Photovoltaic power station intelligent monitoring method and device based on dynamic weight tree diagram and edge-cloud cooperation

Abstract

The invention provides a photovoltaic power station intelligent monitoring system and method based on a dynamic weight tree diagram and edge-cloud collaboration, and relates to the technical field of photovoltaic power station intelligent monitoring, and the method comprises the steps: obtaining environment parameters such as illumination intensity and temperature of a photovoltaic power station, and equipment operation state data such as voltage and current; calculating and adjusting the dynamic weight of each monitoring index in real time according to the illumination / temperature deviation condition and the equipment health state; a multi-level tree diagram is constructed based on the dynamic weight, the upper layer represents the overall operation state of the power station, the lower layer represents a subsystem evaluation result, and the node priority and the hierarchical structure are synchronously updated; the edge computing node preliminarily processes abnormal data, uploads data conforming to a threshold value and adjusts a local control strategy; and the cloud platform issues optimization model parameters after deep diagnosis, and iteratively updates the dynamic weight calculation model. According to the invention, the fault false alarm rate is effectively reduced, and the fault detection precision and response efficiency of the photovoltaic power station in a complex environment are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for photovoltaic power plants, and in particular to an intelligent monitoring method and device for photovoltaic power plants based on dynamic weighted tree diagrams and edge-cloud collaboration. Background Technology

[0002] Photovoltaic power plant monitoring systems, as a crucial technological support in the field of new energy power generation, are widely used in large-scale distributed photovoltaic grid-connected scenarios. With the rapid development of the photovoltaic industry, existing technologies have constructed a monitoring system centered on operating parameters such as voltage and current through the collaborative operation of SCADA systems, fixed-weight monitoring models, and centralized data processing. Specifically, this technological system covers the entire process from data acquisition to fault diagnosis, including key aspects such as equipment status monitoring, environmental parameter analysis, and visualization. However, traditional binary fluctuation analysis methods rely on static weight allocation and typically employ a single threshold judgment mechanism set by expert experience, making it difficult to adapt to varying light intensity (…). ),temperature( Dynamic changes in environmental parameters such as ( ).

[0003] However, existing photovoltaic power plant monitoring methods directly use monitoring indicators with fixed weights and do not establish a real-time correlation mechanism with environmental parameters (such as irradiance and temperature) and equipment status (SOC value). This may lead to a significant increase in false alarm rate under special operating conditions such as cloudy and rainy days (the measured false judgment rate can reach 40%), or the inability to distinguish between performance degradation and equipment failure when there is insufficient sunlight, thereby affecting the operation and maintenance efficiency of the power plant and economic losses. Summary of the Invention

[0004] The present invention aims to at least partially solve one of the technical problems in the related art.

[0005] Therefore, the first objective of this invention is to propose an intelligent monitoring method for photovoltaic power plants based on dynamic weighted tree diagrams and edge-cloud collaboration.

[0006] The second objective of this invention is to propose an intelligent monitoring device for photovoltaic power plants based on dynamic weighted tree diagrams and edge-cloud collaboration.

[0007] The third objective of this invention is to provide an electronic device.

[0008] The fourth objective of this invention is to provide a computer-readable storage medium.

[0009] The fifth objective of this invention is to provide a computer program product.

[0010] To achieve the above objectives, a first aspect of the present invention proposes an intelligent monitoring method for photovoltaic power plants based on dynamic weighted tree diagrams and edge-cloud collaboration, comprising: S1, acquiring real-time environmental parameters and equipment operation state data of the photovoltaic power station, the environmental parameters including light intensity and temperature, and the equipment operation state data including voltage, current, power and equipment state; S2, calculating dynamic weights of each monitoring index based on the environmental parameters and the equipment operation state data, the dynamic weights being adjusted in real time according to a degree of deviation of light intensity from a reference value, a nonlinear influence of temperature deviation from a reference value and an equipment health state; S3, constructing a multi-level tree diagram by using the dynamic weights, an upper node of the tree diagram representing an overall operation state of the power station, and a lower node representing a state evaluation result of each subsystem, and updating a node priority and a hierarchical structure of the tree diagram according to the dynamic weights; S4, performing preliminary processing on abnormal data by an edge computing node, encapsulating abnormal data meeting a preset threshold and uploading the abnormal data to a cloud platform, and adjusting a local control strategy according to the dynamic weights; S5, performing deep fault diagnosis on the received abnormal data by the cloud platform, generating optimized model parameters and delivering the model parameters to the edge computing node, and realizing iterative updating of the dynamic weight calculation model.

[0011] Optionally, the acquiring real-time environmental parameters and equipment operation state data of the photovoltaic power station further comprises: S11, collecting voltage, current, power, temperature, irradiance and equipment state data by multi-source sensor fusion; S12, performing cleaning, standardization and normalization processing on the collected data, and buffering the data to an edge node after encryption and compression.

[0012] Optionally, the calculating dynamic weights of each monitoring index based on the environmental parameters and the equipment operation state data further comprises: S21, calculating a dynamic weight value of an i-th photovoltaic component by using the following formula:

[0013] wherein, is a state health degree, is a real-time temperature, and is a real-time light intensity; S22, realizing real-time adjustment of the weight by using an exponential decay function, the exponential decay function being as follows: wherein, is a decay coefficient, and is an abnormal degree of an i-th index.

[0014] wherein, is a decay coefficient, and is an abnormal degree of an i-th index.

[0015] ​​​​​​Optionally, the preliminary processing of the abnormal data by the edge computing node further includes: S41, a lightweight ELC-RNN model is run using a Raspberry Pi CM4 device to collect device operation data at an interval of 100 ms; S42, when it is detected that the current fluctuation exceeds 15%, the abnormal data is packaged in JSON format and sent to the cloud platform through the MQTT protocol.

[0016] Optionally, the cloud platform performs deep fault diagnosis based on the received abnormal data, and further includes: S51, model training is performed once a week to optimize model parameters using a deep learning model; S52, the optimized model version is distributed to each edge computing node through the HTTP protocol to realize hot updating of the model.

[0017] Optionally, the further includes: S6, a digital twin interaction interface is started, a 3D model of the photovoltaic power station is loaded, and gesture operation is supported; S7, abnormal point labeling information is synchronized to the cloud platform in real time through a multi-person collaborative labeling function, and historical labeling record playback is supported.

[0018] To achieve the above purpose, a second aspect embodiment of the present application proposes a photovoltaic power station intelligent monitoring device based on dynamic weight tree diagram and edge-cloud collaboration, comprising: A data acquisition module is configured to acquire real-time environmental parameters and device operation state data of the photovoltaic power station, wherein the environmental parameters include light intensity and temperature, and the device operation state data includes voltage, current, power and device state; A dynamic weight calculation module is configured to calculate dynamic weights of each monitoring index based on the environmental parameters and device operation state data, wherein the dynamic weights are adjusted in real time according to the degree of deviation of light intensity from a reference value, the nonlinear influence of temperature deviation from a reference value, and the device health status; A tree diagram construction module is configured to construct a multi-level tree diagram using the dynamic weights, wherein the upper nodes of the tree diagram represent the overall operation state of the power station, the lower nodes represent the state evaluation results of each subsystem, and the node priority and hierarchical structure of the tree diagram are updated according to the dynamic weights; An edge computing processing module is configured to preliminarily process abnormal data by an edge computing node, encapsulate abnormal data meeting a preset threshold, and upload the abnormal data to a cloud platform, while adjusting a local control strategy according to the dynamic weights; A cloud platform diagnosis module is configured to perform deep fault diagnosis by the cloud platform based on the received abnormal data, generate optimized model parameters, and distribute the model parameters to the edge computing node to realize iterative updating of the dynamic weight calculation model.

[0019] To achieve the above object, the third aspect of the present application provides an electronic device, comprising: a processor, and a memory connected with the processor in communication; The memory stores computer execution instructions; The processor executes the computer execution instructions stored in the memory to realize the method according to any one of the first aspect.

[0020] To achieve the above object, the fourth aspect of the present application provides a computer readable storage medium, wherein the computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by a processor to realize the method according to any one of the first aspect.

[0021] To achieve the above object, the fifth aspect of the present application provides a computer program product, wherein the computer program is executed by a processor to realize the method according to any one of the first aspect.

[0022] The photovoltaic power station intelligent monitoring method, device, electronic device and storage medium provided by the present application can dynamically adjust the monitoring index weight according to environmental parameters such as illumination and temperature and equipment running state, significantly reduce the fault false alarm rate, and improve the fault detection accuracy and response efficiency of the photovoltaic power station under complex environment.

[0023] Additional aspects and advantages of the present application will be in part apparent and in part pointed out hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0024] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description, taken in conjunction with the accompanying drawings, in which: Figure 1 A flowchart of a photovoltaic power station intelligent monitoring method based on dynamic weight tree diagram and edge-cloud cooperation provided by an embodiment of the present application; Figure 2 A structural diagram of a photovoltaic power station intelligent monitoring device based on dynamic weight tree diagram and edge-cloud cooperation provided by an embodiment of the present application. DETAILED DESCRIPTION

[0025] Embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference signs represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present application, and cannot be understood as a limitation of the present application.

[0026] The embodiment of the application provides a photovoltaic power station intelligent monitoring method based on dynamic weight tree chart and edge-cloud cooperation, Figure 1 A flowchart of a photovoltaic power station intelligent monitoring method based on dynamic weight tree chart and edge-cloud cooperation provided by the embodiment of the application is shown in the figure. Figure 1 The method comprises the following steps: Step S1, acquiring real-time environment parameters and equipment operation state data of the photovoltaic power station, wherein the environment parameters comprise light intensity and temperature, and the equipment operation state data comprises voltage, current, power and equipment state.

[0027] Specifically, in the photovoltaic power station intelligent monitoring system, acquiring real-time environment parameters and equipment operation state data is a basic step for realizing dynamic weight tree chart construction and edge-cloud cooperative analysis. Through the sensor network deployed at the key nodes of the photovoltaic power station, multi-dimensional data including light intensity ( ), environmental temperature ( ), voltage ( ), current ( ), power ( ) and equipment state (such as running / stop, fault code) are collected. Among them, light intensity and temperature as environment parameters directly affect the power generation efficiency of photovoltaic modules, and their collection frequency is usually once every 5 minutes to ensure the sensitive response of the system to environmental changes. The voltage, current and power and other equipment operation parameters are acquired in real time by the SCADA system or the local controller with millisecond-level precision to evaluate the instant operation state of the equipment.

[0028] In terms of technical implementation, the data acquisition module adopts an industrial-grade sensor combined with an edge gateway, supports standard communication protocols such as Modbus, CAN, RS485, and ensures compatibility with photovoltaic equipment of different brands and models. The raw data collected is first preprocessed at the edge node, including outlier detection and elimination (such as triggering an abnormal flag when the voltage mutation exceeds 15% of the rated value), missing value filling (using sliding window mean method or interpolation algorithm), standardization processing (such as normalizing the light intensity to the range of [0, 1000]), data encryption (such as AES-256) and compression (such as GZIP), to reduce transmission bandwidth occupation and improve data security.

[0029] This step plays a key role in the system, and the high-quality data collected provides an input basis for subsequent dynamic weight calculation (such as formula ), directly affecting the priority ordering and abnormality detection accuracy of the tree chart nodes. At the same time, the acquisition and processing capability of real-time data determines whether the edge node can complete preliminary fault judgment within , so as to realize low-latency response and improve the overall intelligent level and operation and maintenance efficiency of the system.

[0030] Further, S1 comprises: S11, collecting voltage, current, power, temperature, irradiance and equipment state data through multi-source sensor fusion.

[0031] Specifically, in one embodiment of the present application, "collecting voltage, current, power, temperature, irradiance and equipment state data through multi-source sensor fusion" is a basic step for implementing data-driven decision-making of the intelligent monitoring system. This step realizes comprehensive perception and data fusion of the power station operation state through multi-type sensors deployed at key nodes of the photovoltaic power station and the real-time processing capability of the edge computing node.

[0032] In terms of technical implementation, the data acquisition module adopts a distributed deployment method, covering core devices such as photovoltaic modules, combiner boxes, inverters, transformers, and grid-connected cabinets. Sensor types include voltage sensors (such as Hall effect sensors), current sensors (such as shunts or current transformers), power sensors (calculated by the product of voltage and current), temperature sensors (such as PT100 or NTC thermistors), irradiance sensors (such as silicon-based photovoltaic sensors or thermoelectric pile sensors), and equipment state acquisition modules (such as PLC or SCADA systems). The acquisition frequency is set according to the characteristics of the device, for example, the voltage and current data acquisition frequency is 100 ms, the irradiance and temperature data is once every 5 minutes, and the equipment state information is uploaded in real time through event-driven mode.

[0033] In terms of parameter indicators, the voltage acquisition accuracy is ±0.5%, the current accuracy is ±1%, the power calculation error is controlled within ±2%; the temperature sensor accuracy is ±0.5°C, and the irradiance sensor measurement error under standard light intensity does not exceed ±5%. The collected data needs to meet the requirements of IEC61724 standard for photovoltaic system monitoring data, and interact with the grid-connected system through IEEE1547 standard.

[0034] In application scenarios, this step is widely used in large ground photovoltaic power stations, distributed rooftop photovoltaic systems, and remote off-grid power stations. Through multi-source data fusion, the system can identify typical faults such as inverter overload, module hot spot, and combiner box mismatch in real time, providing high-quality input for the construction of subsequent dynamic weight tree diagrams and edge-cloud collaborative analysis.

[0035] The technical effect of this step is that through multi-dimensional data acquisition and fusion, the perception accuracy and comprehensiveness of the system for the operation state of the power station are improved, providing a reliable data foundation for subsequent fault detection, performance evaluation, and intelligent decision-making, significantly reducing the false alarm rate and improving the response efficiency.

[0036] S12, clean, standardize and normalize the collected data, and cache the encrypted and compressed data to the edge node.

[0037] Specifically, this step involves cleaning, standardizing and normalizing the collected data, and encrypting and compressing the data after processing, and finally caching to the edge node. This step is a key link of data preprocessing and edge computing cooperation in the whole photovoltaic power station intelligent monitoring system, and its technical implementation directly affects the accuracy and efficiency of subsequent fault detection, model training and visualization analysis.

[0038] In an embodiment of the present application, the data cleaning process adopts an outlier rejection method based on statistical threshold, combined with time series sliding window detection (such as window size minutes), to identify and eliminate abnormal data points outside the range . At the same time, for missing values, the system adopts linear interpolation or collaborative filling strategy based on adjacent nodes to ensure data integrity. The standardization process adopts Z-score standardization method, i.e. , where is the mean, is the standard deviation, and the original data is mapped to the standard normal distribution, which is convenient for subsequent model processing. The normalization process adopts Min-Max normalization method, which linearly maps the data to the interval [0, 1], and the formula is , which is suitable for parameters such as light intensity and temperature that change nonlinearly.

[0039] Further, to ensure the secure transmission of data between the edge node and the cloud platform, the system uses the AES-256 encryption algorithm to encrypt the processed data, with a key length of 256 bits and a CBC (Cipher Block Chaining) encryption mode, to ensure the confidentiality and integrity of the data during transmission. The encrypted data is compressed by GZIP or LZ4 compression algorithm, and the compression ratio can usually reach , significantly reducing the transmission bandwidth occupancy.

[0040] Optionally, the edge node uses Raspberry Pi CM4 or industrial-grade embedded devices with built-in SSD storage modules to support local caching of processed data, with a cache capacity configurable to GB, and the cache strategy adopts LRU (Least Recently Used) mechanism to ensure the persistent storage of high-priority data. This step provides a high-quality, low-latency, secure and reliable data foundation for subsequent real-time analysis of the edge node and collaborative processing of the cloud platform, which is an important prerequisite for realizing the intelligent monitoring closed loop.

[0041] Step S2: Based on the environmental parameters and equipment operating status data, calculate the dynamic weight of each monitoring indicator. The dynamic weight is adjusted in real time according to the degree of light intensity deviating from the benchmark value, the nonlinear effect of temperature deviating from the benchmark value, and the health status of the equipment.

[0042] Specifically, the dynamic weight calculation step of this invention is based on environmental parameters and equipment operating status data, and uses a multi-dimensional fusion algorithm to adjust each monitoring indicator in real time, thereby improving the adaptability and accuracy of the photovoltaic power station intelligent monitoring system. The technical implementation principle of this step mainly relies on the quantitative modeling of light intensity, temperature changes, and equipment health status, and dynamically reflects the operating risks of the equipment under different environmental conditions through weighted fusion.

[0043] The specific operation method is as follows: First, the system collects real-time operating data of the photovoltaic modules, including light intensity. (unit: ),temperature (unit: and health status (Value range [0,1]), where It is calculated by the ratio of the current power generation efficiency to the rated power generation efficiency. For example, if the current efficiency is 80% of the rated value, then... Subsequently, the system uses a preset dynamic weighting formula. The weights of each component are calculated in real time. In this formula, Indicates the first The dynamic weight value of each component ranges from [0,1]. A higher weight indicates a better health status of the device.

[0044] At the parameter level, the baseline value for light intensity is set as follows: ,when When the value deviates from this value, the weight will be adjusted proportionally, for example, during rainy weather. Then the illumination weighting is reduced to This triggers a low light warning. The temperature component utilizes a non-linear influence function. ,when When the temperature deviates from the baseline value, the weight of this part is 1, indicating no abnormal impact; when the temperature deviates from the baseline value, the weight changes in the opposite direction to reflect the potential impact of temperature on power generation efficiency.

[0045] In practical applications, this step is deployed on edge computing nodes (such as Raspberry Pi CM4), updating the weight values ​​every 5 minutes and synchronizing the results to a dynamic weighted tree diagram module for node priority ranking and anomaly detection. This mechanism effectively reduces the false alarm rate of traditional fixed-weight models when the environment changes, and improves the system's responsiveness to complex operating conditions.

[0046] From the technical effect level, this step realizes fine evaluation of the operating state of the photovoltaic assembly by introducing a dynamic weight mechanism, enhances the fault identification capability of the system in scenes such as illumination fluctuation and temperature anomaly. At the same time, this method supports linkage with the edge-cloud coordination module, provides reliable data support for subsequent intelligent decision and remote control, and has significant engineering practical value.

[0047] Further, S2 comprises: S21, adopting the formula to calculate the dynamic weight value of the i th photovoltaic assembly, wherein is the state health degree, is the real-time temperature, is the real-time illumination intensity.

[0048] Specifically, in the dynamic weight calibration process of the application, the health state of the i th photovoltaic assembly is quantitatively evaluated by the formula , so as to realize optimization of dynamic priority sorting and anomaly detection of the tree diagram node. The formula fuses three key performance indicators: state health degree (SOCi), real-time temperature (T), and real-time illumination intensity (I), and gives different weight coefficients to reflect the relative influence of the three indicators on the health state of the assembly.

[0049] In terms of technical implementation, SOCi represents the ratio of the current power generation efficiency of the i th photovoltaic assembly to the rated power generation efficiency, and its value range is [0, 1]. For example, if the current power generation efficiency of the assembly is 80% of the rated value, then = 0.8. This parameter is calculated in real time by the voltage, current, and power data collected by the edge computing node, and has high time resolution and accuracy. The temperature term is used to reflect the degree of deviation of the operating temperature of the assembly from the standard temperature (25°C). The closer the temperature is to the standard value, the higher the weight, and vice versa. This design conforms to the physical characteristics that the performance of the photovoltaic assembly is optimal at the standard temperature, and helps to identify the efficiency decline or potential failure caused by temperature anomaly. The illumination intensity term takes the standard illumination intensity (1000W / m²) as the benchmark to respond sensitively to insufficient illumination. When = 200W / m², the weight of this part decreases to 0.04, and the system can trigger a low-illumination warning mechanism accordingly.

[0050] In terms of parameter setting, the coefficients 0.6, 0.2, and 0.2 in the formula correspond to ​​​​​The contribution ratios of temperature and light intensity to the weights are derived based on the performance sensitivity analysis of photovoltaic modules under different environments, and are supported by engineering experience. Furthermore, the denominator of the temperature term uses absolute value calculations to ensure symmetrical weight changes regardless of whether the temperature is above or below 25°C, enhancing the robustness of the model.

[0051] This step plays a crucial role in the edge-cloud collaborative system. Its output, Wi, serves as the dynamic weight of the tree diagram nodes, directly influencing the priority ranking of fault detection and the sensitivity of anomaly response. By periodically updating the weight values, the system can adapt to environmental changes in real time, improving monitoring accuracy and response efficiency, making it particularly suitable for outdoor photovoltaic power station scenarios with large fluctuations in sunlight and temperature.

[0052] S22, through the exponential decay function To achieve real-time adjustment of weights, where The attenuation coefficient is... For the first The degree of abnormality of each indicator.

[0053] Specifically, in one embodiment of the present invention, through an exponential decay function Real-time adjustment of weights is one of the core technical means of the dynamic weighted tree diagram construction module. This step dynamically corrects the weights of each monitoring indicator based on real-time feedback of the photovoltaic power plant's operating status, thereby improving the system's response accuracy and adaptability to abnormal events.

[0054] At the technical implementation level, this exponential decay function is used to quantize the first... Each indicator at the current moment The weight changes. Among them, This indicates the weight value of the indicator at the previous time step. This is a preset attenuation coefficient used to control the sensitivity of weight adjustment; its typical value range is [value range missing]. This is to ensure that the weight changes are smooth and do not oscillate excessively. Indicates the first The degree of anomaly for each indicator can be calculated based on the deviation between the historical mean and the current value, the number of threshold violations, or quantified through anomaly detection algorithms (such as Isolation Forest or LSTM prediction residuals). The higher the degree of anomaly, the faster the weight decays, thus reducing the priority of that indicator in the dendrogram and preventing it from interfering with the overall state assessment.

[0055] Regarding parameter settings, the system sets parameters during the initialization phase based on expert experience or historical data analysis. The initial value, for example This indicates that each unit of anomaly will result in a weight decrease of approximately 5%. Furthermore, The calculation needs to be combined with the physical meaning of the specific index. For example, for the voltage fluctuation index, the relative change rate of the voltage fluctuation index can be used as the quantitative basis for the degree of abnormality.

[0056] This step is periodically performed in the edge computing node, usually synchronized with the data collection frequency, such as updating the weight every 5 minutes. The updated weight will be used for the priority sorting of each node in the tree diagram, affecting the subsequent abnormality detection and fault diagnosis logic. For example, in high temperature weather, if the temperature of a certain inverter deviates from the standard value , its weight will be dynamically reduced, so that it is marked as a high-risk node in the tree diagram, improving the response efficiency of the operation and maintenance personnel.

[0057] In summary, the introduction of the exponential decay function makes the system have the ability to adapt to the environment and the running state, effectively reduces the false positive rate, and improves the real-time and accuracy of the intelligent monitoring system of the photovoltaic power station. It is a key technical support for realizing dynamic weight adjustment and fault priority sorting.

[0058] Step S3, constructing a multi-level tree diagram using the dynamic weight, the upper nodes of the tree diagram representing the overall operation state of the power station, and the lower nodes representing the state evaluation results of each subsystem, and updating the node priority and hierarchical structure of the tree diagram according to the dynamic weight.

[0059] Specifically, the step of constructing a multi-level tree diagram using a dynamic weight is one of the core logics of the intelligent monitoring system of the present application, and its technical implementation is based on a weight adjustment mechanism driven by real-time data and a dynamic evolution strategy of tree structure. This step realizes the hierarchical evaluation and priority sorting of the overall and subsystem state of the power station by abstracting the running state of the photovoltaic power station into the node structure of the tree diagram.

[0060] From the technical implementation point of view, the upper nodes of the tree diagram represent the overall operation state of the power station, and the weight is calculated by the weighted average of each subsystem node. The lower nodes correspond to the subsystems such as photovoltaic modules, combiner boxes and inverters, and the weight of each node is calculated by the dynamic weight formula , wherein represents the state health degree of the th component, is the real-time temperature, is the real-time light intensity. This formula takes into account the power generation efficiency, temperature deviation and light condition, ensuring that the weight can reflect the actual running state of the equipment under different environments.

[0061] At the parameter index level, the value range of the power generation efficiency is [0, 1], indicating the ratio of the current power generation efficiency of the component to the rated efficiency; the temperature term ​​The degree of temperature deviation from the standard value (25°C) is quantified, and the higher the temperature, the lower the weight; the light term Then, taking the standard light intensity as the reference, the weight significantly decreases when the light is insufficient, thereby triggering the early warning mechanism. The dynamic weight update frequency is once every 5 minutes, ensuring that the tree structure can respond to changes in the operation state of the power station in real time. In the application scenario, this step is mainly used for the local state evaluation of the edge computing node and the display logic of the visualization monitoring module. The edge node calculates the dynamic weight of each subsystem according to the collected voltage, current, temperature, irradiance and other data, and updates the node priority and hierarchical structure of the tree diagram accordingly. The cloud platform further optimizes the global structure of the tree diagram by receiving the weight data uploaded by the edge node, and realizes multi-dimensional and hierarchical monitoring of the operation state of the power station.

[0062] The technical effect of this step is that, through the dynamic weight mechanism, the system can automatically identify high-risk equipment and adjust its priority in the tree diagram, thereby improving the accuracy and response efficiency of fault detection. At the same time, the hierarchical structure of the tree diagram can be adaptively adjusted according to the weight change, supporting the operation and maintenance personnel to quickly locate the abnormal area, and significantly improving the intelligent operation and maintenance level of the photovoltaic power station.

[0063] Step S4, the edge computing node performs preliminary processing on the abnormal data, encapsulates the abnormal data meeting the preset threshold, and uploads it to the cloud platform, and adjusts the local control strategy according to the dynamic weight.

[0064] Specifically, the edge computing node performs preliminary processing on the abnormal data, encapsulates the abnormal data meeting the preset threshold, and uploads it to the cloud platform, and adjusts the local control strategy according to the dynamic weight, which is a key technical step in the present application to realize low-latency response and intelligent decision-making. Based on the cooperative mechanism of the edge computing architecture and the dynamic weight model, the dependence on the cloud platform is effectively reduced, and the adaptability and stability of the system in complex environments are improved.

[0065] From the technical implementation point of view, the edge computing node is usually deployed locally in the photovoltaic power station, uses an industrial-grade embedded device (such as Raspberry Pi CM4) as the hardware platform, and internally embeds a lightweight model (such as ELC-RNN) for real-time data processing. When the sensors collect voltage, current, power, temperature, irradiance and other operating parameters, the edge node first performs data cleaning and standardization processing to remove outliers and fill in missing values. Then, the system identifies potential abnormal events through a preset abnormality detection threshold (such as a current fluctuation of more than 15%). Once an anomaly is detected, the system encapsulates the data in JSON format and uploads it to the cloud platform through the MQTT protocol, ensuring lightweight and low-latency data transmission.

[0066] ​

[0067] In terms of dynamic weight adjustment, the system uses the following formula for weight calculation:

[0068] wherein, represents the dynamic weight value of the th component, is the state health degree, is the real-time temperature (unit: ), is the real-time light intensity (unit: ). The formula comprehensively considers the health status of the component, the influence of temperature change on efficiency, and the fluctuation of light intensity, thereby dynamically adjusting the priority of each node in the tree diagram and optimizing the response logic of the local control strategy.

[0069] In application scenarios, this step is widely applicable to real-time monitoring and fault response scenarios of distributed photovoltaic power stations, especially in remote areas with limited network bandwidth or variable environmental conditions. Through the local processing capability of the edge node, the system can complete abnormality identification and preliminary response within , significantly improving the operation reliability and fault handling efficiency of the power station.

[0070] The technical effect of this step is that through the cooperative processing of the dynamic weight mechanism and edge computing, the system can achieve rapid response and accurate identification of abnormal events, while reducing unnecessary data upload and reducing communication overhead. In addition, real-time adjustment of the local control strategy helps to maintain the basic operation safety of the power station when the cloud platform has not yet intervened, forming an intelligent monitoring closed loop of edge-cloud collaboration.

[0071] Further, S4 includes: S41, using a Raspberry Pi CM4 device to run a lightweight ELC-RNN model to collect device operation data at an interval of 100ms.

[0072] Specifically, the edge computing node (Raspberry Pi CM4) runs a lightweight ELC-RNN model and collects device operation data at a sampling interval of 100ms, which is a key technical link to realize the low-latency and high-real-time monitoring function of the present application. This step uses an embedded edge computing architecture and combines a lightweight neural network model to ensure efficient execution of data collection and preliminary analysis tasks on a resource-limited hardware platform.

[0073] The Raspberry Pi CM4, as an edge computing node, has a hardware configuration including an ARM Cortex-A72 quad-core processor, 1GB / 2GB / 4GB memory options, an Ethernet interface supporting PoE power supply, and an expandable GPIO interface. By deploying a lightweight ELC-RNN (Enhanced Lightweight RNN) model, the device realizes real-time data collection and processing of key equipment such as photovoltaic modules, combiner boxes, and inverters. The ELC-RNN model uses a gating mechanism and parameter sharing strategy in design, with model parameter quantity controlled within 1MB and inference delay less than 50ms, meeting the real-time requirements under a 100ms sampling interval.

[0074] In terms of specific operation, the edge node triggers data collection tasks at fixed time intervals (100ms) through timers or interrupt mechanisms. The collected data includes voltage, current, power, temperature, irradiance, and other operating parameters, as well as device status codes and timestamps. After collection, the data is sent to the ELC-RNN model for time series analysis, and the model outputs the current device's operating status evaluation results. When detecting that the current fluctuation exceeds 15%, the system will trigger an abnormal response mechanism and package the abnormal data into JSON format, and upload it to the cloud platform through the MQTT protocol.

[0075] At the parameter index level, the sampling interval is set to 100ms to ensure rapid response to device status changes. The input dimension of the ELC-RNN model is 12, including 6 real-time operating parameters and 6 environmental parameters, and the output dimension is 1, representing the device's health status score. The model inference frequency is 10Hz, meeting the real-time monitoring requirements. The communication protocol uses MQTT, with QoS level set to 2 to ensure reliable data transmission.

[0076] At the application scenario level, this step is suitable for edge deployment in photovoltaic power stations, especially in remote areas or scenarios with limited network bandwidth. Through local real-time processing and anomaly detection, it effectively reduces dependence on the cloud, improves system response speed and stability. At the same time, this step provides a data basis for subsequent dynamic weight calculation and tree diagram updating, and is a core link to realize edge-cloud collaborative intelligent monitoring.

[0077] In terms of technical effects, this step realizes millisecond-level response to the operating status of photovoltaic power stations through lightweight models and high-frequency sampling mechanisms, significantly improving the timeliness and accuracy of fault detection, and providing reliable data support for subsequent dynamic weight adjustment and remote control.

[0078] S42, when detecting that the current fluctuation exceeds 15%, package the abnormal data into JSON format and send it to the cloud platform through the MQTT protocol.

[0079] Specifically, when the edge computing node detects that the current fluctuation exceeds 15%, the system will start the abnormal data packaging and transmission mechanism to ensure that critical abnormal information can be quickly and reliably uploaded to the cloud platform, providing data support for subsequent fault diagnosis and remote control. The technical implementation of this step is based on the coordinated cooperation of real-time monitoring, data structured packaging, and lightweight communication protocols.

[0080] From the technical implementation level, the edge node collects current data of photovoltaic components in real time through deployed sensors or SCADA systems, and compares it with the reference value at the previous time. If the relative deviation of the current value from the reference value exceeds 15%, i.e. the condition is met, the abnormal response process is triggered. At this time, the system packages abnormal data (including timestamp, device ID, current value, deviation percentage, geographic location, etc.) into JSON format to organize data content in a structured manner, facilitating cloud platform analysis and processing. The JSON data structure usually includes fields such as device_id, timestamp, current_value, deviation_percent, location, etc., ensuring data integrity and traceability.

[0081] At the parameter index level, the current fluctuation threshold is set to 15%, which is based on the normal fluctuation range of photovoltaic systems under standard light and temperature conditions, combined with historical operation data of the power station for optimized configuration. When packaging data, the edge node uses compression algorithms (such as GZIP) to compress JSON data to reduce transmission bandwidth occupancy. At the same time, to ensure data security, the packaging process can use AES-128 or RSA encryption algorithms to encrypt the data, ensuring that the data is not tampered with or leaked during MQTT transmission.

[0082] In terms of application scenarios, this step is widely used in real-time monitoring systems of photovoltaic power stations, especially in scenarios where edge nodes and cloud platforms operate cooperatively. For example, when current abnormalities occur in key devices such as inverters, combiner boxes, or grid-connected cabinets, the edge node can quickly judge and package data locally and upload it to the cloud platform through the MQTT protocol, thereby realizing remote alarm and fault diagnosis. This mechanism is particularly suitable for photovoltaic power stations in areas with limited network bandwidth or remote areas, effectively reducing unnecessary full data upload and improving system response efficiency.

[0083] At the technical effect level, this step realizes the rapid identification and efficient transmission of abnormal events by setting clear abnormal threshold and structured data packaging. Combined with the low delay and lightweight characteristics of the MQTT protocol, it ensures real-time synchronization of data between the edge and the cloud platform, providing a reliable data foundation for subsequent intelligent analysis and decision-making. Further, this mechanism reduces the computational burden of the cloud platform, improving the overall energy efficiency and stability of the system, and is one of the key links to realize edge-cloud collaborative intelligent monitoring.

[0084] Step S5, the cloud platform performs deep fault diagnosis based on the received abnormal data, generates optimized model parameters and issues them to the edge computing node to realize iterative update of the dynamic weight calculation model.

[0085] Specifically, the cloud platform performs deep fault diagnosis based on the received abnormal data and generates optimized model parameters to issue to the edge computing node to realize iterative update of the dynamic weight calculation model. This step is the core link of the edge-cloud collaborative mechanism in the present application, and its technical implementation is based on the joint optimization strategy of distributed computing architecture and machine learning model.

[0086] From the technical implementation level, the cloud platform first receives abnormal data packets from the edge node through a message queue (such as the MQTT or CoAP protocol), which are packaged in JSON format and contain time stamps, device IDs, voltages, currents, powers, temperatures, irradiance, and other key parameters. The deep learning model (such as LSTM or Transformer) built-in the cloud platform extracts features and identifies patterns from the received abnormal data, determining whether it belongs to a known fault type or a potential abnormal trend. During model diagnosis, the system uses a multi-layer perceptron (MLP) to perform nonlinear mapping on the data, combining historical data and current operating status to generate a new set of model parameters.

[0087] At the parameter index level, the cloud platform performs model training once a week, with a training period of 7 days and a training data set containing all abnormal data samples uploaded by edge nodes in the past week. The objective function of model optimization is to minimize fault classification error, and its update formula can be expressed as:

[0088] wherein, represents the weight value of the th monitoring index at time , is the decay coefficient (usually set to ), represents the abnormality degree of the th index. The optimized model parameters are issued to the edge node through the HTTP protocol to realize hot update of the model without the need to shut down to complete model iteration.

[0089] At the application scenario level, this step is widely used in the remote operation and maintenance of photovoltaic power stations, especially in scenarios where network bandwidth is limited or environmental changes are frequent, such as mountainous areas, coastal areas, or desert areas. The cloud platform continuously optimizes model parameters, enabling edge nodes to more accurately identify device anomalies, thereby improving fault response speed and diagnostic accuracy.

[0090] The technical effect of this step is that the centralized deep learning model of the cloud continuously optimizes the lightweight model of the edge node, significantly improving the fault detection capability of the edge side. At the same time, the dynamic updating mechanism of the model parameters effectively adapts to the changes in the operation environment of the power station, reduces the false alarm rate, and improves the intelligent level and operation stability of the system.

[0091] Further, S5 includes: S51, model training is performed once a week, and a deep learning model (such as LSTM, Transformer) is used to optimize model parameters.

[0092] Specifically, performing model training once a week is a key step in the intelligent monitoring system of the present application to achieve continuous optimization and adaptive capability. This step deploys a deep learning model (such as LSTM, Transformer) on the cloud platform, uses real-time and historical data uploaded by the edge node to iteratively update the model parameters, thereby improving the accuracy of fault diagnosis and the intelligent level of the system.

[0093] From the technical implementation perspective, the cloud platform receives structured data packets from the edge node, usually encapsulated in JSON format, containing time stamps, device status, voltage, current, power, temperature, irradiance, and other key parameters. After preprocessing, these data are input into the deep learning model for training. LSTM (Long Short-Term Memory Network) is suitable for processing time series data and can capture the long-term dependence of photovoltaic device operating states. The Transformer model uses a self-attention mechanism to efficiently model multi-dimensional features, especially suitable for large-scale parallel data processing. During model training, cross-entropy loss function or mean square error (MSE) is used as the optimization objective, and the model weights are updated through the backpropagation algorithm to minimize the prediction error.

[0094] At the parameter index level, the model training cycle is set to once a week to ensure that the system can adapt to the slow changes in the power plant operating environment while maintaining low computational load. The training data set usually contains at least 7 days of operating data, with a sampling frequency of every 100 ms, and the data volume is about 10,000-50,000 per device. The model input dimensions include time series features (such as voltage, current, temperature) and static features (such as geographic location, device model), and the output is the fault category or health status prediction value of the device. During the training process, the learning rate is usually set to , the batch size is 64, and the training rounds (epoch) are 10-20, which are adjusted according to the data volume and model convergence.

[0095] At the application scenario level, this step is mainly applied to the model training module of the cloud platform, supporting remote updating of lightweight models (such as ELC-RNN) deployed in edge nodes. Through weekly model training, the system can adapt to seasonal changes in light, device aging trends and other long-term factors, thereby improving the fault detection accuracy and response efficiency of edge nodes. In addition, the trained model can also be used to generate operation and maintenance suggestions, such as cleaning photovoltaic panels and replacing inverters, to further optimize the operating efficiency of the power plant.

[0096] The technical effect of this step is that through periodic model training, the system can continuously learn new fault patterns and operating characteristics, realizing dynamic evolution of the model. Combined with the real-time data processing capability of edge nodes, a collaborative mechanism of "edge rapid response + cloud deep learning" is formed, significantly reducing the false alarm rate and improving the diagnostic accuracy, providing solid technical support for the intelligent operation and maintenance of photovoltaic power plants.

[0097] S52, the optimized model version is distributed to each edge computing node through the HTTP protocol to realize hot updating of the model.

[0098] Specifically, the optimized model version is distributed to each edge computing node through the HTTP protocol to realize hot updating of the model, which is one of the key operation steps in the edge-cloud collaborative module of the present application. This step aims to ensure that the lightweight model (such as ELC-RNN) on the edge node can continuously receive the latest version from the cloud platform, thereby completing online updating of the model without interrupting the current monitoring task, improving the real-time response capability and fault recognition accuracy of the system to the operating state of the power plant.

[0099] From a technical implementation perspective, after completing model training and parameter optimization, the cloud platform packages the model file into a standardized model update package (MUP), usually in lightweight model formats such as onnx or tflite, to adapt to the computing power of edge devices. The model update package is transmitted through the HTTP protocol, and the HTTPS encryption channel is used to ensure data security. During transmission, the cloud platform uses the RESTful API interface to send model update requests to the edge node. After receiving the request, the edge node first verifies the difference between the model version number and the current running version. If there is an update requirement, it downloads the model file through the HTTP GET or POST method and stores it in the local SSD or Flash storage. Then, the edge node loads the new model into the memory through the model loader (ModelLoader), replacing the original model and implementing hot updates.

[0100] At the parameter index level, the size of the model update package is usually controlled within 5MB to adapt to the limited storage and bandwidth resources of the edge node. The update frequency can be configured daily, weekly, or automatically triggered according to model performance changes. It is recommended to update during off-peak hours (such as night) to reduce the impact on real-time monitoring. The model version identifier (model_version), hash check code (checksum), and update timestamp (timestamp) are included in the HTTP request header to ensure the integrity and consistency of the model file.

[0101] At the application scenario level, this step is widely used in distributed photovoltaic power station edge computing nodes, especially in remote areas or scenarios with limited network bandwidth. Through the reliable transmission mechanism of the HTTP protocol, the stability and security of model updates are ensured. After receiving the update, the edge node can immediately apply the new model for real-time data processing and fault detection without restarting the system, thereby realizing seamless switching and continuous monitoring.

[0102] The technical effect of this step is that through the model hot update mechanism, the system can quickly respond to model optimization results and improve the intelligent decision-making ability of the edge node. At the same time, combined with the dynamic weight tree diagram update strategy, the adaptability and accuracy of the model are further enhanced, providing solid technical support for the efficient operation and maintenance of photovoltaic power stations.

[0103] Step S6, start the digital twin interaction interface, load the 3D model of the photovoltaic power station and support gesture operation.

[0104] Specifically, in one embodiment of the present invention, maintenance personnel activate the digital twin interactive interface via a mobile terminal (such as a smartphone app or AR glasses) and load a three-dimensional (3D) model of the photovoltaic power station, while also supporting gesture operations, to achieve intuitive monitoring and interactive analysis of the power station's operating status. This step is one of the core functions of the visualization monitoring module in the overall system architecture and is a key link in realizing human-machine collaboration and improving maintenance efficiency.

[0105] At the technical implementation level, the digital twin interactive interface is built on the Unity3D engine and uses AR (Augmented Reality) technology to map the physical structure of the photovoltaic power station onto the user terminal as a virtual model. The 3D model loading process includes obtaining the power station topology data (such as device coordinates, connection relationships, and real-time status) from the cloud platform and rendering it in combination with locally cached model resources. Gesture operations are achieved through posture recognition and gesture capture using the terminal's built-in gyroscope, accelerometer, and camera, supporting interaction methods including rotation, scaling, panning, and clicking. Clicking can trigger a device information pop-up window, displaying its operating parameters and historical performance curves.

[0106] In terms of specifications, the 3D models supported by the system must meet the Level of Detail (LOD) standard to adapt to the hardware performance of different terminals. The gesture recognition module uses a real-time image processing algorithm based on OpenCV or MediaPipe, and the recognition accuracy must reach [specific accuracy level]. Response latency control within The performance degradation simulation curves are generated based on the device's historical data, with a time window of 48 hours and a sampling frequency of once every 5 minutes. The curve fitting uses linear regression or exponential smoothing methods.

[0107] In application scenarios, this step is suitable for routine inspections, fault location, and remote diagnostics of photovoltaic power plants. Maintenance personnel can visually view the power plant layout on-site or remotely via an AR interface, quickly locate malfunctioning equipment, and obtain detailed operational data through gesture controls, thereby improving response efficiency and decision-making accuracy.

[0108] The technical effect of this step is that by integrating digital twin and AR technologies, an immersive visualization of the photovoltaic power station's operating status is achieved, enhancing the intuitiveness and real-time nature of human-computer interaction. It provides operation and maintenance personnel with an efficient and intuitive monitoring method and is an important supporting module of the intelligent monitoring system of this invention.

[0109] Step S7: Through the multi-person collaborative annotation function, the annotation information of abnormal points is synchronized to the cloud platform in real time, and the playback of historical annotation records is supported.

[0110] Specifically, through the multi-person collaborative labeling function, the abnormal point labeling information is synchronized to the cloud platform in real time, and the historical labeling record playback is supported, which is the key step for the visual monitoring module and the digital twin interaction module to work together. Based on the edge-cloud collaborative architecture, combined with AR (Augmented Reality) technology and distributed data synchronization mechanism, this step realizes real-time labeling and information sharing of photovoltaic power station equipment abnormal points by multiple users in a unified digital twin interface.

[0111] From the technical implementation level, the operation and maintenance personnel can access the digital twin interface through the mobile phone App or AR glasses. The interface is built based on the Unity3D engine and supports gesture recognition and interactive operation. When the user identifies that there is an abnormality in a photovoltaic module or device (such as voltage fluctuation, temperature anomaly, etc.), the user can click on the corresponding position in the 3D model, and the system will automatically pop up a labeling interface, allowing the user to input the abnormal type, severity, note information, etc. The labeling information is packaged in JSON format, including device ID, timestamp, geographic location, user ID and labeling content, and is uploaded to the cloud platform in real time through the MQTT protocol. After receiving the labeling information, the cloud platform writes it into the distributed database (such as Cassandra) and synchronously updates it to the AR interface of all logged-in users, ensuring the consistency and real-time performance of the information.

[0112] At the parameter index level, the synchronization frequency of the labeling information can be configured to be once per second to ensure low-latency response. The size of the labeling data packet is controlled within 500 bytes to adapt to the bandwidth limitation between the edge node and the cloud platform. The system supports up to 10 users to label online at the same time, and each labeling record has a timestamp (format YYYY-MM-DDHH:MM:SS) for subsequent playback and analysis. Historical labeling records can be played back through the timeline, supporting frame-by-frame viewing, and users can select a specific time period or specific device for backtracking to assist fault diagnosis and responsibility tracing.

[0113] At the application scenario level, this step is widely used in the remote collaborative operation and maintenance scene of photovoltaic power stations. For example, in a multi-region distributed photovoltaic power station, operation and maintenance personnel at different locations can label the same power station model through the AR interface, realizing cross-regional collaboration. In addition, during the process of handling sudden failures, multiple experts can access the system at the same time to label and discuss the abnormal points, improving the response efficiency.

[0114] The technical effect of this step is that through the multi-person collaborative labeling and real-time synchronization mechanism, the accuracy and response speed of abnormal point identification are significantly improved. The historical labeling playback function provides reliable data support for fault review, operation and maintenance training, and responsibility analysis, enhancing the traceability and intelligence level of the system.

[0115] In order to realize the above-mentioned embodiment, the application further provides a photovoltaic power station intelligent monitoring device based on dynamic weight tree diagram and edge-cloud cooperation. Figure 2 A structural schematic diagram of a photovoltaic power station intelligent monitoring device based on dynamic weight tree diagram and edge-cloud cooperation provided by the embodiment of the application is shown in the figure. Figure 2 As shown in the figure, the device comprises: A data acquisition module 100 is configured to acquire real-time environmental parameters and equipment operation state data of the photovoltaic power station, wherein the environmental parameters include light intensity and temperature, and the equipment operation state data includes voltage, current, power and equipment state. A dynamic weight calculation module 200 is configured to calculate dynamic weights of each monitoring index based on the environmental parameters and the equipment operation state data, wherein the dynamic weights are adjusted in real time according to the degree of deviation of light intensity from a reference value, the nonlinear influence of temperature deviation from a reference value and the equipment health state. A tree diagram construction module 300 is configured to construct a multi-level tree diagram using the dynamic weights, wherein the upper nodes of the tree diagram represent the overall operation state of the power station, the lower nodes represent the state evaluation results of each subsystem, and the node priority and hierarchical structure of the tree diagram are updated according to the dynamic weights. An edge computing processing module 400 is configured to preliminarily process abnormal data through an edge computing node, encapsulate abnormal data meeting a preset threshold and upload the abnormal data to a cloud platform, and adjust a local control strategy according to the dynamic weights. A cloud platform diagnosis module 500 is configured to perform deep fault diagnosis based on the received abnormal data, generate optimized model parameters and distribute the model parameters to the edge computing node, and realize iterative update of the dynamic weight calculation model.

[0116] As to the device in the above-mentioned embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment related to the method, and will not be described in detail here.

[0117] In order to realize the above-mentioned embodiment, the application further provides an electronic device, comprising a processor and a memory connected with the processor in communication; the memory stores computer execution instructions; the processor executes the computer execution instructions stored in the memory to realize the method provided by the foregoing embodiment.

[0118] In order to realize the above-mentioned embodiment, the application further provides a computer readable storage medium, wherein the computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by a processor to realize the method provided by the foregoing embodiment.

[0119] To achieve the above-mentioned embodiments, the application further provides a computer program product comprising a computer program which, when executed by a processor, implements the method provided by the above-mentioned embodiments.

[0120] The collection, storage, use, processing, transmission, provision and disclosure of the user's personal information involved in the present application comply with the relevant laws and regulations and do not violate public order and good customs.

[0121] It should be noted that the personal information from the user should be collected for legal and reasonable purposes, and should not be shared or sold outside these legal uses. In addition, such collection / sharing should be carried out after the user's informed consent is received, including but not limited to informing the user to read the user agreement / user notice before the user uses the function, and signing the agreement / authorization including authorization of relevant user information. In addition, any necessary steps should be taken to protect and secure access to such personal information data and ensure that other people with access to personal information data comply with their privacy policies and processes.

[0122] The present application is expected to provide embodiments in which the user can selectively prevent the use or access of personal information data. That is, the present disclosure is expected to provide hardware and / or software to prevent or block access to such personal information data. Once the personal information data is no longer needed, the risk is minimized by limiting data collection and deleting data. In addition, such personal information is de-identified, if applicable, to protect the privacy of the user.

[0123] In the foregoing embodiment description, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the specification and the features of the different embodiments or examples, without contradiction.

[0124] In addition, the terms "first", "second" are only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is at least two, for example, two, three, etc., unless otherwise specifically limited.

[0125] Any processes or methods described in the flowcharts or otherwise described herein can be understood as representing modules, segments, or portions of code that include one or more executable instructions for implementing specific logic functions (or steps) and / or can be implemented entirely in hardware. The steps or functions of the preferred embodiments of the application can be embodied in software, firmware, or hardware, and can be embodied within an ASIC or other chipset components, which can be specifically adapted to carry out one or more of the processes or methods described herein. This process can be stored on one or more computer-readable media, such as RAM, ROM, PROM, and / or flash memory, which can be embedded in a computer system, ASIC, chipset, or other hardware, or which can be accessible as a virtual machine, virtual memory, software, or other virtualization techniques. Such computer-readable media can store the instructions necessary for carrying out the processes or methods described herein, and can include the software necessary to carry out the processes or methods described herein.

[0126] Logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be embodied in computer-readable instructions, statements, or a sequence of instructions, steps, or statements, that can be implemented manually, automatically, or a combination thereof, and can be executed in an order other than as shown or discussed, including substantially concurrently or in reverse order, understood by those with skill in the art having the benefit of this disclosure.

[0127] It should be understood that aspects of the application can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, various steps or methods can be implemented in software or firmware that is stored in memory and executed by a suitable instruction execution system. As with a hardware implementation or in another embodiment, any of the techniques known in the art of making or using integrated circuits, programmable logic arrays (PLAs), field programmable gate arrays (FPGAs), and other components can be used to implement the functions of the preferred embodiments of the application.

[0128] Those skilled in the art can understand that all or part of the steps of the method of the above-mentioned embodiments can be instructed by a program to complete the relevant hardware, and the program can be stored in a computer readable storage medium. When the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0129] In addition, each functional unit in each embodiment of the present application can be integrated into one processing module, or each unit can exist physically independently, or two or more units can be integrated into one module. The integrated module can be realized in the form of hardware or in the form of a software functional module. When the integrated module is realized in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer readable storage medium.

[0130] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it should be understood that the above-mentioned embodiments are exemplary and cannot be understood as limiting the present application. Those skilled in the art can make changes, modifications, replacements and variations to the above-mentioned embodiments within the scope of the present application.

[0131] It should be understood that various forms of flow shown above can be used to reorder, add or delete steps. For example, each step described in the present application can be executed in parallel, sequentially or in different order, as long as the desired results of the technical solutions of the present application can be achieved, which is not limited herein.

[0132] The above detailed description does not constitute a limitation on the scope of protection of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations and replacements can be made according to design requirements and other factors. Any modification, equivalent replacement and improvement within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A photovoltaic power station intelligent monitoring method based on dynamic weight tree diagram and edge-cloud collaboration, characterized in that, Comprise the following steps: S1, obtain the real-time environmental parameters and equipment operating state data of the photovoltaic power station, the environmental parameters include light intensity and temperature, and the equipment operating state data include voltage, current, power and equipment state; S2, based on the environmental parameters and equipment operating state data, calculate the dynamic weight of each monitoring index, the dynamic weight is adjusted in real time according to the degree of light intensity deviating from the reference value, the nonlinear influence of temperature deviating from the reference value and the equipment health state; S3, utilize the dynamic weight to build a multi-level tree diagram, the upper nodes of the tree diagram represent the overall operation state of the power station, the lower nodes represent the state evaluation results of each subsystem, and the node priority and hierarchical structure of the tree diagram are updated according to the dynamic weight; S4, the edge computing node preliminarily processes the abnormal data, encapsulates the abnormal data meeting the preset threshold and uploads it to the cloud platform, and adjusts the local control strategy according to the dynamic weight; S5, the cloud platform performs deep fault diagnosis based on the received abnormal data, generates optimized model parameters and sends them to the edge computing node, and realizes iterative update of the dynamic weight calculation model.

2. The method of claim 1, wherein, The obtained real-time environmental parameters and equipment operating state data of the photovoltaic power station also include: S11, collect voltage, current, power, temperature, irradiance and equipment state data through multi-source sensor fusion; S12, clean, standardize and normalize the collected data, and cache them to the edge node after encryption and compression.

3. The method of claim 1, wherein, Based on the environmental parameters and equipment operating state data, the dynamic weight of each monitoring index is calculated, which also includes: S21, the dynamic weight value of the i-th photovoltaic module is calculated by using the following formula: The formula is shown as follows: wherein is the state health, is the real-time temperature, is the real-time light intensity; S22, realize real-time adjustment of the weight through exponential decay function, and the exponential decay function is as follows: wherein is the attenuation coefficient, is the degree of abnormality of the th index.

4. The method of claim 1, wherein, The edge computing node preliminarily processes the abnormal data, which also includes: S41, use Raspberry Pi CM4 device to run lightweight ELC-RNN model to collect equipment operating data at 100ms interval; S42, when detecting that the current fluctuation exceeds 15%, encapsulate the abnormal data into JSON format and send it to the cloud platform through MQTT protocol.

5. The method of claim 1, wherein, The cloud platform performs deep fault diagnosis based on the received abnormal data, which also includes: S51, perform model training once a week, and optimize model parameters by using deep learning model; S52, send the optimized model version to each edge computing node through HTTP protocol to realize hot update of the model.

6. The method of claim 1, wherein, Also include: S6, start the digital twin interaction interface, load the 3D model of the photovoltaic power station and support gesture operation; S7, through the multi-person collaborative labeling function, synchronously label the abnormal point information to the cloud platform in real time, and support historical labeling record playback.

7. A photovoltaic power station intelligent monitoring device based on dynamic weight tree diagram and edge-cloud cooperation, characterized in that, Comprise: A data acquisition module for acquiring real-time environmental parameters and equipment operating state data of a photovoltaic power station, the environmental parameters including light intensity and temperature, and the equipment operating state data including voltage, current, power and equipment state; A dynamic weight calculation module is configured to calculate dynamic weights of each monitoring index based on the environmental parameters and the equipment operation state data, and the dynamic weights are adjusted in real time according to the degree of deviation of the light intensity from a reference value, the nonlinear influence of the temperature deviation from a reference value, and the equipment health state. A tree diagram construction module is configured to construct a multi-level tree diagram by using the dynamic weights, wherein an upper node of the tree diagram represents the overall operation state of the power station, a lower node represents the state evaluation result of each subsystem, and the node priority and the hierarchical structure of the tree diagram are updated according to the dynamic weights. An edge computing processing module is configured to preliminarily process the abnormal data by the edge computing node, encapsulate the abnormal data meeting a preset threshold, and upload the abnormal data to the cloud platform, and meanwhile, adjust the local control strategy according to the dynamic weights. A cloud platform diagnosis module is configured to perform deep fault diagnosis based on the received abnormal data, generate optimized model parameters, and deliver the optimized model parameters to the edge computing node, so as to realize iterative update of the dynamic weight calculation model.

8. An electronic device, comprising: The method comprises: a processor, and a memory connected with the processor in communication; the memory stores computer execution instructions; the processor executes the computer execution instructions stored in the memory to realize the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by the processor to realize the method according to any one of claims 1-6.

10. A computer program product, characterised in that, The computer program is executed by the processor to realize the method according to any one of claims 1-6.

Images