Photovoltaic box transformer substation fault prediction method and system, electronic equipment and storage medium

Abstract

The invention discloses a photovoltaic box transformer substation fault prediction method and system, electronic equipment and a storage medium. During operation of a photovoltaic box transformer substation, various problems occur to the box transformer substation due to objective reasons. Sensors are arranged at a transformer, a circuit breaker and a cable joint of a photovoltaic box transformer substation and inlet wire and outlet wire ends of the photovoltaic box transformer substation, and operation data of the box transformer substation are collected and transmitted to a local edge calculation unit; preprocessing the operation data of the box transformer substation; constructing a local embedded photovoltaic box transformer substation fault prediction model, wherein the model outputs a fault diagnosis result, a fault prediction result and a time window; a standardized data interface is constructed in a local edge calculation unit of the photovoltaic box transformer substation, bidirectional data interaction with a power station digital twinborn platform is carried out, and the power station digital twinborn platform calibrates a digital twinborn model and outputs a fault operation and maintenance decision after receiving operation data of the photovoltaic box transformer substation. According to the invention, fault diagnosis and prediction of the photovoltaic box transformer substation are realized, and the operation and maintenance efficiency of the photovoltaic box transformer substation is improved.

Description

TECHNICAL FIELD

[0002] The application belongs to the technical field of smart grids, and particularly relates to a photovoltaic box transformer fault prediction method and system, an electronic device, and a storage medium. BACKGROUND

[0003] In the long-term operation of a photovoltaic box transformer, due to various objective reasons, the following problems occur after the box transformer is put into operation: cable lap joint fault, busbar fault, operating mechanism fault, lightning arrester fault, transformer fault, insulation flashover fault caused by condensation in the box transformer, waterproof and fireproof problem, anti-theft problem, etc., thereby causing hidden dangers to the safety of the equipment itself and the reliable operation of the power grid, and directly affecting the safe and stable operation of the equipment.

[0004] As the core power distribution equipment of a photovoltaic power station, a photovoltaic box transformer bears the key functions of power conversion, voltage regulation and line protection, and its operating state directly affects the power generation efficiency of the power station and the safety of the power grid. The existing photovoltaic box transformer operation and maintenance has the following technical pain points: Single sensing dimension: the traditional box transformer only relies on basic electrical parameters such as current and voltage for monitoring, and cannot capture internal hidden faults of the equipment, such as transformer winding insulation deterioration, cable joint local overheating, and poor contact of circuit breakers, resulting in sudden failure and shutdown; Low fault diagnosis accuracy: relying on manual inspection and threshold alarm, it is difficult to distinguish complex fault types, such as mixed faults of high resistance grounding and inter-turn short circuit, and misoperation may occur, such as unnecessary shutdown or refusal to act due to misjudgment of faults, and equipment burning due to missed faults; Passive operation and maintenance mode: using a regular inspection combined with post-fault maintenance mode, it is impossible to predict the equipment deterioration trend in advance, the non-planned downtime accounts for a high proportion, up to 8-15 hours per year per unit, and the operation and maintenance cost is high; Lack of digital collaboration: the box transformer operation data is disconnected from the digital twin system of the power station, and it is impossible to realize unified visual management of the equipment state of the whole power station, and the operation and maintenance decision lacks global data support. SUMMARY

[0005] In order to make up for the deficiencies of the prior art, the application provides a photovoltaic box transformer fault prediction method, system, electronic device, and storage medium, which integrates multi-dimensional sensing, intelligent analysis and digital collaboration, realizes fault diagnosis and prediction of the photovoltaic box transformer, and improves the operation and maintenance efficiency of the photovoltaic box transformer.

[0006] In order to achieve the above purpose, the technical scheme adopted by the application is as follows: A photovoltaic box transformer fault prediction method, comprising the following steps: Step S10: deploying sensors at the transformer, circuit breaker, cable joint of the photovoltaic box transformer, and the incoming and outgoing lines of the photovoltaic box transformer, collecting photovoltaic box transformer operation data, and transmitting the photovoltaic box transformer operation data to an edge computing unit installed locally on the photovoltaic box transformer; Step S20: data preprocessing is performed on the photovoltaic box transformer operation data collected in step S10; Step S30: a local embedded photovoltaic box transformer fault prediction model is constructed, historical data of normal operation and fault of the photovoltaic box transformer are collected as a training set to train the model, after the model training is completed, the photovoltaic box transformer operation data after data preprocessing is input, and the model outputs fault diagnosis result, fault prediction result and time window; Step S40: a standardized data interface is constructed in the edge computing unit of the local photovoltaic box transformer, and bidirectional data interaction is performed with the power station digital twin platform, after the power station digital twin platform receives the photovoltaic box transformer operation data, the digital twin model is calibrated and fault operation and maintenance decision is output.

[0007] Further, the step S10 comprises the following steps: Step 1.1: sensor selection and deployment An acoustic fingerprint sensor, a partial discharge sensor, a temperature optical fiber sensor and a vibration sensor are deployed on the transformer of the photovoltaic box transformer, an acoustic fingerprint sensor and a temperature sensor are deployed on the circuit breaker, a temperature optical fiber sensor and a partial discharge sensor are deployed at the cable joint, and a current voltage sensor is deployed at the inlet and outlet of the photovoltaic box transformer; Step 1.2 data acquisition and transmission Each sensor is configured with an independent data acquisition unit, and an edge computing unit is used to realize local data aggregation, and at the same time, backup to a local storage module.

[0008] Further, the step S20 comprises the following steps: Step 2.1: data cleaning Abnormal values are removed, and missing data is completed by using linear interpolation method combined with LSTM prediction completion method; Step 2.2: filtering environmental interference signals The environmental noise in the acoustic fingerprint signal is separated by wavelet transform, and the electromagnetic interference in the temperature signal is eliminated by Kalman filter; Step 2.3: feature extraction and normalization Data features are extracted; Z-Score standardization is used to normalize all features to the interval [-1, 1]; the preprocessed feature data is packaged as a device state feature vector, in the format of {device ID, timestamp, feature name 1-feature value 1,..., feature name N-feature value N}, wherein N represents the number of feature types.

[0009] Further, the step S30 comprises the following steps: 3.1: model training in offline stage The embedded photovoltaic box transformer fault prediction model is constructed, a photovoltaic box transformer fault diagnosis submodel and a photovoltaic box transformer fault prediction submodel double module architecture are adopted, and the embedded photovoltaic box transformer fault prediction model is deployed in a local edge computing unit of the photovoltaic box transformer; the photovoltaic box transformer fault diagnosis submodel adopts an improved convolutional neural network (CNN) combined with a bidirectional long short-term memory (BiLSTM) architecture; and the photovoltaic box transformer fault prediction submodel adopts a time series prediction model; 3.2: Construction of training data set Historical fault data, simulated fault data and historical normal operation data of the photovoltaic box transformer are collected as a training data set of the embedded photovoltaic box transformer fault prediction model; 3.3: Real-time fault diagnosis and prediction in online stage The edge computing unit calls the photovoltaic box transformer fault prediction model once every specified time, inputs the latest device state feature vector to the diagnosis submodel or the feature vector sequence to the prediction submodel, and sets the fault severity classification as 1 to 5 levels; when the fault diagnosis submodel outputs a result of fault severity ≥ 3 levels, a local sound and light alarm is triggered immediately, and fault information is pushed to a power station operation and maintenance platform, including device ID, fault type and severity; when the fault prediction submodel outputs a result of a certain fault probability ≥ 70%, a fault warning report is generated, including a fault probability trend curve, a recommended maintenance time window and a list of spare parts to be replaced; 3.4: Model self-iteration Each new fault or normal data is accumulated, and incremental training is automatically started to complete self-iteration of the photovoltaic box transformer fault prediction model.

[0010] Further, the step S40 comprises the following steps: 4.1: Development of standardized data interface A standardized data interface is developed in the edge computing unit for bidirectional data interaction with the power station digital twin platform; output data of the standardized data interface includes real-time sensing data, output results of the photovoltaic box transformer fault prediction model and device basic information; received data of the standardized data interface includes calibration parameters of the digital twin model and global load scheduling instructions of the power station; 4.2: Calibration of digital twin model and output of operation and maintenance decision After the power station digital twin platform receives real-time data of the photovoltaic box transformer, the real-time data are compared with simulated data of the photovoltaic box transformer digital twin body; when a deviation between the simulated data and the measured data is ≥ 5%, boundary conditions of the digital twin model are automatically adjusted to realize dynamic calibration of the model, ensure state consistency between the digital twin body and the physical box transformer, simulate operation and maintenance scenarios based on the calibrated digital twin body, including shutdown loss and maintenance cost, output an optimal operation and maintenance decision scheme, and feed back the optimal operation and maintenance decision scheme to the edge computing unit ECU of the photovoltaic box transformer, and synchronously push the optimal operation and maintenance decision scheme to a mobile terminal of an operation and maintenance personnel to guide on-site operation.

[0011] A photovoltaic box transformer fault prediction system comprises a photovoltaic box transformer operation data acquisition and transmission module: sensors are arranged at the transformer, circuit breaker and cable joint of the photovoltaic box transformer and the inlet and outlet of the photovoltaic box transformer to acquire photovoltaic box transformer operation data and transmit the data to an edge computing unit installed locally in the photovoltaic box transformer. A photovoltaic box transformer operation data preprocessing module: used for data preprocessing of the acquired photovoltaic box transformer operation data; A photovoltaic box transformer fault prediction module: used for constructing a local embedded photovoltaic box transformer fault prediction model, collecting historical data of normal operation and fault of the photovoltaic box transformer as a training set to train the model, inputting the data preprocessed photovoltaic box transformer operation data after the model training is completed, and outputting fault diagnosis results, fault prediction results and a time window; A photovoltaic box transformer fault operation and maintenance decision module: used for constructing a standardized data interface in the edge computing unit locally in the photovoltaic box transformer, performing bidirectional data interaction with a power station digital twin platform, calibrating a digital twin model after the power station digital twin platform receives the photovoltaic box transformer operation data, and outputting fault operation and maintenance decisions.

[0012] An electronic device comprising a memory for storing a computer program; A processor for executing the computer program to implement the steps of the photovoltaic box transformer fault prediction method according to any one of claims 1 to 5.

[0013] A computer readable storage medium having a computer program stored therein, wherein the computer program, when executed by a processor, implements the steps of the photovoltaic box transformer fault prediction method.

[0014] The present application has the following advantages: 1) The present application realizes real-time, all-around and non-intrusive monitoring of the operation state of the key equipment of the photovoltaic box transformer by integrating more and more accurate sensors; from traditional fault maintenance, it is upgraded to predictive maintenance and condition-based maintenance, the device degradation trend is found in advance, sudden failure is avoided, unplanned downtime is reduced, device life is prolonged, and operation and maintenance strategy is optimized; at the same time, by combining AI algorithm, the fault diagnosis and prediction accuracy is improved, and more intelligent fault handling is realized; 2) The present application creates a digital twin of the photovoltaic box transformer by using digital twin technology, realizes full-dimensional perception, not only improves the operation and maintenance efficiency and management level of the box transformer, but also reduces the operation and maintenance cost and risk, and provides a strong guarantee for the stable operation of the power system. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 The flowchart of the photovoltaic box transformer fault prediction method of the present application; Figure 2 This is a schematic diagram of the photovoltaic transformer substation fault prediction system of the present invention; Figure 3 This is a schematic block diagram of the photovoltaic transformer fault prediction electronic device of the present invention. Detailed Implementation

[0016] The present invention will now be described in detail with reference to specific embodiments.

[0017] This invention provides a method, system, electronic device, and storage medium for predicting faults in photovoltaic transformer substations. It integrates multi-dimensional perception, intelligent analysis, and digital collaboration to achieve fault diagnosis and prediction of photovoltaic transformer substations, thereby improving the operation and maintenance efficiency of photovoltaic transformer substations.

[0018] like Figure 1 As shown, the photovoltaic transformer substation fault prediction method of the present invention includes the following steps: Step S10: Deploy sensors at the transformer, circuit breaker, cable joints, and incoming and outgoing lines of the photovoltaic transformer substation to collect operating data and transmit it to the edge computing unit installed locally within the photovoltaic transformer substation; including the following steps: Step 1.1: Sensor Selection and Deployment Acoustic sensors, partial discharge sensors, temperature fiber optic sensors, and vibration sensors are deployed on the transformer of the photovoltaic transformer substation. The acoustic sensors are installed on the outer wall of the transformer tank, with a sampling rate ≥44.1kHz and a monitoring range of 20Hz-20kHz, used to capture abnormal acoustic signatures caused by core loosening and winding deformation. The partial discharge sensors are installed at the high-voltage bushings of the transformer, with a sensitivity ≤1pC, used to monitor partial discharge signals caused by insulation degradation. The temperature fiber optic sensors are embedded along the transformer winding axis, with a temperature measurement accuracy of ±0.5℃ and a spatial resolution of 1m, realizing the measurement of temperature within the winding. A distributed temperature monitoring system is implemented. Vibration sensors are installed on the transformer base, with a measurement range of 0-500Hz, to monitor abnormal core vibration. Acoustic and temperature sensors are deployed on the circuit breakers; the acoustic sensors are installed on the circuit breaker operating mechanism housing to monitor abnormal opening and closing sounds; the temperature sensors are non-contact infrared thermometers with an accuracy of ±1℃, used to monitor contact overheating. Temperature fiber optic sensors and partial discharge sensors are deployed at cable joints, and current and voltage sensors are deployed at the photovoltaic transformer's incoming and outgoing lines. The temperature fiber optic sensors are wound-mounted, with a temperature range of -40℃ to 125℃, monitoring overheating caused by excessive joint contact resistance; the partial discharge sensors are attached to the surface to monitor partial discharge caused by insulation damage. High-precision current and voltage sensors are deployed at the photovoltaic transformer's incoming and outgoing lines, with current measurement accuracy of 0.2% and voltage measurement accuracy of 0.2%, used to collect basic current and voltage data. All deployed sensors are non-intrusive to avoid damaging the original insulation structure and sealing performance of the transformer.

[0019] Step 1.2: Data acquisition and transmission Each sensor is configured with an independent data acquisition unit, which implements local data aggregation using an edge computing unit, supports LoRa / Wi-Fi / Ethernet communication, and adjusts the acquisition frequency according to the monitored object, such as 1 time / second for electrical parameters, 1 time / 10 seconds for temperature / vibration, 1 time / minute for voiceprint / partial discharge, and automatically increases to 1 time / 10 seconds in abnormal conditions. Data is transmitted to the local edge computing unit ECU through an encryption protocol, such as the MQTT-SN protocol, and is simultaneously backed up to a local storage module with a storage capacity of ≥1TB, supporting data lossless power failure.

[0020] Step S20: Data preprocessing of the photovoltaic transformer operation data collected in step S10; including the following steps: Step 2.1: Data cleaning Abnormal values are removed using the 3σ criterion, such as data outside the normal range caused by sensor failure, and missing data is completed using linear interpolation combined with LSTM prediction completion method. Linear interpolation is used to handle short-term missing data, such as missing time ≤5 minutes, and LSTM prediction completion method is used for long-term missing data, such as missing time of 5-30 minutes; Step 2.2: Filter environmental interference signals Separate environmental noise in the voiceprint signal through wavelet transform, such as db4 wavelet basis, and eliminate electromagnetic interference in the temperature signal through Kalman filtering; Step 2.3: Feature extraction and normalization Extract data features; for voiceprint signals, extract 12 time or frequency domain features such as Mel frequency cepstral coefficients MFCC, spectral centroid, and spectral bandwidth; for partial discharge data, extract 8 features such as discharge quantity peak, discharge frequency, average discharge current, and discharge phase distribution; for temperature / vibration / current / voltage data, extract 5 statistical features such as mean, variance, maximum, minimum, and rate of change; Use Z-Score standardization to normalize all features to the [-1, 1] interval, eliminating the impact of dimension differences on subsequent AI models; encapsulate the preprocessed feature data as a device state feature vector in the format {device ID, timestamp, feature name1-feature value1,..., feature nameN-feature valueN}, where N represents the number of feature types.

[0021] Step S30: Construct a local embedded photovoltaic transformer fault prediction model, collect historical data of normal operation and fault of the photovoltaic transformer as a training set to train the model, input the photovoltaic transformer operation data after data preprocessing after the model training is completed, and the model outputs fault diagnosis results, fault prediction results, and time windows; including the following steps: 3.1: Model Training in the Offline Phase An embedded photovoltaic transformer substation fault prediction model is constructed, adopting a dual-module architecture of a photovoltaic transformer substation fault diagnosis submodel and a photovoltaic transformer substation fault prediction submodel, deployed on the local edge computing unit of the photovoltaic transformer substation to avoid cloud transmission delays; the photovoltaic transformer substation fault diagnosis submodel adopts an improved convolutional neural network (CNN) combined with a bidirectional long short-term memory network (BiLSTM) architecture; the photovoltaic transformer substation fault prediction submodel adopts a time series prediction model, such as Temporal Fusion Transformer (TFT).

[0022] 3.2: Constructing the training dataset Historical fault data, simulated fault data, and historical normal operation data of photovoltaic transformer substations were collected as training datasets for an embedded AI photovoltaic transformer substation fault prediction model. The historical fault data consisted of fault cases from 500 photovoltaic transformer substations, covering 12 typical faults and 3 composite faults, including high-resistance grounding, inter-turn short circuit, cable joint overheating, and circuit breaker malfunction. The simulated fault data was generated by building a transformer substation fault simulation platform in the laboratory to generate simulated data with different degrees of degradation, supplementing the scarce composite fault samples in the historical data. The normal operation data consisted of three consecutive months of normal operation data from 1,000 transformer substations.

[0023] 3.3: Real-time Fault Diagnosis and Prediction in the Online Phase The edge computing unit calls the photovoltaic transformer fault prediction model every specified time interval, inputs the latest equipment status feature vector into the diagnostic sub-model or the feature vector sequence into the prediction sub-model, and sets the fault severity level to 1 to 5. When the fault diagnosis sub-model outputs a fault severity level ≥ 3, a local audible and visual alarm is immediately triggered, and fault information is pushed to the power plant operation and maintenance platform, including the equipment ID, fault type, and severity. When the fault prediction sub-model outputs a fault probability ≥ 70%, a fault warning report is generated, which includes a fault probability trend curve, a recommended maintenance time window, and a list of spare parts to be replaced.

[0024] The input to the photovoltaic transformer substation fault diagnosis submodel is the equipment state feature vector output from step S20. The output is the fault type and fault severity, where the fault severity is divided into 1-5 levels, with level 1 being slight degradation and level 5 being severe fault. During training, an attention mechanism is used to strengthen the weight of key features, such as the phase distribution features of partial discharge signals, to improve the accuracy of distinguishing complex faults. The input to the photovoltaic transformer substation fault prediction submodel is the equipment state feature vector over a certain period of time (the equipment state feature vector over the past 72 hours). The output is the probability of various faults occurring in the equipment within a certain period of time in the future and the fault critical time window (e.g., the probability of transformer winding insulation degradation fault ≥85% within the next 15-20 days). At the same time, equipment life cycle parameters, such as years of operation and cumulative load rate, are introduced as auxiliary features to improve prediction robustness.

[0025] 3.4: Model Self-Iteration For every 100 new fault or normal data points accumulated, incremental training is automatically initiated. This is used only to update model parameters and does not reconstruct the network structure. This completes the self-iteration of the photovoltaic transformer fault prediction model, ensuring that the model adapts to dynamic scenarios such as transformer aging and environmental changes.

[0026] Step S40: Construct a standardized data interface in the local edge computing unit of the photovoltaic transformer substation to conduct bidirectional data interaction with the power station digital twin platform. After receiving the photovoltaic transformer substation's operating data, the power station digital twin platform calibrates the digital twin model and outputs fault operation and maintenance decisions; including the following steps: 4.1: Standardized Data Interface Development A standardized data interface based on the OPCUA protocol was developed in the edge computing unit (ECU) for bidirectional data interaction with the power plant digital twin platform. The output data of the standardized data interface includes real-time sensing data, the output results of the AI ​​photovoltaic transformer substation fault prediction model, and basic equipment information. The real-time sensing data includes the raw values ​​of all sensors and the feature values ​​of the preprocessed data. The output results of the AI ​​photovoltaic transformer substation fault prediction model include fault diagnosis results, fault prediction probability, and time window. The basic equipment information includes model, years of operation, and cumulative operating time. The received data of the standardized data interface includes the calibration parameters of the digital twin model and the global load dispatching instructions of the power plant. The calibration parameters of the digital twin model include the ambient temperature correction factor and the equipment aging factor, etc. 4.2: Digital Twin Model Calibration and Operation and Maintenance Decision Output After receiving real-time data from the photovoltaic transformer substation, the power station's digital twin platform compares it with the simulated data of the digital twin of the photovoltaic transformer substation, such as the transformer winding temperature calculated based on a thermodynamic model and the circuit breaker vibration value calculated based on a mechanical model. When the deviation between the simulated data and the measured data is ≥5%, the platform automatically adjusts the boundary conditions of the digital twin model, such as correcting the heat dissipation coefficient and contact resistance parameters, to achieve dynamic model calibration and ensure the consistency of the state between the digital twin and the physical transformer substation. Based on the calibrated digital twin, the platform simulates operation and maintenance scenarios, such as simulating the economic comparison between repairing the transformer 10 days in advance and repairing it after a failure, including downtime losses and maintenance costs. The platform outputs the optimal operation and maintenance decision plan and feeds it back to the edge computing unit (ECU) of the photovoltaic transformer substation, which is then simultaneously pushed to the mobile terminal of the operation and maintenance personnel to guide on-site operations.

[0027] like Figure 2 As shown, the photovoltaic transformer substation fault prediction system of the present invention includes a photovoltaic transformer substation operation data acquisition and transmission module, a photovoltaic transformer substation operation data preprocessing module, a photovoltaic transformer substation fault prediction module, and a photovoltaic transformer substation fault operation and maintenance decision module. 1) Photovoltaic transformer box operation data acquisition and transmission module: This module is used to deploy sensors at the transformer, circuit breaker, cable joints, and incoming and outgoing terminals of the photovoltaic transformer box to collect the operation data of the photovoltaic transformer box and transmit it to the edge computing unit installed locally in the photovoltaic transformer box.

[0028] 2) Photovoltaic transformer substation operation data preprocessing module: used to preprocess the collected photovoltaic transformer substation operation data; 3) Photovoltaic transformer fault prediction module: used to build a local embedded photovoltaic transformer fault prediction model. It collects historical data of normal operation and fault occurrence of photovoltaic transformers as training set to train the model. After the model is trained, it inputs the photovoltaic transformer operation data after data preprocessing, and the model outputs fault diagnosis results, fault prediction results and time window. 4) Photovoltaic transformer box fault operation and maintenance decision module: It is used to build a standardized data interface in the local edge computing unit of the photovoltaic transformer box, and to conduct bidirectional data interaction with the power station digital twin platform. After receiving the photovoltaic transformer box operation data, the power station digital twin platform calibrates the digital twin model and outputs fault operation and maintenance decisions.

[0029] The present invention also provides an electronic device, including a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the photovoltaic transformer fault prediction method of the present invention.

[0030] like Figure 3As shown in the illustration, in one embodiment of the present invention, a structural schematic diagram of a photovoltaic transformer substation fault prediction device based on intelligent state perception, suitable for implementing the embodiments of this application, is presented. The photovoltaic transformer substation fault prediction device based on intelligent state perception in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 3 The photovoltaic transformer fault prediction device based on intelligent state perception shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0031] Figure 3 The illustrated photovoltaic transformer substation fault prediction device based on intelligent state awareness may include a processor 1001 (e.g., a central processing unit, graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1002 or a program loaded from storage device 1003 into machine-readable storage medium (RAM) 1004. The machine-readable storage medium 1004 also stores various programs and data required for the operation of the photovoltaic transformer substation fault prediction device based on intelligent state awareness. The processor 1001, ROM 1002, and machine-readable storage medium 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and a communication unit 1009. Communication unit 1009 allows the intelligent state-aware photovoltaic substation fault prediction device to exchange data with other devices wirelessly or via wired communication. Although the figure shows an intelligent state-aware photovoltaic substation fault prediction device with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0032] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication unit, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processor 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0033] The photovoltaic substation fault prediction device based on intelligent state perception provided in this application adopts the photovoltaic substation fault prediction method based on intelligent state perception in the above embodiments, which can solve the technical problems of low fault diagnosis accuracy, lack of digital collaboration, and low fault repair efficiency in the operation and maintenance of existing photovoltaic substations. Compared with the prior art, the beneficial effects of the photovoltaic substation fault prediction device based on intelligent state perception provided in this application are the same as the beneficial effects of the photovoltaic substation fault prediction method based on intelligent state perception provided in the above embodiments, and other technical features in the photovoltaic substation fault prediction device based on intelligent state perception are the same as the features disclosed in the method of the previous embodiment, and will not be repeated here.

[0034] The various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0035] This invention also provides a computer-readable storage medium storing a computer program, wherein when executed by a processor, the computer program implements the steps of the photovoltaic transformer substation fault prediction method of this invention. The computer-readable storage medium provided by this invention can solve the technical problems of low fault diagnosis accuracy, lack of digital collaboration, and low fault repair efficiency in the operation and maintenance of existing photovoltaic transformer substations.

[0036] Those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims and their equivalents, this invention is also intended to include these modifications and variations.

Claims

1. A method for predicting faults in photovoltaic transformer substations, characterized in that: Includes the following steps: Step S10: Deploy sensors at the transformer, circuit breaker, cable joints, and incoming and outgoing terminals of the photovoltaic transformer box to collect the operating data of the photovoltaic transformer box and transmit it to the edge computing unit installed locally in the photovoltaic transformer box; Step S20: Perform data preprocessing on the photovoltaic transformer operation data collected in step S10; Step S30: Construct a local embedded photovoltaic transformer substation fault prediction model. Collect historical data of normal operation and fault occurrence of photovoltaic transformer substation as training set to train the model. After the model is trained, input the photovoltaic transformer substation operation data after data preprocessing. The model outputs fault diagnosis results, fault prediction results and time window. Step S40: Build a standardized data interface in the local edge computing unit of the photovoltaic transformer to conduct bidirectional data interaction with the power station digital twin platform. After receiving the photovoltaic transformer operation data, the power station digital twin platform calibrates the digital twin model and outputs fault operation and maintenance decisions.

2. The photovoltaic transformer substation fault prediction method according to claim 1, characterized in that: Step S10 includes the following steps: Step 1.1: Sensor Selection and Deployment Sound signature sensors, partial discharge sensors, temperature fiber optic sensors, and vibration sensors are deployed on the transformers of the photovoltaic transformer substation; sound signature sensors and temperature sensors are deployed on the circuit breakers; temperature fiber optic sensors and partial discharge sensors are deployed at the cable joints; and current and voltage sensors are deployed at the incoming and outgoing terminals of the photovoltaic transformer substation. Step 1.2 Data Acquisition and Transmission Each sensor is equipped with an independent data acquisition unit, and uses an edge computing unit to achieve local data aggregation, while backing up to a local storage module.

3. The photovoltaic transformer substation fault prediction method according to claim 2, characterized in that: Step S20 includes the following steps: Step 2.1: Data Cleaning Outliers are removed, and missing data is filled using a combination of linear interpolation and LSTM prediction. Step 2.2: Filtering environmental interference signals Environmental noise in the acoustic signature signal is separated by wavelet transform, and electromagnetic interference in the temperature signal is eliminated by Kalman filtering. Step 2.3: Feature Extraction and Normalization Extract data features; Z-Score standardization is used to normalize all features to the range [-1,1]. The preprocessed feature data is encapsulated into a device status feature vector with the format {Device ID, Timestamp, Feature Name 1 - Feature Value 1, ..., Feature Name N - Feature Value N}, where N represents the number of feature types.

4. The photovoltaic transformer substation fault prediction method according to claim 3, characterized in that: Step S30 includes the following steps: 3.1: Model Training in the Offline Phase An embedded photovoltaic transformer substation fault prediction model is constructed, adopting a dual-module architecture of a photovoltaic transformer substation fault diagnosis submodel and a photovoltaic transformer substation fault prediction submodel, deployed on the local edge computing unit of the photovoltaic transformer substation; the photovoltaic transformer substation fault diagnosis submodel adopts an improved convolutional neural network (CNN) combined with a bidirectional long short-term memory network (BiLSTM) architecture; the photovoltaic transformer substation fault prediction submodel adopts a time-series prediction model. 3.2: Constructing the training dataset Historical fault data, simulated fault data, and historical normal operation data of photovoltaic transformer substations are collected as training datasets for embedded photovoltaic transformer substation fault prediction models. 3.3: Real-time Fault Diagnosis and Prediction in the Online Phase The edge computing unit calls the photovoltaic transformer fault prediction model every specified time, inputs the latest equipment status feature vector into the diagnostic sub-model or the feature vector sequence into the prediction sub-model, and sets the fault severity level to 1 to 5. When the fault diagnosis sub-model outputs a fault severity level ≥ 3, it immediately triggers a local audible and visual alarm and pushes fault information to the power plant operation and maintenance platform, including equipment ID, fault type and severity. When the fault prediction sub-model outputs a fault probability of ≥70%, a fault warning report is generated, which includes a fault probability trend curve, a recommended maintenance time window, and a list of spare parts to be replaced. 3.4: Model Self-Iteration Each time new fault or normal data is accumulated, incremental training is automatically started to complete the self-iteration of the photovoltaic transformer fault prediction model.

5. A method for predicting faults in photovoltaic transformer substations according to claim 4, characterized in that: Step S40 includes the following steps: 4.1: Standardized Data Interface Development A standardized data interface is developed in the edge computing unit for bidirectional data interaction with the power plant digital twin platform. The output data of the standardized data interface includes real-time sensing data, the output results of the photovoltaic transformer fault prediction model, and basic equipment information. The received data of the standardized data interface includes the calibration parameters of the digital twin model and the power plant global load dispatching instructions. 4.2: Digital Twin Model Calibration and Operation and Maintenance Decision Output After receiving real-time data from the photovoltaic transformer, the power station's digital twin platform compares it with the simulated data of the photovoltaic transformer's digital twin. When the deviation between the simulated data and the measured data is ≥5%, the platform automatically adjusts the boundary conditions of the digital twin model to ensure the consistency between the digital twin and the physical transformer's state. Based on the calibrated digital twin, the platform simulates operation and maintenance scenarios, including downtime losses and maintenance costs, and outputs the optimal operation and maintenance decision-making scheme. The digital twin platform feeds back the optimal operation and maintenance decision-making scheme to the edge computing unit of the photovoltaic transformer and simultaneously pushes it to the mobile terminal of the operation and maintenance personnel to guide on-site operations.

6. A photovoltaic transformer substation fault prediction system, characterized in that: Includes a photovoltaic transformer substation operation data acquisition and transmission module: used to deploy sensors at the transformer, circuit breaker, cable joints and the incoming and outgoing ends of the photovoltaic transformer substation to collect photovoltaic transformer substation operation data and transmit it to the edge computing unit installed locally in the photovoltaic transformer substation; Photovoltaic transformer substation operation data preprocessing module: used to preprocess the collected photovoltaic transformer substation operation data; Photovoltaic transformer fault prediction module: It is used to build a local embedded photovoltaic transformer fault prediction model. It collects historical data of normal operation and fault occurrence of photovoltaic transformers as training set to train the model. After the model is trained, it inputs the photovoltaic transformer operation data after data preprocessing and the model outputs fault diagnosis results, fault prediction results and time window. Photovoltaic transformer fault operation and maintenance decision module: It is used to build a standardized data interface in the local edge computing unit of the photovoltaic transformer, and to conduct bidirectional data interaction with the power station digital twin platform. After receiving the photovoltaic transformer operation data, the power station digital twin platform calibrates the digital twin model and outputs fault operation and maintenance decisions.

7. An electronic device, characterized in that: Includes memory used to store computer programs; A processor is configured to execute the computer program to implement the steps of the photovoltaic transformer fault prediction method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, it implements the steps of the photovoltaic transformer fault prediction method as described in any one of claims 1 to 5.

Images