the high-dimensional space. The predicted probability is fitted to the decision function value of the support vector machine by using logistic regression to generate the probability of belonging to the positive class. Problems that require predicted probability include binary classification problems and multi-classification problems.

[0047] The binary classification problem directly outputs the probability value belonging to the positive class; the multi-classification problem adopts an extended method based on a one-to-one strategy, treating the classification task between each category pair as a binary classification problem, performing probability calibration on each binary classification model respectively, and integrating all calibrated results into a multi-classification probability distribution.

[0048] In this embodiment, the features of multi-source data such as user electricity consumption behavior and geographic location are extracted, and the area to which each user belongs is predicted through a machine learning model based on a support vector machine. The radial basis kernel function is selected to map the user's feature data from the original space to a high-dimensional space. In the high-dimensional space, the support vector machine algorithm achieves optimal classification by finding a classification hyperplane that maximizes the interval between users in different areas. This hyperplane not only classifies users into different areas, but also maximizes the distance from the sample to the classification boundary, thereby ensuring the robustness and generalization ability of the classification.

[0049] Based on the prediction of the area, combined with the probability prediction model, on the basis of the classification of the area to which the user belongs, the probability calibration based on logistic regression is used to convert the classification results into probability outputs, and the probability distribution of the area to which the user belongs is calculated to improve the interpretability and flexibility of recognition. Logistic regression is used to fit the decision function value of the support vector machine to generate the probability of belonging to the positive class. For each data point x i And the decision function output value f(x i ), the probability calculation formula is:

[0050]

[0051] Among them, A and B are the parameters of the logistic regression model that need to be learned. In order to solve the optimal values ​​of A and B, the following log-likelihood function is defined:

[0052]

[0053] By maximizing the log-likelihood function, the optimal values ​​of parameters A and B can be obtained. The trained logistic regression model is applied to the decision function output of the support vector machine, and the decision function value of each data point is calibrated using formula (1) to generate the probability value of the data point belonging to the positive class. For binary classification problems, the probability value belonging to the positive class is directly output; for multi-classification problems, an extension method based on a one-to-one strategy is adopted, that is, for a classification problem with K categories, the classification task between each category pair is regarded as a binary classification problem, and a total of K(K-1) / 2 support vector machine models are generated. Then, the probability of each binary classification model is calibrated separately, and all calibrated results are integrated into a multi-classification probability distribution. Through calibration based on logistic regression, the model output changes from a single classification label to a probability distribution. The prediction result of each data point is not only the divided area, but also the probability of the data point belonging to different areas.

[0054] The power supply of the transformer is calculated by retrieving the historical operation data of the transformer in the substation, wherein the historical operation data includes the load curve of the transformer, the maximum power supply power, the maximum output power, the average load value and the average power output during daily operation; the upper limit of the total power supply load of the substation is calculated based on the maximum output power and the average load value.

[0055] The adjustment of the probability prediction is comprehensively corrected according to the power deviation, the power deviation is used as a feedback signal and the user is reallocated to other possible areas, and the area affiliation of the user is adjusted to reduce the power deviation.

[0056] In this embodiment, after obtaining the probability distribution of the substation to which the user belongs, the prediction results are further verified and corrected by the energy conservation principle to identify and correct possible prediction errors, ensuring that the prediction results are more accurate and reliable. The energy conservation principle means that the total load of a substation should be equivalent to the actual output power of the transformer connected to the substation, that is, the power supply of the transformer is consistent with the total power consumption of all users in the substation.

[0057] Calculate the total power upper limit of each substation, which is determined by the actual power supply capacity of the transformer connected to the substation. Retrieve the historical operation data of the substation transformer to obtain the actual record of its power supply capacity, including the transformer's load curve, maximum power supply power, and average power output during daily operation. Calculate the total power supply load upper limit of the substation based on the transformer's maximum output power and the average load value in historical operation. If the transformer's historical power supply data is abnormal, such as sudden failure or power shortage, these data will be eliminated or smoothed to ensure that the power supply power data used can accurately reflect the power supply capacity under normal circumstances.

[0058] Based on the power calculation of the substation, the total power of the users is calculated. Based on the high-probability matching results given by the probability prediction model, the users identified in the preliminary prediction stage are classified according to their respective substations. Through the user's electricity consumption information collection system, the historical electricity consumption data of each user is obtained, including daily electricity consumption curves, peak loads, valley electricity consumption and other information. The electricity consumption data of each user in the substation is accumulated to calculate the total user electricity power of the substation. This power value represents the total load of the users identified in the prediction stage within a certain time period. In order to ensure the accuracy of the calculation results, the time period of the user power sum needs to be consistent with the power supply power data of the substation transformer. Usually the same time window, such as one hour, one day or one week, is selected for power comparison.

[0059] Compare the predicted total power of users with the power supply power of the transformer in the substation area to identify possible deviations in the prediction results and provide a basis for further adjustments. Directly compare the power supply power of the transformer in the substation area with the total power of users. If the total power of the user is close to the power supply power of the transformer, it means that the preliminary predicted user allocation result is relatively accurate; conversely, if the gap is large, there may be an error. In order to quantify the power difference, the deviation value between the power supply power and the total power of the user is calculated, and the deviation tolerance is set with reference to the line loss in the power supply area. Deviations within this range can be considered that energy conservation is basically established, while deviations beyond this range indicate that the prediction needs further adjustment.

[0060] In order to ensure that the final area identification results comply with the principle of energy conservation, a power correction mechanism based on probability redistribution and iterative optimization is proposed. Through continuous adjustment and feedback improvement, the area identification results are gradually optimized to make them more accurate. For areas where the power exceeds or is far below the power supply capacity, users are reallocated to other possible areas by adjusting the probability distribution of the area. When adjusting the area belonging to high-probability users, it is necessary to check the neighboring areas to which these users may be assigned at the same time to ensure that the neighboring areas have sufficient power carrying capacity. For areas with insufficient power, the allocation results of users with low and medium probabilities are first checked. The power consumption characteristics of these users may not completely match the characteristics of the area, but the possibility that they belong to the area is not ruled out. Based on the verification of low-probability users, some low-probability users are reallocated to areas with insufficient power, so that the power load in the area is closer to the power supply capacity of the transformer.

[0061] We further introduce an iterative optimization mechanism based on feedback signals, taking the power deviation as the feedback signal, gradually updating the area identification model, adjusting the user's area affiliation to reduce the power deviation and improve the accuracy of area allocation. Figure 2As shown. During the iteration process, the model is optimized by adjusting the decision boundary and smoothing the probability distribution according to the energy conservation error. In the support vector machine classification model, the setting of the decision boundary will affect the probability of the user's substation belonging. By feeding back the energy conservation error and adjusting the decision boundary, the user's allocation is more consistent with the actual load distribution of the substation. For the substation identification probability distribution, the smoothing technology is applied to make the model gradually converge during the iteration process to avoid fluctuations caused by drastic adjustments. At the same time, the smoothing of the probability distribution can also enhance the robustness of the model and reduce the errors caused by noise or abnormal data.

[0062] After each adjustment of the user allocation result, the system recalculates the total power consumption of the users and compares it with the power supply of the transformer. When the power deviation is controlled within a reasonable range, it indicates that the prediction model and correction mechanism have achieved the expected effect. Through multiple iterative corrections, the system gradually optimizes the accuracy of user allocation to ensure that the final result is consistent with the actual power supply situation in the substation.

[0063] Example 2

[0064] The example is 225 users in two adjacent areas, and the electricity consumption data collected from the users within one month is analyzed and judged. After processing the data and extracting the characteristic parameters, a probability prediction model is established, and the model is tested on the test set. The prediction results are shown in Tables 1 and 2. Among them, Table 1 gives the test results of the training model on the test set, where the classification accuracy, precision and recall are 0.81818, 0.76923 and 0.90909 respectively. Compared with the category prediction probabilities given in Table 2, the probability gap between the two categories of some users is small, which is the main source of error.

[0065] Table 1 Confusion matrix of initial prediction model of substation area

[0066]

[0067] Table 2 Class prediction probability of the test set

[0068]

[0069]

[0070] Based on the above prediction results, the electricity consumption of users in the substation was summed up and it was found that the total electricity consumption of users in the first substation in one month exceeded the power supply of this substation, which did not conform to the laws of physics. The category prediction probability results were adjusted and energy conservation was verified, as shown in Table 3. Among them, the energy conservation satisfaction of the prediction results of users 4 and 7 adjusted by scheme 1 was the best. The final model prediction results are shown in Table 4. The classification accuracy, precision and recall rates were 0.90909, 0.84615 and 1 respectively, and the accuracy was significantly improved.

[0071] Table 3 Energy conservation verification results

[0072]

[0073]

[0074] Note: The unit of electricity consumption is kWh.

[0075] Table 4 Confusion matrix of initial prediction model of substation area

[0076]

[0077] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for identifying household relationships based on probability prediction and energy conservation verification, characterized in that: The method comprises the following steps: S1. Extract features of user electricity usage behavior and geographic location; S2, dividing the area to which the user belongs through a machine learning model based on a support vector machine according to the extracted features; S3. Based on the classification of the area to which the user belongs, convert the classification result into a probability output and calculate the predicted probability; S4. Calculate the total power consumption of users based on the predicted probability and verify the predicted results in combination with the principle of energy conservation to identify and correct the existing prediction errors; S5. Calculate the deviation between the transformer's power supply and the user's total power consumption, and set the deviation tolerance with reference to the line loss in the power supply area. If the deviation exceeds the range, prompt for adjustment and correction; S6. After each adjustment, recalculate the total power consumption of users and compare it with the power supply of the transformer.

2. According to claim 1, a method for identifying household relationships based on probability prediction and energy conservation verification is characterized in that: Extract user electricity consumption behavior characteristics based on user electricity load data.

3. The method for identifying household relationships based on probability prediction and energy conservation verification according to claim 2, characterized in that: The power load data is subjected to time series analysis and power consumption time characteristics are extracted, wherein the power consumption time characteristics include peak load, peak power consumption period, daily average load, nighttime power consumption ratio and power consumption volatility.

4. The method for identifying household relationships based on probability prediction and energy conservation verification according to claim 2, characterized in that: The power load data is analyzed in the frequency domain to extract high-frequency and low-frequency components, thereby capturing the periodic changes in the user's power consumption.

5. The method for identifying household relationships based on probability prediction and energy conservation verification according to claim 1, characterized in that: The geographical location includes the user installation address and the electrical equipment installation address.

6. The method for identifying household relationships based on probability prediction and energy conservation verification according to claim 1, characterized in that: In the station area division, the support vector machine model uses a radial basis kernel function to map the user's feature data from the original space to a high-dimensional space.

7. The method for identifying household relationships based on probability prediction and energy conservation verification according to claim 6, characterized in that: The probability prediction is performed by fitting the decision function value of the support vector machine using logistic regression to generate the probability of belonging to the positive class. Problems that require probability prediction include binary classification problems and multi-classification problems.

8. The method for identifying household relationships based on probability prediction and energy conservation verification according to claim 7, characterized in that: The binary classification problem directly outputs the probability value belonging to the positive class; the multi-classification problem adopts an extended method based on a one-to-one strategy, treating the classification task between each category pair as a binary classification problem, performing probability calibration on each binary classification model respectively, and integrating all calibrated results into a multi-classification probability distribution.

9. The method for identifying household relationships based on probability prediction and energy conservation verification according to claim 1, characterized in that: The power supply of the transformer is calculated by retrieving the historical operation data of the transformer in the substation, wherein the historical operation data includes the load curve of the transformer, the maximum power supply power, the maximum output power, the average load value and the average power output during daily operation; the upper limit of the total power supply load of the substation is calculated based on the maximum output power and the average load value.

10. The method for identifying household relationships based on probability prediction and energy conservation verification according to claim 1, characterized in that: The adjustment of the predicted probability is comprehensively corrected according to the power deviation, the power deviation is used as a feedback signal and the user is reallocated to other possible areas, and the area affiliation of the user is adjusted to reduce the power deviation.

Images