Distributed photovoltaic anomaly identification method based on artificial intelligence technology

Abstract

The invention discloses a distributed photovoltaic anomaly identification method based on an artificial intelligence technology, which overcomes the problem of low detection accuracy of distributed photovoltaic operation states of users in the prior art, and comprises the following steps: obtaining and processing total data of resident distributed photovoltaic users to obtain original data; user historical similar date power generation data and adjacent user power generation data are selected from the original data, and a resident distributed photovoltaic state evaluation model is constructed; adjusting the key parameters of the positioning state category by combining offline troubleshooting; collecting user distributed photovoltaic real-time operation data, judging whether the operation data is abnormal or not and the data abnormality type, and evaluating the resident distributed photovoltaic state; different service processing strategies are adopted for users in different abnormal states, and abnormal operation data are repaired. Big data filtering and screening and human-assisted screening and troubleshooting are used to accurately identify distributed photovoltaic real-time operation abnormal data and a user distributed photovoltaic state.

Description

Technical Field

[0001] The present invention relates to the field of distributed photovoltaic detection technology, and in particular to a distributed photovoltaic anomaly recognition method based on artificial intelligence technology. Background Art

[0002] With the large-scale development and utilization of new energy in the domestic market, the photovoltaic industry has become increasingly mature. Distributed photovoltaic power generation system, also known as distributed power generation or distributed energy supply, refers to the configuration of a smaller photovoltaic power generation system at the user site or near the power consumption site to meet the needs of specific users, support the economic operation of the existing distribution network, or meet both requirements at the same time. The basic equipment of the distributed photovoltaic power generation system includes photovoltaic cell components, photovoltaic array brackets, DC junction boxes, DC distribution cabinets, grid-connected inverters, AC distribution cabinets and other equipment, as well as power supply system monitoring devices and environmental monitoring devices. Its operating mode is that under the condition of solar radiation, the solar cell module array of the photovoltaic power generation system converts the output electricity of solar energy into electricity through the DC junction box and sends it to the DC distribution cabinet, which is then converted into AC power by the grid-connected inverter to supply the building's own load. The excess or insufficient power is regulated by connecting to the power grid.

[0003] At present, the management methods for distributed photovoltaic power generation for residents are relatively simple. In terms of dealing with photovoltaic overcapacity power generation, there is only an audit rule of "monthly power generation ≥ annual maximum sunshine hours / 12" in the audit profession, and the accuracy rate is not high; in terms of dealing with low photovoltaic power generation efficiency, no corresponding countermeasures have been taken, which may cause users to believe that the amount of subsidies and electricity purchase funds issued is incorrect, thus causing the risk of complaints. Therefore, it is urgent to take effective measures to improve the management level of residential photovoltaic power generation and ensure the healthy and orderly development of the residential photovoltaic industry. Summary of the invention

[0004] The purpose of the present invention is to overcome the problems in the prior art of low accuracy in detecting the operating status of users' distributed photovoltaics, single management means and low management level. A distributed photovoltaic anomaly identification method based on artificial intelligence technology is provided, which uses big data filtering and screening and manual assisted screening and investigation to accurately identify distributed photovoltaic real-time operating abnormal data and user distributed photovoltaic status. At the same time, classified measures are taken for the power generation status of distributed photovoltaics, effectively intercepting the default behavior of over-capacity power generation, and actively pushing information to users with low power generation efficiency such as equipment failure, thereby reducing unnecessary work orders and complaints, and improving satisfaction and customer stickiness.

[0005] In order to achieve the above object, the present invention adopts the following technical solution: a distributed photovoltaic anomaly identification method based on artificial intelligence technology, comprising the following steps:

[0006] S1: Obtain the full amount of data of residential distributed photovoltaic users, and process the acquired data to obtain the original data;

[0007] S2: Select the power generation data of users on similar historical dates and the power generation data of neighboring users from the original data to build a residential distributed photovoltaic status evaluation model;

[0008] S3: Combined with offline investigation, adjust the key parameters of the positioning status category;

[0009] S4: Collecting real-time operation data of distributed photovoltaics of users, using the model to determine whether the operation data is abnormal and the type of data abnormality, and evaluating the state of distributed photovoltaics of residents according to the type of data abnormality;

[0010] S5: Different service processing strategies are adopted for users with different abnormal states, and the abnormal operation data are repaired by using historical similar date data and power generation data of neighboring users.

[0011] A status evaluation model for residential distributed photovoltaics is proposed. Through the characteristic expression of residential distributed photovoltaic power generation characteristics, the method of using big data filtering and screening and manual assisted screening and investigation is studied, which can accurately identify the real-time abnormal operation data of distributed photovoltaics and the status of user distributed photovoltaics, and effectively block the default behavior of over-capacity power generation such as private capacity increase.

[0012] Preferably, in step S4, repairing the abnormal operation data includes:

[0013] If the abnormal operation data is data loss or higher than the normal range: obtain the user's historical power generation data on similar dates and the power generation data of neighboring users as training samples, input them into the BP neural network for training, use the BP neural network to predict the abnormal data, and obtain the abnormal data repair value;

[0014] If the abnormal operating data is lower than the normal range: analyze the possible causes of abnormal power generation of the user's rooftop photovoltaic panels and send maintenance suggestions to the user.

[0015] Effectively block overcapacity power generation violations, proactively push information to users with low power generation efficiency due to equipment failures, reduce unnecessary work orders and complaints, and improve satisfaction and customer stickiness.

[0016] Preferably, the neighboring user power generation data includes screening neighboring users in the area and selecting users with high similarity to the users who need to perform data repair:

[0017] Calculate the similarity between the adjacent users and the current user in the area;

[0018] Sort the neighbors from high to low according to similarity, and select the user with the highest similarity as the sample.

[0019] Residential users within a certain area can be considered to have the same lighting conditions. Therefore, when adjacent users have similar photovoltaic installation capacity, their photovoltaic power generation is also similar. In view of this, it is first necessary to screen adjacent users in the area, select users with high similarity to the users who need data repair, and use these user data as the basis to improve the accuracy of data repair.

[0020] Preferably, the historical similar date power generation data includes selecting historical similar day data with weather similar to the current weather from historical data:

[0021] Calculate the power output of the kth user on the hth day And normalize the power output to get

[0022] Calculate the average power generation of all users in the tth period of the hth day

[0023] Calculate the average power generation of all users in all periods of day h

[0024] Calculate the average photovoltaic power generation of users on the current date

[0025] calculate and Pearson product-moment correlation coefficient, and sort all historical data days from high to low according to the correlation coefficient, and select the historical similar days with the highest similarity as samples

[0026] Historical data plays an important role in repairing abnormal data, but historical data involves photovoltaic power generation data under different weather conditions, and different weather conditions have a great impact on photovoltaic power generation. Therefore, in order to accurately repair the abnormal data of current data, it is necessary to select historical similar day data with weather similar to the current weather from the historical data.

[0027] Preferably, the step S4 further comprises:

[0028] S4.1: Collect the real-time operation data of the user's distributed photovoltaic power generation, and determine whether the operation data is lost;

[0029] S4.2: If the operation data is not lost, determine whether the operation data is higher than a normal range, and if so, evaluate the state of the residential distributed photovoltaic power generation as an over-capacity power generation state;

[0030] S4.3: If the operating data is not higher than the normal range, determine whether the operating data is lower than the normal range, and if so, evaluate the state of the residential distributed photovoltaic power generation as a low-efficiency power generation state;

[0031] S4.4: If the operating data is not lower than the normal range, the operating data is normal, and the state of the residents' distributed photovoltaic power generation is evaluated as a normal power generation state.

[0032] Data loss is manifested as data gaps, which are mainly caused by photovoltaic data collection devices or communication interruptions. Data higher than the normal range is manifested as significantly exceeding the power generation of adjacent users with similar installed capacity, mainly due to factors such as users using other power generation methods (such as diesel engine power generation) to disguise photovoltaic power generation to defraud subsidies or installing photovoltaic power generation arrays beyond capacity. Data lower than the normal range is manifested as significantly lower than the power generation of adjacent users with similar installed capacity, mainly due to low power generation efficiency caused by aging photovoltaic equipment or excessive dust accumulation on the surface of photovoltaic panels.

[0033] Preferably, in step 4.1, determining whether the operation data is lost is as follows:

[0034] Calculate the power output of user n in T time periods:

[0035] If the user's output power in time period t The output power during period t is Lost.

[0036] Data loss is manifested as data gaps, which are mainly caused by photovoltaic data acquisition devices or communication interruptions.

[0037] Preferably, in step S4.2, judging whether the operating data is higher than a normal range is:

[0038] Calculate the output power of user n in time period t and will Normalized, we get

[0039] Calculate the output power of user n's neighboring user m with similar photovoltaic installation capacity during period t Calculating output power The photovoltaic installed capacity S of the adjacent user m m Business

[0040] like: This indicates that the operating data is higher than the normal state, where k up is a constant coefficient; M is the number of adjacent users.

[0041] The data is higher than the normal range, which is manifested as significantly exceeding the power generation of neighboring users with similar installed capacity. This is mainly caused by factors such as users using other power generation methods (such as diesel engine power generation) to disguise photovoltaic power generation to obtain subsidies or installing photovoltaic power arrays beyond capacity.

[0042] Preferably, in step S4.3, judging whether the operating data is lower than a normal range is:

[0043] Calculate the output power of user n in time period t and will Normalized, we get

[0044] Calculate the output power of user n's neighboring user m with similar photovoltaic installation capacity during period t Calculating output power The photovoltaic installed capacity S of the adjacent user m m Business

[0045] like This indicates that the operating data is lower than the normal state, where k low is a constant coefficient.

[0046] The data is lower than the normal range, which is significantly lower than the power generation of neighboring users with similar installed capacity. The main reason is the low power generation efficiency caused by the aging of photovoltaic equipment or excessive dust accumulation on the surface of photovoltaic panels.

[0047] Preferably, in step S4, repairing the abnormal operation data includes:

[0048] If the abnormal data is data loss: perform reasonable repair processing based on relevant data at the time point;

[0049] If the abnormal data is higher than the normal range: analyze by time point and warn the user if abnormal data is found;

[0050] If the abnormal data is lower than the normal range: the user will be reminded and warned, and the user will be guided to remove dust and replace the photovoltaic panels.

[0051] In the distributed photovoltaic operation data, whether it is data loss caused by technical reasons during data collection and measurement, or due to some management problems, such as users pretending to generate electricity to defraud subsidies, low power generation efficiency of distributed photovoltaic equipment, etc., it will be reflected in the abnormal distributed photovoltaic power generation load data. These data anomalies need to be identified and processed, the lost data needs to be supplemented, and the default power generation such as pretending to generate electricity to defraud subsidies and over-capacity power generation needs to be handled in a timely manner. The data anomalies caused by the low power generation efficiency of photovoltaic equipment need to provide users with maintenance suggestions.

[0052] Preferably, in step S5, adopting different service processing strategies for users in different abnormal states includes:

[0053] For users whose power generation data is lost, complete the power generation data;

[0054] For users with over-capacity power generation, users are reminded that they have increased capacity privately to defraud subsidies;

[0055] For users in the state of inefficient power generation, we proactively send service text messages to remind them to promptly operate and manage their own power generation equipment, and provide proactive door-to-door services to guide users in cleaning and dust removal.

[0056] An active service model for residential distributed photovoltaic users is proposed, and policies are implemented based on the power generation status of distributed photovoltaics. The over-capacity power generation breach is effectively blocked. Information is actively pushed to users with low power generation efficiency such as equipment failure, reducing unnecessary work orders and complaints, and improving customer satisfaction and stickiness.

[0057] Therefore, the present invention has the following beneficial effects: 1. A status evaluation model for residential distributed photovoltaics is proposed. Through the characteristic expression of the characteristics of residential distributed photovoltaic power generation, a method of using big data filtering and screening and manual assisted screening and investigation is studied, which can accurately identify the real-time abnormal operation data of distributed photovoltaics and the status of user distributed photovoltaics, and effectively block the over-capacity power generation default behaviors such as private capacity increase; 2. An active service model for residential distributed photovoltaic users is proposed, and policies are implemented according to the classification of the power generation status of distributed photovoltaics, effectively blocking the over-capacity power generation default behaviors, and actively pushing information to users with low power generation efficiency such as equipment failures, reducing unnecessary work orders and complaints, and improving customer satisfaction and customer stickiness. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 The figure is a specific operation flow chart of the method of the present invention.

[0059] Figure 2 This is a schematic diagram of the first type of abnormal data in Example 1.

[0060] Figure 3 This is a schematic diagram of the second type of abnormal data in Example 1.

[0061] Figure 4 This is a schematic diagram of the third type of abnormal data in Example 1.

[0062] Figure 5 This is a flow chart of the abnormal data repair steps in Example 2. DETAILED DESCRIPTION

[0063] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments:

[0064] like Figure 1 In the illustrated embodiment, a distributed photovoltaic anomaly identification method based on artificial intelligence technology can be seen, and its operation process is as follows: Step 1, obtain the full amount of data of residential distributed photovoltaic users, and process the obtained data to obtain the original data; Step 2, select the user's historical similar date power generation data and the neighboring user power generation data from the original data to build a residential distributed photovoltaic status evaluation model; Step 3, combined with offline investigation, adjust the key parameters of the positioning status category in the residential distributed photovoltaic status evaluation model to verify the accuracy of the residential distributed photovoltaic status evaluation model; Step 4, collect the user's distributed photovoltaic real-time operation data, use the resident distributed photovoltaic status evaluation model to determine whether the operation data is abnormal and the data anomaly type, and evaluate the resident distributed photovoltaic status according to the data anomaly type; Step 5, adopt different service processing strategies for users with different abnormal states, and use historical similar date data and neighboring user power generation data to repair the abnormal operation data.

[0065] A status evaluation model for residential distributed photovoltaics is proposed. Through the characteristic expression of residential distributed photovoltaic power generation characteristics, the method of using big data filtering and screening and manual assisted screening and investigation is studied, which can accurately identify the real-time abnormal operation data of distributed photovoltaics and the status of user distributed photovoltaics, and effectively block the default behavior of over-capacity power generation such as private capacity increase.

[0066] The technical solution of the present application is further illustrated below through specific examples.

[0067] Step 1: Obtain the full data of residential distributed photovoltaic users, and process the acquired data to obtain the original data.

[0068] This embodiment relies on the marketing system and the collection system to obtain the full amount of data of residential distributed photovoltaic users including the reported capacity, power generation and consumption mode, power supply station and other related fields. And adopts data processing methods such as data cleaning, data filling, and data filtering to form the original data.

[0069] Step 2: Select the user's historical power generation data on similar dates and the power generation data of neighboring users from the original data to build a residential distributed photovoltaic status evaluation model.

[0070] Based on the historical power generation data of the selected users and the power generation data of nearby users, the power generation characteristic attributes are studied, and a residential distributed photovoltaic status evaluation model is constructed. The status of residential distributed photovoltaic power generation users is defined as normal power generation, overcapacity power generation, and inefficient power generation.

[0071] Modeling process:

[0072] The power generation data of users with similar historical dates and the power generation data of neighboring users are selected from the original data as the sample set, and the sample set is randomly divided into training sample set and test sample set in a ratio of 7:3. The neural network is used to train the residential distributed photovoltaic status evaluation model, and the trained residential distributed photovoltaic status evaluation model is used to test the samples in the test sample set.

[0073] Step 3: Combined with offline investigation, adjust the key parameters of the positioning status category to verify the accuracy of the model.

[0074] By combining offline troubleshooting, key parameters of the positioning status category are adjusted to verify the accuracy of the model and improve the accuracy of the model.

[0075] Step 4: Collect real-time operation data of user distributed photovoltaics, use the resident distributed photovoltaic status evaluation model to determine whether the operation data is abnormal and the type of data abnormality, and evaluate the resident distributed photovoltaic status according to the data abnormality type.

[0076] In the distributed photovoltaic operation data, whether it is the data missing caused by technical reasons during the data collection and measurement process, or due to some management problems, such as users disguising power generation to obtain subsidies, and the low power generation efficiency of distributed photovoltaic equipment, it will be reflected in the abnormal distributed photovoltaic power generation load data.

[0077] All of these data anomalies need to be identified and processed, lost data need to be supplemented, and false power generation to obtain subsidies, power generation beyond capacity and other illegal power generation need to be dealt with in a timely manner. If the low power generation efficiency of photovoltaic equipment causes data anomalies, users need to be given maintenance suggestions.

[0078] For the analysis of user distributed photovoltaic operation data, this embodiment divides the distributed photovoltaic abnormal data into three categories:

[0079] Obtain the user's real-time distributed photovoltaic operation data, determine whether the operation data is the first type of abnormal data, and if so, repair the abnormal data; if it is not the first type of abnormal data, determine whether the operation data is the second type of abnormal data, and if so, repair the abnormal data; if it is not the second type of abnormal data, determine whether the operation data is the third type of abnormal data, and if so, repair the abnormal data; if not, it means that the operation data is normal.

[0080] Specifically, the three types of distributed photovoltaic abnormal data include:

[0081] (1) The first type of abnormal data: data loss, such as Figure 2 As shown, it manifests as data gaps, which are mainly caused by photovoltaic data acquisition devices or communication interruptions.

[0082] (2) The second type of abnormal data: data higher than the normal range, such as Figure 3 As shown, the power generation is significantly higher than that of adjacent users with similar installed capacity. This is mainly caused by factors such as users using other power generation methods (such as diesel engine power generation) to disguise photovoltaic power generation to obtain subsidies or installing photovoltaic power generation arrays in excess of capacity. The corresponding residential distributed photovoltaic status is an over-capacity power generation state.

[0083] (3) The third type of abnormal data: data is lower than the normal range, such as Figure 4 As shown, the power generation is significantly lower than that of adjacent users with similar installed capacity. The main reason is the low power generation efficiency caused by the aging of photovoltaic equipment or excessive dust accumulation on the surface of photovoltaic panels. The corresponding residential distributed photovoltaic status is a low-efficiency power generation state.

[0084] Distributed photovoltaic abnormal data identification:

[0085] In view of the above three types of abnormal data, this embodiment proposes the following targeted data identification methods, and the specific mathematical models are as follows:

[0086] (1) Identification of the first type of abnormal data (data loss)

[0087] Assume that the power output of user n in T time periods is When the user's output power in period t satisfy: , it indicates that there is a loss of output power during period t.

[0088] (2) Identification of the second type of abnormal data (data higher than the normal range)

[0089] Assume that the power output of user n in T time periods is The power output of neighboring user m with similar photovoltaic installation capacity of user n in T time periods is Output power of user n in time period t After normalization, we get:

[0090]

[0091]

[0092]

[0093] in, and The normalized running data of users n and m are respectively, n and S m are the photovoltaic installed capacities of users n and m respectively.

[0094] When user n runs the data to meet:

[0095]

[0096] , it indicates that the operating data is higher than the normal value.

[0097] Among them, k up is a constant coefficient; M is the number of adjacent users.

[0098] (3) Identification of the third type of abnormal data (data below the normal range)

[0099] Calculate the power output X of neighboring user m with similar photovoltaic installation capacity of user n in T time periods m And the output power of adjacent user m in time period t

[0100] Will Normalized, we get Calculating output power The photovoltaic installed capacity S of the adjacent user m m Business

[0101] When user n runs the data to meet:

[0102] , it indicates that the operating data is lower than the normal value.

[0103] Among them, k low is a constant coefficient. It should be noted here that k up and k low Generally it is set according to the actual situation.

[0104] Distributed photovoltaic abnormal data repair:

[0105] Among the above three types of abnormal data, for the first type of abnormal data, this embodiment performs reasonable repair processing according to relevant data at time points. For the second type of abnormal data, this embodiment performs analysis at time points and promptly issues warnings when abnormal data is found. The third type of data is caused by data abnormalities due to low power generation efficiency caused by aging or improper management of the user's photovoltaic equipment. This type of data abnormality also requires warnings, and the user is guided to take measures such as dust removal and replacement of the photovoltaic panels to solve the problem.

[0106] Step 5: Adopt different service processing strategies for users in different abnormal states and repair abnormal operation data.

[0107] Different service processing strategies are adopted for users in different abnormal states. For users whose power generation data is lost, the power generation data is supplemented to promote the improvement of user data quality. For users of overcapacity power generation, users are reminded that they may have privately increased capacity to defraud subsidies. For users of inefficient power generation, service text messages are proactively sent to remind users to timely operate and manage their own power generation equipment to avoid reductions in user subsidies and electricity purchase fees due to aging or improper management of photovoltaic equipment. The power generation efficiency of photovoltaic users is also improved through proactive door-to-door services to guide users to clean and remove dust and other measures.

[0108] This embodiment can also build a distributed photovoltaic operation data monitoring system based on the proposed residential distributed photovoltaic status evaluation model (photovoltaic abnormal data identification and repair model), so as to effectively block suspicious power generation loopholes and actively push classified reminders of power generation customers' equipment failure information and subsidy issuance status.

[0109] The solution of the present invention has the following advantages:

[0110] 1. Propose a status evaluation model for residential distributed photovoltaics. Through the characteristic expression of residential distributed photovoltaic power generation characteristics, study the use of big data filtering and screening and manual assisted screening methods to effectively block private capacity increase and other over-capacity power generation default behaviors.

[0111] 2. Propose an active service model for residential distributed photovoltaic users, classify and implement policies based on the power generation status of distributed photovoltaics, effectively block the breach of contract for over-capacity power generation, and actively push information to users with low power generation efficiency such as equipment failures, reduce unnecessary work orders and complaints, and improve satisfaction and customer stickiness.

[0112] Embodiment 2:

[0113] This embodiment provides another method for repairing abnormal data based on the first embodiment:

[0114] For the first and second types of abnormal data, this embodiment judges the data according to the time point, and needs to use the power generation data of adjacent users and the power generation data of similar historical days for comprehensive judgment, such as Figure 5 As shown, the main steps are as follows:

[0115] 1) Input the real-time operation data of the current user's distributed photovoltaic;

[0116] 2) Filter out missing data in the data based on the first type of abnormal data identification method;

[0117] 3) Screen out the other two types of data based on the second and third types of abnormal data identification methods;

[0118] 4) Based on the historical user photovoltaic data and the current adjacent user data, the first and second types of abnormal data are predicted, and the first type of abnormal data is repaired; at the same time, corresponding suggestions are given to users for the third type of data abnormalities, including the possible causes of abnormal power generation of rooftop photovoltaic panels and other maintenance suggestions.

[0119] Among them, the method involved in abnormal data repair in step 4) mainly involves inputting historical data and adjacent user data as training samples into the BP neural network model for model training, so as to obtain the repair value of the abnormal data.

[0120] Model training process:

[0121] Historical data and adjacent user data are used as sample sets, and the sample sets are randomly divided into training sample sets and test sample sets in a ratio of 8:2. The training sample sets are used to train the abnormal data repair model using a neural network, and the trained abnormal data is used to repair the model until the termination condition is met.

[0122] The samples in the test sample set are tested, and the abnormal data repair rate and the abnormal data repair success rate are calculated. Then, the abnormal data repair model with a high abnormal data repair rate and a high abnormal data repair success rate is selected as the final abnormal data repair model.

[0123] In view of the data anomalies of existing distributed photovoltaic power generation devices, this embodiment divides the distributed photovoltaic anomaly data into three categories and proposes methods for identifying each type of data; then, a method of combining historical date power generation data with neighboring user power generation data is proposed to repair the anomaly data.

[0124] Embodiment three:

[0125] This embodiment provides a method for selecting similar users and selecting historical similar days based on the first and second embodiments.

[0126] 1. Similarity user selection

[0127] Residential users within a certain area can be considered to have the same lighting conditions. Therefore, when adjacent users have similar photovoltaic installation capacities, their photovoltaic power generation powers are also similar.

[0128] In view of this, it is first necessary to screen the neighboring users in the area, select the neighboring users with high similarity to the current user who needs data repair, and use the data of these neighboring users as a basis to improve the accuracy of data repair.

[0129] This embodiment uses the Pearson product-moment correlation coefficient to measure the similarity of user data.

[0130] Specific:

[0131] Taking the normalized photovoltaic data of users n and m as an example, the similarity between the two users is:

[0132]

[0133] in, and They represent the average photovoltaic power generation power of users n and m respectively.

[0134] The similarity between neighboring users and the current user in the area is calculated by the above formula, and they are sorted from high to low according to the similarity. Then, the neighboring users with the highest similarity to the current user are selected as samples. The number of selected users can be determined according to the actual situation.

[0135] 2. Selection of historically similar days

[0136] Historical data plays an important role in repairing abnormal data, but historical data involves photovoltaic power generation output data under different weather conditions, and different weather conditions have a great impact on photovoltaic power generation.

[0137] Therefore, in order to more accurately repair the abnormal data of the current data, it is necessary to select historical data of similar days with weather similar to the current weather from the historical data.

[0138] However, since there is less relevant data related to weather information in the current power grid system platform, this embodiment proposes the following method to select historical similar days:

[0139] Assuming that the number of historical data days available for data repair is H days, the power output of the kth user on the hth day is for:

[0140]

[0141] After normalization:

[0142]

[0143] The average power generation of all users in the tth period of the hth day is:

[0144]

[0145] Where K represents the number of users covered in the historical data.

[0146] Further, we can get the average value of all periods on the hth day

[0147]

[0148] Similarly, the average photovoltaic power generation of users on the current date is

[0149]

[0150] in

[0151] Calculate in sequence and Pearson product-moment correlation coefficient, and sort all historical data days from high to low according to the correlation coefficient, and select the historical similar days with the highest similarity as samples.

[0152] In addition, if the number of samples of similar days is too small, it will be difficult to train the residential distributed photovoltaic status evaluation model and the subsequent abnormal data repair model (neural network model). If the number of samples of similar days is too large, the repair accuracy will also decrease. Therefore, the number of similar days should be selected according to the actual situation.

[0153] The above-described embodiment is only a preferred solution of the present invention and does not limit the present invention in any form. There are other variations and modifications without exceeding the technical solution described in the claims.

Claims

1. A distributed photovoltaic anomaly identification method based on artificial intelligence technology, characterized in that: The following steps are involved: S1: Obtain the full amount of data of residential distributed photovoltaic users, and process the acquired data to obtain the original data; S2: Select the historical power generation data of users on similar dates and the power generation data of neighboring users from the original data to build a residential distributed photovoltaic status evaluation model; S3: Combined with offline investigation, key parameters of the positioning status category in the model are adjusted; S4: Collecting real-time operation data of distributed photovoltaics of users, using the model to determine whether the operation data is abnormal and the type of data abnormality, and evaluating the state of distributed photovoltaics of residents according to the type of data abnormality; S5: Different service processing strategies are adopted for users with different abnormal states, and the abnormal operation data are repaired by using historical similar date data and power generation data of neighboring users.

2. According to claim 1, a distributed photovoltaic anomaly identification method based on artificial intelligence technology is characterized in that: In step S4, repairing the abnormal operation data includes: If the abnormal operation data is data loss or higher than the normal range: obtain the user's historical power generation data on similar dates and the power generation data of neighboring users as training samples, input them into the BP neural network for training, use the BP neural network to predict the abnormal data, and obtain the abnormal data repair value; If the abnormal operating data is lower than the normal range: analyze the possible causes of abnormal power generation of the user's rooftop photovoltaic panels and send maintenance suggestions to the user.

3. A distributed photovoltaic anomaly identification method based on artificial intelligence technology according to claim 1 or 2, characterized in that: The neighboring user power generation data includes screening neighboring users in the area and selecting users with high similarity to the users who need data repair: Calculate the similarity between the adjacent users and the current user in the area; Sort the neighbors from high to low according to similarity, and select the user with the highest similarity as the sample.

4. A distributed photovoltaic anomaly identification method based on artificial intelligence technology according to claim 1 or 2, characterized in that: The historical similar date power generation data includes selecting historical similar day data with weather similar to the current weather from historical data: Calculate the power output of the kth user on the hth day And normalize the power output to get Calculate the average power generation of all users in the tth period of the hth day Calculate the average power generation of all users in all periods of day h Calculate the average photovoltaic power generation of users on the current date calculate and Pearson product-moment correlation coefficient, and sort all historical data days from high to low according to the correlation coefficient, and select the historical similar days with the highest similarity as samples.

5. According to claim 1, a distributed photovoltaic anomaly identification method based on artificial intelligence technology is characterized in that: The step S4 further comprises: S4.1: Collect the real-time operation data of the user's distributed photovoltaic power generation, and determine whether the operation data is lost; S4.2: If the operation data is not lost, determine whether the operation data is higher than a normal range, and if so, evaluate the state of the residential distributed photovoltaic power generation as an over-capacity power generation state; S4.3: If the operating data is not higher than the normal range, determine whether the operating data is lower than the normal range, and if so, evaluate the state of the residential distributed photovoltaic power generation as a low-efficiency power generation state; S4.4: If the operating data is not lower than the normal range, the operating data is normal, and the state of the residents' distributed photovoltaic power generation is evaluated as a normal power generation state.

6. The distributed photovoltaic anomaly identification method based on artificial intelligence technology according to claim 5 is characterized in that: In step 4.1, determining whether the operation data is lost is as follows: Calculate the power output of user n in T time periods: If the user's output power in time period t The output power during period t is Lost.

7. The distributed photovoltaic anomaly identification method based on artificial intelligence technology according to claim 5 is characterized in that: In step S4.2, whether the operating data is higher than the normal range is determined as follows: Calculate the output power of user n in time period t and will Normalized, we get Calculate the output power of neighboring user m with similar photovoltaic installation capacity of user n during period t Calculating output power The photovoltaic installed capacity S of the adjacent user m m Business like: This indicates that the operating data is higher than the normal state, where k up is a constant coefficient; M is the number of adjacent users.

8. A distributed photovoltaic anomaly identification method based on artificial intelligence technology according to claim 5, 6 or 7, characterized in that: In step S4.3, whether the operating data is lower than the normal range is determined as follows: Calculate the output power of user n in time period t and will Normalized, we get Calculate the output power of neighboring user m with similar photovoltaic installation capacity of user n during period t Calculating output power The photovoltaic installed capacity S of the adjacent user m m Business like This indicates that the operating data is lower than the normal state, where k low is a constant coefficient.

9. The distributed photovoltaic anomaly identification method based on artificial intelligence technology according to claim 1 is characterized in that: In step S4, repairing the abnormal operation data includes: If the abnormal data is data loss: perform reasonable repair processing based on relevant data at the time point; If the abnormal data is higher than the normal range: analyze by time point and warn the user if abnormal data is found; If the abnormal data is lower than the normal range: the user will be reminded and warned, and the user will be guided to remove dust and replace the photovoltaic panels.

10. A distributed photovoltaic anomaly identification method based on artificial intelligence technology according to claim 1 or 2 or 5 or 9, characterized in that: In step S5, different service processing strategies are adopted for users in different abnormal states, including: For users whose power generation data is lost, complete the power generation data; For users with over-capacity power generation, users are reminded that they have increased capacity privately to defraud subsidies; For users in the state of inefficient power generation, we proactively send service text messages to remind them to promptly operate and manage their own power generation equipment, and provide proactive door-to-door services to guide users in cleaning and dust removal.

Images