Intelligent monitoring early warning and operation and maintenance management system and method for photovoltaic system

Abstract

The invention discloses an intelligent monitoring, early warning and operation and maintenance management method for a photovoltaic system. The method comprises the following steps: S1, dynamic fusion and edge collaborative acquisition of multi-source heterogeneous data: constructing a full-dimensional data acquisition system covering a photovoltaic module, an inverter, a combiner box, a meteorological environment and a power grid side; a framework of'edge preprocessing + cloud fine fusion 'is adopted, edge nodes realize second-level data synchronization through a lightweight time sequence alignment algorithm, and sensor noise is eliminated by adopting an improved density-based abnormal value filtering algorithm; and the cloud end allocates dynamic weights to data of different frequencies and different types based on a multi-modal data fusion model of an attention mechanism, so that high-precision data fusion is realized. By constructing a full-dimensional data acquisition system and an innovative fusion framework, the problem of data fragmentation of traditional photovoltaic monitoring is solved, the data fusion precision is improved, and a high-quality data basis is provided for subsequent analysis and diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photovoltaic system operation and maintenance, in particular to a photovoltaic system intelligent monitoring and early warning and operation and maintenance management system and method. BACKGROUND

[0002] In the field of photovoltaic system operation and maintenance management, the traditional monitoring method has many drawbacks. On the one hand, the data collection dimension is single and scattered, only isolated data acquisition is carried out for photovoltaic components, inverters and other equipment, forming a "data fragmentation" problem, which cannot reflect the overall operation of the photovoltaic system, making it difficult for operation and maintenance personnel to find deep-seated potential faults and risks based on one-sided data.

[0003] On the other hand, the existing fault diagnosis and early warning mechanism lacks systematicness and accuracy. For component-level hidden abnormalities, such as local shading, hidden cracks and other conditions, traditional detection relying on single physical rules or simple machine learning models has a high misjudgment rate; in terms of device-level fault diagnosis, new power stations often lack sufficient local labeled data, making it difficult to improve fault diagnosis accuracy, and it is difficult to effectively share fault knowledge between different power stations; and at the system level, there is also a lack of means to comprehensively analyze the correlation of each device and link, thereby effectively warning and locating the root cause of systemic risks. In addition, the functions of the current photovoltaic operation and maintenance software are also limited, and cannot fully meet the diversified needs of intelligent operation and maintenance, such as deep mining and analysis of multi-dimensional operation data of photovoltaic systems, and dynamic development of adaptive strategies based on power grid load.

[0004] Therefore, we propose a photovoltaic system intelligent monitoring and early warning and operation and maintenance management system and method to solve the above problems. SUMMARY

[0005] The present application proposes the following technical solutions to address the problems in the prior art: The photovoltaic system intelligent monitoring and early warning and operation and maintenance management method comprises the following steps: S1: Dynamic fusion and edge collaborative collection of multi-source heterogeneous data: a full-dimensional data collection system covering photovoltaic components, inverters, combiner boxes, meteorological environments and power grid sides is constructed; a "edge preprocessing + cloud precise fusion" architecture is adopted, the edge node realizes second-level data synchronization through a lightweight time alignment algorithm, and an improved anomaly value filtering algorithm based on density is used to eliminate sensor noise; a multi-modal data fusion model based on attention mechanism is used in the cloud to assign dynamic weights to data of different frequencies and types, and realize high-precision data fusion; S2: Hierarchical anomaly detection and root cause intelligent diagnosis: S21: Component-level anomaly detection: A "physical rule + deep learning" dual-criterion model is used, a basic threshold is set based on the theoretical power calculation of the photovoltaic component I-V characteristic curve, and an improved CNN-LSTM model is used to identify hidden anomalies; S22: Device-level fault diagnosis: A "fault feature library + transfer learning" diagnostic framework is built for inverters and combiner boxes, an XGBoost classification model is pre-trained, and a federal learning mechanism is used to share fault knowledge across power stations; S23: System-level risk warning: A photovoltaic system topology correlation model is built based on a graph neural network, the correlation between power flow and current flow in the system is analyzed, systemic risks are identified, and the SHAP value explanation model output is used to locate the key nodes of risk propagation; S3: Intelligent operation and maintenance decision and multi-module collaborative application: The output results of step S2 are applied to: S31: Dynamic power generation prediction and accurate power generation calculation for driving wind farms and photovoltaic power stations; S32: Generate operation and maintenance work orders to guide multi-dimensional operation monitoring and abnormal warning intelligent operation and maintenance management of photovoltaic systems; S33: Based on the dynamic perception of power grid load, generate adaptive charging and discharging strategies to optimize the operation and revenue of energy storage systems; S34: Provide data support and decision basis for the whole process of new energy project quantity control and multi-module intelligent management; S35: Support the planning and design of zero-carbon microgrids in industrial parks, efficient use of clean energy, and operation management; S36: Provide data terminal support for the automatic management of photovoltaic supply chains, equipment health rating, and intelligent recommendation of spare parts.

[0006] A photovoltaic system intelligent monitoring and early warning and operation and maintenance management system for implementing the above method, comprising: a data acquisition and fusion module for executing step S1, including edge computing nodes deployed on site and cloud fusion servers located in a data center; An intelligent diagnosis and early warning module for executing step S2, including a component-level detection unit, a device-level diagnosis unit, and a system-level early warning unit; An operation and maintenance management and collaborative application module for executing step S3, integrated with a power generation prediction sub-module, an operation and maintenance management sub-module, a power grid interaction sub-module, an engineering management sub-module, a microgrid design and operation sub-module, and a supply chain management sub-module.

[0007] The beneficial effects of the present application are: By constructing a full-dimensional data acquisition system and an innovative fusion architecture, the data fragmentation problem of traditional photovoltaic monitoring is solved, the data fusion accuracy is improved, and a high-quality data foundation is provided for subsequent analysis and diagnosis.

[0008] The hierarchical anomaly detection and diagnosis model greatly improves the detection and identification capability of photovoltaic system anomalies and faults, from the component level, device level to system level, layer by layer. The high identification rate and low false alarm rate of the component level can avoid unnecessary manpower investigation. The device level diagnosis framework significantly improves the diagnosis efficiency of new power station. The system level model can control the risk from the whole and locate the key node, and ensure the stable operation of the photovoltaic system.

[0009] The intelligent operation and maintenance management software integrated with multiple functions meets the intelligent management needs of photovoltaic operation and maintenance and other aspects of the new energy field, helps to promote the operation and maintenance and management level of the whole new energy industry to the intelligent and efficient direction, and has good application prospect and economic benefit. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 The overall architecture schematic diagram provided by the embodiment of the application is shown. Figure 2 The multi-source data dynamic fusion process schematic diagram provided by the embodiment of the application is shown. Figure 3 The hierarchical anomaly detection and diagnosis model architecture diagram provided by the embodiment of the application is shown. Figure 4 The intelligent operation and maintenance decision and multi-module collaborative application schematic diagram provided by the embodiment of the application is shown. DETAILED DESCRIPTION

[0011] In order to make the purpose, technical scheme and advantages of the embodiment of the application more clear, the technical scheme of the application will be described clearly and completely in combination with the embodiments.

[0012] As shown in Figure 1 , Figure 2 , Figure 3 and Figure 4 , in the embodiment of the application: the photovoltaic system intelligent monitoring and early warning and operation and maintenance management system first deploys various sensors (current and voltage sensors, temperature sensors, irradiation instruments, anemometers, etc.) and edge collection nodes (built-in lightweight algorithms) on the photovoltaic power station site.

[0013] The edge node locally caches the high-frequency data collected by the components (1 second / time) and the weather (1 minute / time), and runs a lightweight time alignment algorithm (such as a simplified version of dynamic time warping DTW) to synchronize the multi-source data in a sliding window.

[0014] After synchronization, the improved DBSCAN algorithm (for example, introducing time dimension as part of the distance measurement) is used to cluster the data, identify and eliminate noise points generated due to instantaneous failure of sensors or environmental interference.

[0015] The preprocessed data is uploaded to the cloud together with low-frequency data (such as power grid data, 5 minutes / time).

[0016] The cloud-deployed multi-modal fusion model based on attention mechanism takes as input numerical time series data (current, voltage), image data (EL detection images), and state data (breaker opening and closing state).

[0017] The model automatically learns and assigns different data sources and different time points a contribution weight (i.e., attention score) to the final fused state through an attention layer, thereby achieving accurate fusion and outputting a high-confidence system state matrix.

[0018] The fused high-quality data is fed into a hierarchical diagnosis module; the component-level unit first calculates the theoretical I-V curve and maximum power point under the condition of real-time irradiance and temperature through physical formulas, and sets a reasonable fluctuation range as the physical threshold; at the same time, the historical and real-time sequences of component temperature and current are input into an improved CNN-LSTM model (CNN is used to extract local spatial features in current fluctuations, and LSTM is used to capture long-term time dependence), and the model outputs an anomaly probability. The final alarm is determined by both the physical threshold and the AI probability, and an alarm is confirmed only when both indicate an anomaly, greatly reducing false alarms.

[0019] The device-level unit uses historical fault data from multiple power stations to train an XGBoost classification model that can diagnose 12 typical faults (such as inverter overheating, IGBT open circuit, and busbar grounding fault) as a base model in the cloud; when deployed for a new power station (such as power station B), the base model is not directly used, but a federated learning mechanism is adopted: the local data of power station B is not exported, only the encrypted update gradient of the model parameters is uploaded to the cloud server, and the gradient is securely aggregated with other power stations to update the global model, and then the updated model is distributed to power station B; through such iteration, the diagnosis model of power station B quickly obtains the "experience" of other power stations under the premise of protecting data privacy, and the accuracy is rapidly improved.

[0020] The system-level unit regards the entire photovoltaic station as a graph structure, with nodes being components, strings, inverters, and transformers, and edges representing electrical connection relationships (power flow, current flow); a graph neural network (GNN) is used to learn the representation of nodes and edges, so that the propagation path of faults or anomalies in the system can be simulated; when a power anomaly occurs in an inverter, the GNN model can infer that the anomaly is caused by the shading of multiple components in a string connected below the inverter, and locate the most likely fault source string; through SHAP value analysis, the contribution of each component node to the current system abnormal state can be quantified, thereby intuitively displaying the "key nodes".

[0021] Finally, all diagnostic results, early warning information and health status reports will be pushed to the operation and maintenance and collaborative application module; this module is a comprehensive software platform that can call diagnostic results for power generation prediction, automatically generate operation and maintenance work orders and distribute them to mobile terminals to guide on-site defect elimination (intelligent operation and maintenance management), develop the optimal energy storage charging and discharging strategy according to the grid dispatching instructions and real-time electricity price, combined with the system state (revenue optimization), and provide data intelligent services for the design, construction, supply chain management and other whole life cycle links of the power station.

[0022] As Figure 1 , Figure 2 , Figure 3 and Figure 4 , the intelligent monitoring and early warning and operation and maintenance management method of the photovoltaic system has the following specific steps: A full-dimensional data acquisition system covering photovoltaic modules, inverters, combiner boxes, meteorological environments and grid-side parameters is constructed; The multi-source heterogeneous data collected by the full-dimensional data acquisition system is dynamically fused by using an edge preprocessing and cloud fine fusion architecture, wherein the edge preprocessing includes using a lightweight time alignment algorithm to realize data synchronization and using an improved density-based clustering algorithm for noise filtering, and the cloud fine fusion includes using a model based on attention mechanism to assign dynamic weights to multi-modal and multi-frequency data for fusion; Based on the fused data, hierarchical anomaly detection and root cause diagnosis are performed, including component-level anomaly detection based on physical rules and CNN-LSTM deep learning model, device-level fault diagnosis based on pre-trained fault feature library and transfer learning framework, and system-level risk warning based on graph neural network topology model and explainable artificial intelligence technology; The results of the hierarchical anomaly detection and root cause diagnosis are output to at least one application module, including: a power generation power dynamic prediction and power generation capacity measurement module, an intelligent operation and maintenance management module, an adaptive charging and discharging strategy adjustment and revenue optimization module, a new energy engineering whole process intelligent management module, a zero-carbon microgrid design and operation management module, and a supply chain automation management and intelligent rating recommendation module.

[0023] The above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them.

Claims

1. A method for intelligent monitoring, early warning, and operation and maintenance management of photovoltaic systems, characterized in that, include: Construct a comprehensive data acquisition system covering photovoltaic modules, inverters, combiner boxes, meteorological environment, and grid-side parameters; An architecture combining edge preprocessing and cloud-based fine fusion is adopted to dynamically fuse multi-source heterogeneous data collected by the full-dimensional data acquisition system. Edge preprocessing includes using a lightweight temporal alignment algorithm to achieve data synchronization and using an improved density-based clustering algorithm for noise filtering. Cloud-based fine fusion includes using an attention-based model to dynamically assign weights to multimodal and multi-frequency data for fusion. Based on the fused data, hierarchical anomaly detection and root cause diagnosis are performed, including component-level anomaly detection based on physical rules and CNN-LSTM deep learning models, equipment-level fault diagnosis based on pre-trained fault feature libraries and transfer learning frameworks, and system-level risk warning based on graph neural network topology models and interpretable artificial intelligence technologies. The results of the hierarchical anomaly detection and root cause diagnosis are output to at least one application module, which includes: a power generation dynamic prediction and power generation calculation module, an intelligent operation and maintenance management module, an adaptive charging and discharging strategy adjustment and revenue optimization module, a new energy project full-process intelligent management module, a zero-carbon microgrid design and operation management module, and a supply chain automation management and intelligent rating and recommendation module.

2. The intelligent monitoring, early warning, and operation and maintenance management method for photovoltaic systems according to claim 1, characterized in that, The lightweight temporal alignment algorithm is a sliding window dynamic matching algorithm, and the improved density-based clustering algorithm is the DBSCAN algorithm that introduces the time dimension as a distance metric.

3. The intelligent monitoring, early warning, and operation and maintenance management method for photovoltaic systems according to claim 1, characterized in that, The attention-based model can assign differentiated fusion weights to numerical, image, and state data, as well as data collected at different frequencies.

4. The intelligent monitoring, early warning, and operation and maintenance management method for photovoltaic systems according to claim 1, characterized in that, The physical rules in the component-level anomaly detection are theoretical power calculation rules based on the photovoltaic component IV characteristic curve and considering temperature and irradiance corrections. The input of the CNN-LSTM model includes component temperature and current fluctuation time series data.

5. The intelligent monitoring, early warning, and operation and maintenance management method for photovoltaic systems according to claim 1, characterized in that, The transfer learning framework in the equipment-level fault diagnosis is a federated learning mechanism, which is used to achieve collaborative training and knowledge sharing of fault diagnosis models across power plants.

6. The intelligent monitoring, early warning, and operation and maintenance management method for photovoltaic systems according to claim 1, characterized in that, The interpretable artificial intelligence technology in the system-level risk warning is SHAP value analysis, which is used to locate key nodes in the risk propagation path.

7. The intelligent monitoring, early warning, and operation and maintenance management system for photovoltaic systems according to any one of claims 1-6, characterized in that, include: The data acquisition and fusion module includes multiple edge computing nodes deployed on-site and a cloud fusion server, used to realize the dynamic fusion and collaborative acquisition of multi-source heterogeneous data; The intelligent diagnosis and early warning module includes a first unit for component-level anomaly detection, a second unit for equipment-level fault diagnosis, and a third unit for system-level risk early warning. The operation and maintenance management and collaborative application module integrates multiple sub-modules that communicate with the intelligent diagnosis and early warning module, and is used to receive diagnostic results and execute corresponding operation and maintenance decisions and collaborative applications.

8. The intelligent monitoring, early warning, and operation and maintenance management system for photovoltaic systems according to claim 7, characterized in that, The sub-modules of the operation and maintenance management and collaborative application module include: Sub-module for dynamic prediction and power generation calculation software for wind farms / photovoltaic power plants; Sub-module of intelligent operation and maintenance management software for multi-dimensional operation data monitoring and anomaly early warning of photovoltaic systems; A software submodule for adaptive charging and discharging strategy adjustment and revenue optimization based on dynamic perception of power grid load; Sub-module of software for full-process engineering quantity control and multi-module intelligent management of new energy projects; Sub-module of software for design, operation and management of zero-carbon microgrids and efficient utilization of clean energy in industrial parks; Supply chain automation management and intelligent rating and recommendation terminal sub-module.

Images