CN-122026805-A - Building photovoltaic remote diagnosis and collaborative operation and maintenance method and system based on artificial intelligence

CN122026805ACN 122026805 ACN122026805 ACN 122026805ACN-122026805-A

Abstract

The invention discloses a building photovoltaic remote diagnosis and collaborative operation and maintenance method and system based on artificial intelligence, which belong to the technical field of building photovoltaic diagnosis and operation and maintenance, wherein a building photovoltaic three-dimensional digital twin model is established by acquiring accurate building information and photovoltaic distribution information, data of each photovoltaic module are associated with the photovoltaic module to finish deep fusion of the building information and the photovoltaic data, abnormal state probability distribution is analyzed by using a state transition probability matrix, fault type probability distribution is calculated by using a photovoltaic fault diagnosis model, remote diagnosis of building photovoltaic is performed, analysis of abnormal states of the photovoltaic module is guaranteed, fault types are considered, the finally obtained diagnosis result accurately distinguishes the abnormal states and the fault states, accurate remote fault diagnosis is realized, the photovoltaic module is positioned according to the diagnosis result, corresponding operation and maintenance instructions are generated, the difficulty of operation and maintenance work is reduced, and the operation and maintenance efficiency is effectively improved.

Inventors

HONG WEI
HAN LIN
Deng Quanbin
WANG DONG
LI YANGHUA
PAN JIASHENG
ZHANG NAIYU

Assignees

广州发展新能源集团股份有限公司

Dates

Publication Date: 20260512
Application Date: 20251205

Claims (10)

1. The building photovoltaic remote diagnosis and collaborative operation and maintenance method based on artificial intelligence is characterized by comprising the following steps of: Building information and photovoltaic distribution information are obtained, and a building photovoltaic three-dimensional digital twin model is built according to the building information and the photovoltaic distribution information; Acquiring real-time electric state data and real-time physical state data of each photovoltaic module at each moment in a preset sampling time window, and respectively associating the real-time electric state data and the real-time physical state data of each photovoltaic module on the corresponding photovoltaic module in the building photovoltaic three-dimensional digital twin model to form a standard photovoltaic data stream of each photovoltaic module; Integrating the standard photovoltaic data flow of each photovoltaic module into photovoltaic diagnosis data, carrying out abnormal data analysis and extraction on the photovoltaic diagnosis data to construct a conventional state data chain according to the photovoltaic diagnosis data after the abnormal data extraction is completed, constructing an abnormal state data chain according to the extracted abnormal data, and establishing a state transition probability matrix based on the conventional state data chain and the abnormal state data chain so as to calculate abnormal state probability distribution of building photovoltaic by using the state transition probability matrix; Inputting the photovoltaic diagnosis data into a pre-trained photovoltaic fault diagnosis model to output fault type probability distribution of building photovoltaic according to the photovoltaic fault diagnosis model; Building photovoltaic diagnosis is carried out according to the abnormal state probability distribution and the fault type probability distribution, building photovoltaic diagnosis results are obtained, the building photovoltaic diagnosis results are marked into the building photovoltaic three-dimensional digital twin model, diagnosis result positioning is completed, corresponding operation and maintenance instructions are generated according to the diagnosis result positioning and the building photovoltaic diagnosis results, and the operation and maintenance instructions are uploaded to a cloud platform to complete operation and maintenance task dispatching.
2. The artificial intelligence based building photovoltaic remote diagnosis and collaborative operation and maintenance method according to claim 1, wherein obtaining building information and photovoltaic distribution information to build a building photovoltaic three-dimensional digital twin model from the building information and the photovoltaic distribution information includes: Acquiring a building BIM design model of a current building through a cloud platform, acquiring a building scene model in a preset influence range of the current building through three-dimensional point cloud scanning, and fusing and splicing the building BIM design model and the building scene model to form building information of the current building; Acquiring a photovoltaic module installation record and a photovoltaic module installation position of a current building through a cloud platform as photovoltaic distribution information of the current building, wherein the photovoltaic module installation record comprises a photovoltaic module model and a photovoltaic module electrical connection relation, and the photovoltaic module installation position comprises a photovoltaic module physical layout, a photovoltaic module installation dip angle and a photovoltaic module orientation azimuth angle; according to the building information of the current building, a building main body three-dimensional digital twin model of the current building is built, and according to the photovoltaic distribution information of the current building, each photovoltaic module is positioned and placed on the building main body three-dimensional digital twin model so as to build the building photovoltaic three-dimensional digital twin model.
3. The artificial intelligence-based building photovoltaic remote diagnosis and collaborative operation and maintenance method according to claim 1, wherein acquiring real-time electrical state data and real-time physical state data of each photovoltaic module at each moment in a preset sampling time window, and calculating standard photovoltaic data of each photovoltaic module according to the real-time electrical state data and the real-time physical state data to form standard photovoltaic data flow of each photovoltaic module comprises: Collecting real-time electric state data of each photovoltaic module at each moment in real time on the direct current side of each photovoltaic module in a preset sampling time window, wherein the real-time electric state data comprise real-time voltage, real-time current and real-time electric power; Real-time working temperature data and real-time irradiance receiving data of each photovoltaic module at each moment are detected in real time through a temperature sensing unit and an irradiance sensing unit which are arranged on each photovoltaic module, and the real-time working temperature data and the real-time irradiance receiving data are integrated to form real-time physical state data of each photovoltaic module at each moment; for each photovoltaic module, carrying out standardized conversion on the real-time electric state data by utilizing the real-time physical state data to obtain standard real-time electric state data; adding positioning information to the standard real-time electrical state data of each photovoltaic module by using the building photovoltaic three-dimensional digital twin model to form standard photovoltaic data of each photovoltaic module at each moment; and integrating the standard photovoltaic data of each piece of photovoltaic module according to the time sequence to form a standard photovoltaic data stream corresponding to each piece of photovoltaic module.
4. The artificial intelligence-based building photovoltaic remote diagnosis and collaborative operation and maintenance method according to claim 1, wherein integrating the standard photovoltaic data stream of each photovoltaic module into photovoltaic diagnosis data, performing abnormal data analysis and extraction on the photovoltaic diagnosis data to construct a conventional state data chain according to the photovoltaic diagnosis data after the abnormal data extraction is completed, and constructing an abnormal state data chain according to the extracted abnormal data, comprising: Carrying out standardization processing on the standard photovoltaic data flow of each photovoltaic module to obtain photovoltaic diagnosis data, wherein the standardization processing is completed by adopting zero-mean calculation; Arranging the photovoltaic diagnosis data into a photovoltaic diagnosis data point sequence according to the numerical value order, and determining an upper quartile and a lower quartile in the photovoltaic diagnosis data point sequence by using a linear interpolation method; Acquiring a preset abnormality analysis coefficient, calculating a quartile distance according to an upper quartile and a lower quartile in the photovoltaic diagnosis data point sequence, and calculating an abnormality judgment boundary value according to the abnormality analysis coefficient, the lower quartile and the quartile distance; performing anomaly analysis on each photovoltaic diagnosis data point in the photovoltaic diagnosis data by using the anomaly determination boundary value, taking each photovoltaic diagnosis data point which is lower than the anomaly determination boundary value in the photovoltaic diagnosis data as an anomaly data point, adding an anomaly tag on the anomaly data point, taking each photovoltaic diagnosis data point which is not lower than the anomaly determination boundary value in the photovoltaic diagnosis data as a regular data point, and adding a regular tag on the regular data point, wherein the anomaly tag and the regular tag are both binary tags; Extracting each abnormal data point, sorting each abnormal data point according to a time sequence to form a pre-abnormal state data chain, and sorting each conventional data point according to a time sequence to form a pre-conventional state data chain; Performing discrete verification on the pre-routine state data chain and the pre-abnormal state data chain respectively, taking the pre-routine state data chain passing the discrete verification as a routine state data chain, and taking the pre-abnormal state data chain passing the discrete verification as an abnormal state data chain; Correspondingly, based on the regular state data chain and the abnormal state data chain, a state transition probability matrix is established, so that abnormal state probability distribution of the building photovoltaic is calculated by using the state transition probability matrix, and the method comprises the following steps: Calculating the probability of the conventional state data transferring to the abnormal state data, the probability of the abnormal state data transferring to the conventional state data and the probability of the abnormal state data transferring to the abnormal state data according to the conventional state data chain and the abnormal state data chain, and establishing a state transition probability matrix by taking the probability of the conventional state data transferring to the abnormal state data, the probability of the abnormal state data transferring to the conventional state data, the probability of the conventional state data transferring to the conventional state data and the probability of the abnormal state data transferring to the abnormal state data as elements in the matrix; and carrying out multi-iteration limit probability calculation on the state transition probability matrix to obtain limit probability distribution, wherein the limit probability distribution comprises conventional state probability distribution and abnormal state probability distribution.
5. The artificial intelligence based building photovoltaic remote diagnosis and co-operation maintenance method according to claim 1, wherein the pre-training method of the photovoltaic fault diagnosis model comprises the following steps: acquiring historical photovoltaic diagnosis data and historical photovoltaic diagnosis records of a current building through a cloud platform, generating a corresponding historical photovoltaic module current and voltage correlation curve according to the historical photovoltaic diagnosis data, and integrating the historical photovoltaic module current and voltage correlation curve and the historical photovoltaic diagnosis records into first photovoltaic fault diagnosis branch model training data; Selecting a one-dimensional convolutional neural network model as a first base model, and training the first base model for multiple times by utilizing training data of the first photovoltaic fault diagnosis branch model to obtain a first photovoltaic fault diagnosis branch model; Integrating the historical photovoltaic diagnosis data and the historical photovoltaic diagnosis record into second photovoltaic fault diagnosis branch model training data; Selecting a long-period memory network model as a second base model, and performing multiple times of training on the second base model by utilizing training data of the second photovoltaic fault diagnosis branch model to obtain a second photovoltaic fault diagnosis branch model; Arranging the first photovoltaic fault diagnosis branch model and the second photovoltaic fault diagnosis branch model in parallel, taking an input layer of the first photovoltaic fault diagnosis branch model and an input layer of the second photovoltaic fault diagnosis branch model as a first input channel and a second input channel respectively, connecting an output layer of the first photovoltaic fault diagnosis branch model and an output layer of the second photovoltaic fault diagnosis branch model through a full connection layer to form a full connection fusion layer so as to obtain a trained photovoltaic fault diagnosis model, wherein the photovoltaic fault diagnosis model further comprises a fault type diagnosis output layer for outputting a photovoltaic fault type score, and the photovoltaic fault type score is a quantification score of a photovoltaic module predicted by the photovoltaic fault diagnosis model to generate various types of faults; and inputting the photovoltaic fault type score output by the fully-connected fusion layer into the fault type diagnosis output layer so as to generate and output corresponding fault type confidence coefficient for the photovoltaic fault characteristics by utilizing the fault type diagnosis output layer.
6. The artificial intelligence based building photovoltaic remote diagnosis and co-operation method according to claim 5, wherein inputting the photovoltaic diagnosis data into a pre-trained photovoltaic fault diagnosis model to output a fault type probability distribution of building photovoltaics according to the photovoltaic fault diagnosis model comprises: Generating a corresponding photovoltaic module current and voltage correlation curve according to the photovoltaic diagnosis data, taking the photovoltaic module current and voltage correlation curve as a first input quantity, and taking the photovoltaic diagnosis data as a second input quantity; Obtaining a pre-trained photovoltaic fault diagnosis model, inputting the first input quantity into the first input channel of the photovoltaic fault diagnosis model to process the first input quantity through a first photovoltaic fault diagnosis branch model in the photovoltaic fault diagnosis model, calculating a corresponding first photovoltaic fault characteristic, and inputting the second input quantity into the second input channel of the photovoltaic fault diagnosis model to process the second input quantity through a second photovoltaic fault diagnosis branch model in the photovoltaic fault diagnosis model, and calculating a corresponding second photovoltaic fault characteristic; And performing feature splicing and quantification on the first photovoltaic fault feature and the second photovoltaic fault feature by using the fully-connected fusion layer in the photovoltaic fault diagnosis model to obtain a photovoltaic fault type score, and calculating a corresponding fault type confidence coefficient for the photovoltaic fault type score by using the fault type diagnosis output layer in the photovoltaic fault diagnosis model to generate fault type probability distribution according to the fault type confidence coefficient.
7. The artificial intelligence-based building photovoltaic remote diagnosis and collaborative operation and maintenance method according to claim 1, wherein performing building photovoltaic diagnosis according to the abnormal state probability distribution and the fault type probability distribution to obtain building photovoltaic diagnosis results includes: Acquiring a preset state judgment probability threshold, and judging the abnormal state type of the abnormal state probability distribution of each photovoltaic module by using the state judgment probability threshold, wherein when the abnormal state probability of a certain photovoltaic module in the abnormal state probability distribution is higher than the state judgment probability threshold, the abnormal state of the certain photovoltaic module is judged; Acquiring a preset fixed abnormal state duration threshold, and in the sampling time window, if the duration of the abnormal state of a certain photovoltaic module exceeds the fixed abnormal state duration threshold, considering the abnormal state as a fixed abnormal state, and if the duration of the abnormal state of the certain photovoltaic module does not exceed the fixed abnormal state duration threshold, considering the abnormal state as a random abnormal state; Acquiring a preset fault confidence coefficient threshold value, and performing fault judgment on the fault type probability distribution of each photovoltaic module by using the fault confidence coefficient threshold value, wherein when the fault type confidence coefficient in the fault type probability distribution of a certain photovoltaic module is higher than the fault confidence coefficient threshold value, the fault of the photovoltaic module of the corresponding type is judged to be in a fault state; If the probability of the abnormal state of the certain photovoltaic module in the abnormal state probability distribution is higher than the state judgment probability threshold, and the confidence of the fault type of the certain photovoltaic module in the fault type probability distribution is higher than the fault confidence threshold, judging that the certain photovoltaic module is in a normal state; and integrating the fixed abnormal state, the random abnormal state, the fault state or the normal state of each photovoltaic module to form a building photovoltaic diagnosis result.
8. The artificial intelligence based architectural photovoltaic remote diagnosis and collaborative operation and maintenance method according to claim 7, wherein the operation and maintenance instructions include a first operation and maintenance instruction, a second operation and maintenance instruction, and a third operation and maintenance instruction; Correspondingly, labeling the building photovoltaic diagnosis result into the building photovoltaic three-dimensional digital twin model, completing diagnosis result positioning, and generating a corresponding operation and maintenance instruction according to the diagnosis result positioning and the building photovoltaic diagnosis result, wherein the operation and maintenance instruction comprises: According to the building photovoltaic diagnosis result, in the building photovoltaic three-dimensional digital twin model, corresponding labeling is carried out on each photovoltaic module, and each photovoltaic module is respectively labeled as a fixed abnormal state, a random abnormal state, a fault state or a normal state; If the mark of the photovoltaic module is in a fixed abnormal state, generating a pre-first operation and maintenance instruction, and combining the diagnosis result to locate to form a first operation and maintenance instruction, wherein the first operation and maintenance instruction is used for controlling an operation and maintenance end to monitor the continuous state of the photovoltaic module at a specific position and generate a long-term operation and maintenance record; If the mark of the photovoltaic module is in a random abnormal state, generating a pre-second operation and maintenance instruction, and combining the diagnosis result to locate to form a second operation and maintenance instruction, wherein the second operation and maintenance instruction is used for controlling an operation and maintenance end to monitor the continuous state of the photovoltaic module at a specific position; if the mark of the photovoltaic module is in a fault state, generating a pre-third operation and maintenance instruction, and combining the diagnosis result to locate to form a third operation and maintenance instruction, wherein the third operation and maintenance instruction is used for controlling an operation and maintenance end to carry out inspection and fault removal on the photovoltaic module at a specific position; If the mark of the photovoltaic module is in a normal state, no operation and maintenance instruction is generated.
9. The artificial intelligence based building photovoltaic remote diagnosis and collaborative operation and maintenance method according to claim 8, wherein uploading the operation and maintenance instructions to a cloud platform to complete operation and maintenance task dispatch includes: uploading the operation and maintenance instruction generated for each photovoltaic module to a cloud platform, recording the received operation and maintenance instruction by the cloud platform, and generating a corresponding operation and maintenance task according to the received operation and maintenance instruction; If the received operation and maintenance instruction is a first operation and maintenance instruction, a continuous state monitoring task and a long-term operation and maintenance recording task are distributed to a monitoring and maintenance team of an operation and maintenance end through a mobile intelligent terminal; If the received operation and maintenance instruction is a second operation and maintenance instruction, a continuous state monitoring task is distributed to a monitoring and maintenance team at an operation and maintenance end through the mobile intelligent terminal; and if the received operation and maintenance instruction is a second operation and maintenance instruction, distributing a site inspection task and a fault removal task to a monitoring operation and maintenance team of the operation and maintenance end through the mobile intelligent terminal.
10. The building photovoltaic remote diagnosis and cooperative operation and maintenance system based on artificial intelligence is characterized by being applied to the building photovoltaic remote diagnosis and cooperative operation and maintenance method based on artificial intelligence as claimed in any one of claims 1 to 9, and comprising the following steps: the three-dimensional model building unit is used for obtaining building information and photovoltaic distribution information so as to build a building photovoltaic three-dimensional digital twin model according to the building information and the photovoltaic distribution information; The photovoltaic data acquisition unit is used for acquiring real-time electric state data and real-time physical state data of each photovoltaic module at each moment in a preset sampling time window, and respectively associating the real-time electric state data and the real-time physical state data of each photovoltaic module on the corresponding photovoltaic module in the building photovoltaic three-dimensional digital twin model to form a standard photovoltaic data stream of each photovoltaic module; The abnormal state analysis unit is used for integrating the standard photovoltaic data flow of each photovoltaic module into photovoltaic diagnosis data, carrying out abnormal data analysis and extraction on the photovoltaic diagnosis data, constructing a conventional state data chain according to the photovoltaic diagnosis data after the abnormal data extraction is completed, constructing an abnormal state data chain according to the extracted abnormal data, and establishing a state transition probability matrix based on the conventional state data chain and the abnormal state data chain so as to calculate abnormal state probability distribution of building photovoltaic by using the state transition probability matrix; The fault state identification unit is used for inputting the photovoltaic diagnosis data into a pre-trained photovoltaic fault diagnosis model so as to output fault type probability distribution of building photovoltaic according to the photovoltaic fault diagnosis model; The remote diagnosis and operation and maintenance unit is used for carrying out building photovoltaic diagnosis according to the abnormal state probability distribution and the fault type probability distribution to obtain building photovoltaic diagnosis results, marking the building photovoltaic diagnosis results into the building photovoltaic three-dimensional digital twin model to finish diagnosis result positioning, generating corresponding operation and maintenance instructions according to the diagnosis result positioning and the building photovoltaic diagnosis results, and uploading the operation and maintenance instructions to a cloud platform to finish operation and maintenance task dispatching.

Description

Building photovoltaic remote diagnosis and collaborative operation and maintenance method and system based on artificial intelligence Technical Field The invention belongs to the technical field of building photovoltaic diagnosis and operation and maintenance, and particularly relates to a building photovoltaic remote diagnosis and collaborative operation and maintenance method and system based on artificial intelligence. Background Along with the acceleration of global energy transformation, photovoltaic power generation, in particular to a building photovoltaic system which is deeply integrated with a building, is rapidly developed. However, due to the unique integration and environmental dependence of the building photovoltaic system, the operation and maintenance aspect faces more serious challenges than the traditional ground power station, the existing operation and maintenance mode is often highly dependent on manual periodic inspection and experience judgment, the response is slow, the cost is high, the manual inspection efficiency is extremely low for the building photovoltaic system with wide distribution and complex installation position, and the early warning and accurate positioning of faults cannot be realized. In recent years, some modern remote diagnosis systems are presented, wherein the more common is an infrared monitoring photovoltaic detection and diagnosis system which usually needs expensive professional equipment and is easy to be disturbed by environment, and is difficult to be applied in a complex building environment in a large scale, and the monitoring system and the operation and maintenance system are two systems which are separated, so that the conditions of information disconnection, transmission errors and the like can occur between the photovoltaic monitoring and operation and maintenance decision, the operation and maintenance accuracy and efficiency of building photovoltaic are seriously influenced, and an effective integral closed loop for monitoring, diagnosis and cooperative operation and maintenance cannot be formed. In addition, because the photovoltaic module fault and performance attenuation mode of the building photovoltaic system are more complex and hidden than those of a general photovoltaic system, the normal working state of some photovoltaic modules in the building photovoltaic system is influenced randomly by cloud layers, flying birds and the like and is also influenced by the fixed effect of the inherent structures of the building structure, the ventilating duct, the surrounding trees and other specific positions on the photovoltaic modules at some positions, therefore, in the building photovoltaic system, the situation that the working state of some photovoltaic modules is normal, but the current and voltage characteristics change due to the influence of external conditions, the abnormal state occurs to other photovoltaic modules, the specific actual state of the photovoltaic modules is difficult to accurately distinguish by the traditional current and voltage characteristic analysis, the abnormal photovoltaic modules can be subjected to large-area inspection, investigation and maintenance, the working pressure of the operation and maintenance end is extremely large, and the operation and maintenance efficiency is greatly reduced. From the foregoing, how to provide a building photovoltaic remote diagnosis and collaborative operation and maintenance method and system based on artificial intelligence, which can deeply fuse building information to realize accurate remote fault diagnosis and improve operation and maintenance efficiency, has become a subject to be studied in the field. Disclosure of Invention The invention aims to provide an artificial intelligence-based building photovoltaic remote diagnosis and collaborative operation and maintenance method and system, which are used for solving the problems in the prior art. In order to achieve the above purpose, the present invention adopts the following technical scheme: in a first aspect, the invention provides an artificial intelligence-based building photovoltaic remote diagnosis and collaborative operation and maintenance method, which comprises the following steps: Building information and photovoltaic distribution information are obtained, and a building photovoltaic three-dimensional digital twin model is built according to the building information and the photovoltaic distribution information; Acquiring real-time electric state data and real-time physical state data of each photovoltaic module at each moment in a preset sampling time window, and respectively associating the real-time electric state data and the real-time physical state data of each photovoltaic module on the corresponding photovoltaic module in the building photovoltaic three-dimensional digital twin model to form a standard photovoltaic data stream of each photovoltaic module; Integrating the standard photovoltaic data flow of e