Regional distributed photovoltaic power station intelligent management method based on cloud end architecture

Abstract

The invention discloses an area-level distributed photovoltaic power station intelligent management method based on a cloud side end architecture, and the method comprises the steps: building a layered architecture: the layered architecture building comprises a sensing layer, a transmission layer, an edge calculation layer, a platform layer and an application layer, and relates to the technical field of energy management and information technology crossing. By deploying various high-precision sensors in the sensing layer, the equipment operation state and environmental factors are sensed comprehensively, and richer data can be obtained. In the aspect of equipment performance analysis, indexes such as average power generation efficiency and the like are calculated, multi-dimensional comparison is carried out by applying time sequence analysis and data normalization processing, and equipment performance abnormity can be found more timely and accurately.

Description

Technical Field

[0001] The present invention relates to the field of the intersection of energy management and information technology, with a particular focus on utilizing a cloud-edge integrated architecture and integrating advanced data processing, model building, and intelligent analysis technologies. Specifically, it is a method for intelligent management of regional distributed photovoltaic power stations based on a cloud-edge architecture. Background Art

[0002] Against the backdrop of the global push for a clean energy transition, distributed photovoltaic power generation, with its advantages of local consumption and flexible deployment, has become a key development direction in the energy sector. However, with the large-scale construction and integration of distributed photovoltaic projects, a series of serious problems have emerged.

[0003] From the perspective of grid structure adaptation, distributed PV systems generally have smaller capacities and are often connected to the distribution network, significantly different from the traditional grid architecture, which is primarily based on centralized power generation. Traditional grid-side monitoring and management systems are primarily designed for large, centralized power sources, making it difficult to effectively monitor and manage the numerous, widely distributed, and small-capacity distributed PV systems, resulting in a long-standing regulatory blind spot. This makes it difficult for grid operators to obtain timely, critical information such as the real-time operating status and power generation of distributed PV systems, hindering accurate power dispatch and load balancing.

[0004] As massive amounts of distributed photovoltaic power generation are integrated into the power grid, problems caused by a lack of monitoring and management continue to accumulate. The randomness, volatility, and uncertainty of distributed photovoltaic output power pose a significant threat to the safe and stable operation of the power grid. Weather changes, such as cloud cover and sudden changes in sunlight intensity, can cause distributed photovoltaic power generation to fluctuate significantly over short periods of time. Once connected to the grid, this unstable power output can cause voltage fluctuations and frequency deviations, impacting the normal operation of other equipment in the grid and even causing localized grid failures. Especially with the rapid advancement of large-scale projects such as county-wide photovoltaic projects, the proportion of distributed photovoltaic power generation in localized areas of the grid capacity has increased significantly, making power generation capacity prediction and global optimal output assessment extremely urgent. If the power generation capacity of distributed photovoltaic power generation cannot be accurately predicted, the grid may face power shortages or surpluses when allocating power, reducing the economic and reliability of grid operation.

[0005] From a monitoring technology perspective, existing monitoring equipment and technologies are insufficient to meet the monitoring needs of distributed photovoltaic systems. Traditional power monitoring equipment faces challenges with high installation costs and data transmission difficulties in the dispersed layout of distributed photovoltaic systems. Furthermore, distributed photovoltaic systems involve multiple devices and complex operating environments, requiring more sophisticated and intelligent monitoring technologies to comprehensively assess their operating status. However, current monitoring technologies still have significant shortcomings in fault diagnosis and performance evaluation of distributed photovoltaic modules, failing to promptly identify potential safety hazards and performance degradation.

[0006] At the policy and market levels, with increasing national support for clean energy, the distributed photovoltaic market is rapidly expanding. However, the lack of effective monitoring and management tools makes it difficult to accurately assess the actual benefits of distributed photovoltaic projects and their impact on the power grid, hindering the formulation of scientific and rational industrial policies and subsidy strategies. Furthermore, for investors and operators of distributed photovoltaic projects, the inability to implement efficient monitoring and management makes it difficult to ensure stable returns and long-term operation.

[0007] Currently, existing research on wide-area distributed photovoltaic power generation power prediction and global output characteristics is relatively lagging. On the one hand, the industry is rapidly developing, and the scale and complexity of research objects far exceed those of previous studies. Existing research is unable to cope with such a massive amount of distributed photovoltaics. On the other hand, the scale and capacity of distributed photovoltaics abroad lags significantly behind those in China. Research results lack empirical evidence and application in real-world scenarios in China, and are unable to effectively address the unique challenges facing the development of distributed photovoltaics in my country. Therefore, the development of a regional distributed photovoltaic cluster power station monitoring and management system suitable for my country's national conditions is urgent. Summary of the Invention

[0008] In response to the shortcomings of existing technologies, the present invention provides a regional distributed photovoltaic power station intelligent management method based on a cloud-edge architecture, which solves the existing cluster power station monitoring and management problems that cannot be faced with massive distributed photovoltaics.

[0009] To achieve the above objectives, the present invention is implemented through the following technical solutions: a regional distributed photovoltaic power station intelligent management method based on cloud-fog-edge architecture, the method steps include building a layered architecture:

[0010] The layered architecture includes: perception layer, transport layer, edge computing layer, platform layer and application layer;

[0011] The perception layer consists of densely deploying various types of high-precision sensors on and within each photovoltaic panel, inverter, and combiner box in a distributed photovoltaic power station, based on different monitoring requirements. In addition to conventional sensors for collecting basic data such as light intensity, temperature, voltage, and current, these sensors also include optical sensors for monitoring the cleanliness of photovoltaic panel surfaces, stress sensors for detecting stress changes in key inverter components, and sensors for monitoring meteorological parameters such as ambient humidity and wind speed. These sensors work together to comprehensively perceive the power station's operating status and environmental factors from multiple dimensions.

[0012] The transport layer uses software-defined networking (SDN) technology as its core architecture to build a highly flexible and intelligent transmission network. In the backbone network, high-fiber, high-bandwidth single-mode optical fibers are used to establish high-speed and stable data transmission channels. This ensures that data can be transmitted over long distances within the region at ultra-high speeds and low latency, meeting the needs of large-scale real-time data transmission.

[0013] The edge computing layer consists of deploying high-performance micro servers based on the advanced ARM architecture at edge nodes close to distributed photovoltaic power plants. These servers are compact, energy-efficient, and have powerful computing capabilities and local storage capabilities. They can cache raw data and preliminary processing results from the past 24 hours.

[0014] The platform layer builds an ultra-large-scale data lake based on big data technology, using an innovative storage architecture that combines a distributed file system with a columnar storage database to achieve efficient storage and rapid retrieval of massive power plant operation data. The data lake can store not only structured numerical data, but also unstructured text data and semi-structured JSON data. This comprehensive data storage method provides a rich data source for subsequent data mining and analysis.

[0015] The application layer involves developing diversified application services for different user groups. For power plant operators, a powerful intelligent operation and maintenance app integrates real-time equipment status monitoring, fault warning notifications, maintenance work order generation, and maintenance knowledge base querying. Furthermore, an intelligent operation and maintenance assistant is introduced. Leveraging natural language processing technology, operators can query equipment information and obtain maintenance recommendations through voice input, and even engage in conversations with the intelligent operation and maintenance assistant to quickly resolve issues encountered during operation and maintenance. Furthermore, the app features equipment inspection route planning. Based on the distribution and operating status of power plant equipment, a path planning algorithm is used to plan optimal inspection routes for operators, improving inspection efficiency. For managers, a decision support system is built, providing visual power plant operation reports, key performance indicator (KPI) analysis, and cost-benefit analysis. Through intuitive data visualization charts and intelligent analysis reports, managers are assisted in making scientific and rational strategic decisions. The system also incorporates intelligent forecasting and risk assessment capabilities, leveraging time series analysis and machine learning algorithms to predict key indicators such as future power generation, equipment failure rates, and cost expenditures.

[0016] Preferably, the perception layer also needs to introduce an intelligent sensor with advanced adaptive acquisition frequency adjustment function. The sensor has a built-in high-performance microprocessor and a data analysis algorithm based on deep learning. By analyzing the fluctuation of the collected data in real time, when the data fluctuation exceeds the preset dynamic threshold, it is judged that the operation status of the power station has changed drastically. The intelligent sensor automatically increases the acquisition frequency to a maximum of 10 times per second, so as to capture the equipment status changes more timely and accurately. When the operation status is stable, the acquisition frequency is reduced to 1 time per minute. While ensuring data integrity, it effectively reduces the data transmission volume and processing pressure, thereby reducing system energy consumption. When σ≤σ threshold When the lower acquisition frequency f low , the acquisition frequency formula can be expressed as:

[0017]

[0018] The degree of data fluctuation is measured by the standard deviation σ. When σ>σ threshold When the acquisition frequency f high .

[0019] Preferably, the transmission layer also needs to target distributed photovoltaic power stations in remote areas or with complex terrain and difficult wiring. The access layer adopts wireless communication technologies such as NB-IoT (narrowband Internet of Things) and 5G, and takes advantage of the low power consumption, wide coverage and low connection cost of NB-IoT technology to achieve stable connection and data collection for small distributed photovoltaic equipment; for scenarios with large data volumes and extremely high real-time requirements, such as real-time operation data transmission of inverters, the high speed and low latency characteristics of 5G networks are used to ensure that data can be transmitted quickly and accurately to the upper node; a powerful SDN controller is deployed to continuously monitor the key performance indicators of each transmission link in real time. The SDN controller automatically switches data traffic to the optimal backup link within milliseconds based on a preset complex link switching strategy. To ensure the security and confidentiality of data during transmission, the controller uses advanced quantum encryption algorithms to encrypt the transmitted data. The untapped and unbreakable nature of quantum key distribution ensures that data cannot be stolen or tampered with during transmission, providing a solid guarantee for secure data transmission in power plants. When selecting a link, the formula for calculating the comprehensive link score is as follows:

[0020]

[0021] Where B is bandwidth, D is delay, L is packet loss rate, and R is reliability. w1, w2, w3, and w4 are the weights of each factor, and w1+w2+w3+w4=1 selects the link with the highest score for data transmission.

[0022] Preferably, the edge computing layer also includes using a distributed hashing (DHT) algorithm to quickly classify and filter the massive amount of raw data collected by the perception layer, and map the data to different storage buckets based on multiple attributes such as data type, timestamp, and device identification. For real-time control data, such as the power adjustment instructions of the inverter, and important data closely related to the safe operation of the power station, such as abnormal current and voltage data, after preliminary analysis and processing by the edge computing device, they are preferentially sent to the platform layer through the transport layer; for general historical operation data, they are locally cached and uploaded to the platform layer in batches at a time period of once an hour, which greatly reduces the data transmission pressure of the core network and significantly improves the timeliness of data processing and the overall operation efficiency of the system. The location h of the data mapped to the storage bucket is calculated by the hash function H:

[0023] h=H(data_type, timestamp, device_id)

[0024] Among them, data_type is the data type, timestamp is the timestamp, and device_id is the device number.

[0025] Preferably, the platform layer also uses a deep learning algorithm based on convolutional neural network (CNN) for data fusion. Assuming that there are n different types of data, the feature vectors of each data extracted by CNN are F1, F2, ..., F n , the fused data feature vector F fusion It can be calculated by the following formula:

[0026]

[0027] Among them, wi is the weight of the i-th data feature vector, and The weights are continuously optimized through model training to achieve the best data fusion effect, effectively improve the accuracy and reliability of the data, and provide a solid data foundation for subsequent analysis and decision-making.

[0028] Preferably, the application layer also includes a risk assessment model to quantitatively assess the market risks, technical risks, policy risks, etc. that may be faced during the operation of the power station, providing managers with a comprehensive decision-making reference. When predicting power generation, the ARIMA (p, d, q) model is used, and its formula is:

[0029]

[0030] Among them, y t is the observed value of power generation at time t, B is the backward operator, Φ p (B) is an autoregressive polynomial, is a d-order difference operator, Θ q (B) is the moving average polynomial, is a white noise sequence.

[0031] Preferably, the method further includes IaaS resource allocation, which is divided into computing resource allocation, storage resource management and network resource optimization:

[0032] The computing resource allocation described above: uses container orchestration technology to achieve dynamic allocation of computing resources. Physical server resources are divided into multiple independent containerized service instances, each of which has an independent operating environment and resource configuration. According to the real-time changes in the power station's business load, Kubernetes automatically adjusts the number of container instances and resource allocation through built-in resource monitoring and scheduling algorithms. For example, during periods of sufficient sunlight and peak power generation, when the amount of data processing and analysis tasks increases dramatically, Kubernetes automatically and quickly increases the number of container instances related to data processing and analysis, and allocates more CPU cores and memory resources to them; at night or during periods of low power generation, when the business load is reduced, Kubernetes automatically reduces unnecessary container instances, releases idle resources, and reduces energy consumption and costs;

[0033] The storage resource management described above: Storage uses distributed object storage (such as Ceph), which stores data in a dispersed manner on multiple storage nodes. Redundant backup and fault-tolerant processing of data are achieved through a copy mechanism and erasure coding technology. Each storage node has independent read and write capabilities and can process data requests in parallel, greatly improving the data read and write speed. Ceph provides a unified storage interface to facilitate docking with upper-layer applications and platforms. At the same time, based on the access frequency and importance of data, a hierarchical storage strategy is adopted to store frequently accessed hot data on high-performance solid-state drives (SSDs), while storing less frequently accessed cold data on large-capacity mechanical hard drives (HDDs). While ensuring data access performance, it effectively reduces storage costs. The data storage location decision formula is:

[0034]

[0035] Among them, f a is the data access frequency, f threshold is the access frequency threshold;

[0036] Network resource optimization: Software-Defined Wide Area Network (SD-WAN) technology enables flexible cross-regional network configuration and optimization. The SD-WAN controller dynamically adjusts network traffic routing strategies based on the network requirements and link status of power plants in different regions. For example, for monitoring data transmission with high real-time requirements, low-latency links are prioritized; for batch data transmission tasks such as historical data backup and software upgrade package downloads, lower-cost links are selected to achieve optimal utilization of network resources. SD-WAN technology also enables network slicing management, allocating independent network resources to different business applications, ensuring network quality and security for critical services.

[0037] Preferably, the method steps also include PaaS platform construction, including machine learning platform construction, fault diagnosis and prediction, blockchain service application and intelligent decision-making engine construction:

[0038] The machine learning platform construction: Build a feature-rich, easily scalable machine learning platform that integrates multiple mainstream deep learning frameworks and traditional machine learning algorithm libraries to provide powerful data analysis and model training capabilities for power plant operation and management. The platform supports the import and preprocessing of multiple data formats and has a visual model training and evaluation interface, allowing users to quickly build and optimize machine learning models.

[0039] The fault diagnosis and prediction: Deep Belief Network (DBN) is used for fault diagnosis. The DBN model is composed of multiple restricted Boltzmann machines (RBMs). Assuming that the input equipment operation data is \(X\), the feature vector obtained after feature extraction of \(k\) layers of RBM is H k , it is calculated using the following formula:

[0040] H1=sigmoid(W1X+b1)

[0041] H2=sigmoid(W2H1+b2)

[0042] …

[0043] H k =sigmoid(W k H k-1 +b k )

[0044] Among them, W i is the weight matrix of layer i, b i is the bias vector of the i-th layer, sigmoid is the activation function, and the model parameters W are continuously optimized by combining unsupervised learning and supervised learning. i and b i , improve the accuracy and reliability of fault diagnosis. When the equipment operation data is monitored to be abnormal, the DBN model can quickly and accurately determine the fault type and fault location, issue an early warning, and provide corresponding fault solution suggestions. When calculating the fault probability, the characteristic vector H output by the DBN model can be used. k , through the logistic regression model:

[0045] P(fault)=sigmoid(β0+β1H k1 +…+β n H kn )

[0046] Among them, β0, β1,…, β n is the regression coefficient;

[0047] The blockchain service application: introduces blockchain services, uses the decentralized, tamper-proof and traceable characteristics of blockchain to achieve trusted sharing and traceability of data. In the data management of distributed photovoltaic power stations, key data is stored and shared in the form of blockchain. Each data block contains the hash value of the previous data block, forming a chain structure to ensure the integrity and consistency of the data. Any modification to the data needs to be verified by consensus of multiple nodes, and all operation records are permanently saved, which is convenient for regulatory authorities and users to trace and audit data, and enhance the credibility and security of data. The hash value of the blockchain data block is Hash. block calculate:

[0048] Hash block =H(data+Hash prev +timestamp+nonce),

[0049] Among them, H is the hash function, data is the data in the data block, Hash prev It is the hash value of the previous data block, timestamp is the timestamp, and nonce is a random number;

[0050] The intelligent decision-making engine is constructed as follows: an intelligent decision-making engine is constructed, which combines operations research algorithms (such as linear programming and integer programming) with artificial intelligence technology to formulate optimal power generation and operation and maintenance strategies based on multiple sources of information such as power plant operation data, market electricity price information, and weather forecast data. For example, when formulating a power generation plan, the difference in electricity prices in different time periods and the impact of weather changes on power generation are taken into account. Assuming that there are m power generation time periods, the electricity price of each time period is pi, the predicted power generation is ei, and the power generation cost is ci, the objective function for maximizing power generation revenue can be expressed as:

[0051]

[0052] At the same time, a series of constraints are met, such as equipment power generation capacity constraints, power balance constraints, etc. The objective function is solved by linear programming algorithm to obtain the optimal power distribution plan and maximize the power generation benefits. In terms of operation and maintenance management, the integer programming algorithm is used to reasonably arrange the equipment maintenance plan, comprehensively considering factors such as equipment service life, maintenance cost, and failure risk. Under the premise of ensuring the normal operation of the equipment, the operation and maintenance cost is minimized. At the same time, the reinforcement learning algorithm is introduced to enable the decision engine to continuously optimize the decision strategy based on real-time feedback and adapt to the complex and changeable power station operation environment. In the equipment maintenance plan, the equipment maintenance cost C is set. maintenance , equipment failure loss cost C failure , maintenance time t, then the total operation and maintenance cost C total It can be expressed as:

[0053]

[0054] Solving the optimal maintenance time t by integer programming i .

[0055] Preferably, the method also includes improving the intelligent operation and maintenance APP: the intelligent operation and maintenance APP developed for power station operation and maintenance personnel is further improved on the basis of the original functions. In addition to the functions of real-time equipment status monitoring, fault warning push, maintenance work order generation, maintenance knowledge base query, intelligent operation and maintenance assistant and inspection route planning, it also adds equipment performance analysis and comparison functions. By setting a series of key performance indicators (KPIs) to quantitatively evaluate equipment performance, for example, calculating the average power generation efficiency η of the equipment avg , the formula is:

[0056]

[0057] Among them, n is the number of data points in the statistical time period, P out,i is the output power corresponding to the i-th data point, P in,i is the input power at the same moment (such as the equivalent value of light power);

[0058] For performance comparison in different time periods, the time series analysis method is used to arrange historical data in chronological order. The moving average method is used to eliminate short-term fluctuations in the data and highlight long-term trends. Assuming the moving average period is m, the moving average power generation efficiency is:

[0059]

[0060] By comparing the moving average power generation efficiency in different time periods, we can clearly see the changing trend of the equipment power generation efficiency. If the moving average power generation efficiency in the current time period is Compared with the average power generation efficiency in the same period of history Drops below a certain threshold Δη threshold ,Right now:

[0061]

[0062] If the device performance is abnormal, the APP will automatically push a detailed performance analysis report to the operation and maintenance personnel. The report includes information such as the time period of performance degradation and possible causes.

[0063] When abnormal equipment performance is detected, the app automatically associates the equipment's historical maintenance records and fault data, and uses fault tree analysis (FTA) to graphically display the cause and effect of the fault, helping operation and maintenance personnel quickly locate the root cause of the fault. At the same time, the app is linked with the intelligent spare parts management system to automatically recommend the required spare parts list based on the equipment fault type and historical maintenance data, and display the spare parts inventory location and estimated arrival time, thereby improving fault repair efficiency.

[0064] The present invention provides a regional distributed photovoltaic power station intelligent management method based on the cloud-fog-edge architecture, which has the following beneficial effects:

[0065] By deploying a variety of high-precision sensors at the perception layer, comprehensive awareness of equipment operating status and environmental factors is achieved, enabling richer data to be acquired. In terms of equipment performance analysis, not only are indicators such as average power generation efficiency calculated, but time series analysis and data normalization are also used for multi-dimensional comparison, enabling more timely and accurate detection of equipment performance anomalies. When an anomaly is detected, fault tree analysis is used to quickly locate the root cause. This, coupled with integration with an intelligent spare parts management system, significantly improves fault repair efficiency and reduces equipment downtime, a feat unattainable with existing technologies. This invention also supports the system's use of simulation technology and Monte Carlo simulation to simulate and analyze different power plant development and operational strategies. It can predict multiple possible development scenarios, comprehensively considering factors such as the policy environment, market electricity price fluctuations, and technological advancements, providing managers with more scientific and comprehensive decision-making recommendations. The transport layer utilizes SDN technology and quantum encryption algorithms to ensure secure, stable, and efficient data transmission. The platform layer builds a data lake based on big data technology, combining a distributed file system and a column-based storage database to efficiently store and retrieve massive amounts of data. Data fusion employs a deep learning algorithm based on CNN, improving data accuracy and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is a flow chart of the intelligent management method of regional distributed photovoltaic power stations based on the cloud-edge architecture in the present invention. DETAILED DESCRIPTION

[0067] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0068] See also Figure 1The present invention provides a technical solution: a method for intelligent management of regional distributed photovoltaic power stations based on a cloud-edge architecture, the method steps including building a layered architecture:

[0069] The layered architecture includes: perception layer, transport layer, edge computing layer, platform layer and application layer;

[0070] The perception layer consists of densely deploying various types of high-precision sensors on and within each photovoltaic panel, inverter, and combiner box in a distributed photovoltaic power station, based on different monitoring requirements. In addition to conventional sensors for collecting basic data such as light intensity, temperature, voltage, and current, these sensors also include optical sensors for monitoring the cleanliness of photovoltaic panel surfaces, stress sensors for detecting stress changes in key inverter components, and sensors for monitoring meteorological parameters such as ambient humidity and wind speed. These sensors work together to comprehensively perceive the power station's operating status and environmental factors from multiple dimensions.

[0071] The transport layer uses software-defined networking (SDN) technology as its core architecture to build a highly flexible and intelligent transmission network. In the backbone network, high-fiber, high-bandwidth single-mode optical fibers are used to establish high-speed and stable data transmission channels. This ensures that data can be transmitted over long distances within the region at ultra-high speeds and low latency, meeting the needs of large-scale real-time data transmission.

[0072] The edge computing layer consists of deploying high-performance micro servers based on the advanced ARM architecture at edge nodes close to distributed photovoltaic power plants. These servers are compact, energy-efficient, and have powerful computing capabilities and local storage capabilities. They can cache raw data and preliminary processing results from the past 24 hours.

[0073] The platform layer builds a large-scale data lake based on big data technology, using an innovative storage architecture that combines a distributed file system (such as CephFS) and a columnar storage database (such as HBase) to achieve efficient storage and rapid retrieval of massive power plant operation data. The data lake can store not only structured numerical data, but also unstructured text data (such as equipment logs and maintenance records) and semi-structured JSON data (such as sensor configuration information and device associations). This comprehensive data storage method provides a rich data source for subsequent data mining and analysis.

[0074] The application layer involves developing diversified application services for different user groups. For power plant operators, a powerful intelligent operation and maintenance app integrates real-time equipment status monitoring, fault warning notifications, maintenance work order generation, and maintenance knowledge base querying. Furthermore, an intelligent operation and maintenance assistant is introduced. Leveraging natural language processing technology, operators can query equipment information and obtain maintenance recommendations through voice input, and even engage in conversations with the intelligent operation and maintenance assistant to quickly resolve issues encountered during operation and maintenance. Furthermore, the app features equipment inspection route planning. Based on the distribution and operating status of power plant equipment, a path planning algorithm is used to plan optimal inspection routes for operators, improving inspection efficiency. For managers, a decision support system is built, providing visual power plant operation reports, key performance indicator (KPI) analysis, and cost-benefit analysis. Through intuitive data visualization charts and intelligent analysis reports, managers are assisted in making scientific and rational strategic decisions. The system also incorporates intelligent forecasting and risk assessment capabilities, leveraging time series analysis and machine learning algorithms to predict key indicators such as future power generation, equipment failure rates, and cost expenditures.

[0075] Furthermore, the perception layer also needs to introduce intelligent sensors with advanced adaptive acquisition frequency adjustment functions. The sensors have built-in high-performance microprocessors and data analysis algorithms based on deep learning. By analyzing the fluctuations of the collected data in real time, when the data fluctuations exceed the preset dynamic threshold (this threshold is dynamically adjusted according to the different operating stages of the power station and the characteristics of the equipment), it is judged that the operating status of the power station has changed drastically. The intelligent sensor automatically increases the acquisition frequency to a maximum of 10 times per second to capture equipment status changes more timely and accurately. When the operating status is stable, the acquisition frequency is reduced to 1 time per minute. While ensuring data integrity, it effectively reduces the amount of data transmitted and the processing pressure, thereby reducing system energy consumption. When σ≤σ threshold When the lower acquisition frequency f low , the acquisition frequency formula can be expressed as:

[0076]

[0077] The degree of data fluctuation is measured by the standard deviation σ. When σ>σ threshold When the acquisition frequency f high .

[0078] Furthermore, the transmission layer also needs to target distributed photovoltaic power stations in remote areas or those with complex terrain and difficult wiring. The access layer adopts wireless communication technologies such as NB-IoT (narrowband Internet of Things) and 5G, and takes advantage of the low power consumption, wide coverage and low connection cost of NB-IoT technology to achieve stable connection and data collection for small distributed photovoltaic equipment; for scenarios with large data volumes and extremely high real-time requirements, such as real-time operation data transmission of inverters, the high speed (theoretical peak rate can reach 20Gbps) and low latency characteristics (end-to-end latency as low as 1 millisecond) of the 5G network are used to ensure that data can be transmitted quickly and accurately to the upper-level node; a powerful SDN controller is deployed to continuously monitor the key performance indicators of each transmission link in real time, including bandwidth usage. When a link fails, such as a fiber break, wireless signal interruption, or link congestion causing excessive delay (exceeding the preset delay threshold), the SDN controller automatically switches data traffic to the optimal backup link within milliseconds based on the preset complex link switching strategy (this strategy comprehensively considers factors such as the link's remaining bandwidth, delay, reliability, and the priority of current business needs). At the same time, to ensure the security and confidentiality of data during transmission, advanced quantum encryption algorithms are used to encrypt the transmitted data. The untapped and unbreakable characteristics of quantum key distribution are used to ensure that data is not stolen or tampered with during transmission, providing a solid guarantee for the secure transmission of power station data. When selecting a link, the formula for calculating the comprehensive link score is as follows:

[0079]

[0080] Where B is bandwidth, D is delay, L is packet loss rate, and R is reliability. w1, w2, w3, and w4 are the weights of each factor, and w1+w2+w3+w4=1 selects the link with the highest score for data transmission.

[0081] Furthermore, the edge computing layer also includes the use of a distributed hashing (DHT) algorithm to quickly classify and filter the massive amount of raw data collected by the perception layer, and map the data to different storage buckets based on multiple attributes such as the data type (such as sensor type, device type to which the data belongs), timestamp, and device identification. For real-time control data, such as the power adjustment instructions of the inverter, and important data closely related to the safe operation of the power station, such as abnormal current and voltage data, after preliminary analysis and processing by the edge computing device, they are preferentially sent to the platform layer through the transport layer; for general historical operation data, they are locally cached and uploaded to the platform layer in batches at a time period of once an hour, which greatly reduces the data transmission pressure of the core network and significantly improves the timeliness of data processing and the overall operation efficiency of the system. The location h of the data mapped to the storage bucket is calculated by the hash function H:

[0082] h=H(data_type, timestamp, device_id)

[0083] Among them, data_type is the data type, timestamp is the timestamp, and device_id is the device number.

[0084] Furthermore, the platform layer also uses a deep learning algorithm based on convolutional neural network (CNN) for data fusion. Assuming that there are n different types of data, the feature vectors of each data extracted by CNN are F1, F2, ..., F n , the fused data feature vector F fusion It can be calculated by the following formula:

[0085]

[0086] Among them, w i is the weight of the eigenvector of the i-th data, and The weights are continuously optimized through model training to achieve the best data fusion effect, effectively improve the accuracy and reliability of the data, and provide a solid data foundation for subsequent analysis and decision-making.

[0087] Furthermore, the application layer also includes a risk assessment model to quantitatively assess the market risks, technical risks, and policy risks that may be faced during the operation of the power station, providing managers with a comprehensive decision-making reference. When predicting power generation, the ARIMA (p, d, q) model is used, and its formula is:

[0088]

[0089] Among them, y t is the observed value of power generation at time t, B is the backward operator, Φ p (B) is an autoregressive polynomial, is a d-order difference operator, Θ q (B) is the moving average polynomial, is a white noise sequence.

[0090] Furthermore, the method also includes IaaS resource allocation, which is divided into computing resource allocation, storage resource management and network resource optimization:

[0091] The computing resource allocation described above: uses container orchestration technology to achieve dynamic allocation of computing resources. Physical server resources are divided into multiple independent containerized service instances, each of which has an independent operating environment and resource configuration. According to the real-time changes in the power station's business load, Kubernetes automatically adjusts the number of container instances and resource allocation through built-in resource monitoring and scheduling algorithms. For example, during periods of sufficient sunlight and peak power generation, when the amount of data processing and analysis tasks increases dramatically, Kubernetes automatically and quickly increases the number of container instances related to data processing and analysis, and allocates more CPU cores and memory resources to them; at night or during periods of low power generation, when the business load is reduced, Kubernetes automatically reduces unnecessary container instances, releases idle resources, and reduces energy consumption and costs;

[0092] The storage resource management described above: Storage uses distributed object storage (such as Ceph), which stores data in a dispersed manner on multiple storage nodes. Redundant backup and fault-tolerant processing of data are achieved through a copy mechanism and erasure coding technology. Each storage node has independent read and write capabilities and can process data requests in parallel, greatly improving the data read and write speed. Ceph provides a unified storage interface to facilitate docking with upper-layer applications and platforms. At the same time, based on the access frequency and importance of data, a hierarchical storage strategy is adopted to store frequently accessed hot data on high-performance solid-state drives (SSDs), while storing less frequently accessed cold data on large-capacity mechanical hard drives (HDDs). While ensuring data access performance, it effectively reduces storage costs. The data storage location decision formula is:

[0093]

[0094] Among them, f a is the data access frequency, f threshold is the access frequency threshold;

[0095] Network resource optimization: Software-Defined Wide Area Network (SD-WAN) technology enables flexible cross-regional network configuration and optimization. The SD-WAN controller dynamically adjusts network traffic routing strategies based on the network requirements and link status of power plants in different regions. For example, for monitoring data transmission with high real-time requirements, low-latency links are prioritized; for batch data transmission tasks such as historical data backup and software upgrade package downloads, lower-cost links are selected to achieve optimal utilization of network resources. SD-WAN technology also enables network slicing management, allocating independent network resources to different business applications, ensuring network quality and security for critical services.

[0096] Furthermore, the method also includes PaaS platform construction, including machine learning platform construction, fault diagnosis and prediction, blockchain service application and intelligent decision-making engine construction:

[0097] The machine learning platform construction: Build a feature-rich, easily scalable machine learning platform that integrates multiple mainstream deep learning frameworks and traditional machine learning algorithm libraries to provide powerful data analysis and model training capabilities for power plant operation and management. The platform supports the import and preprocessing of multiple data formats and has a visual model training and evaluation interface, allowing users to quickly build and optimize machine learning models.

[0098] The fault diagnosis and prediction: Deep Belief Network (DBN) is used for fault diagnosis. The DBN model is composed of multiple restricted Boltzmann machines (RBMs). Assuming that the input equipment operation data is \(X\), the feature vector obtained after feature extraction of \(k\) layers of RBM is H k , it is calculated using the following formula:

[0099] H1=sigmoid(W1X+b1)

[0100] H2=sigmoid(W2H1+b2)

[0101] …

[0102] H k =sigmoid(W k H k-1 +b k )

[0103] Among them, W i is the weight matrix of layer i, b i is the bias vector of the i-th layer, sigmoid is the activation function, and the model parameters W are continuously optimized by combining unsupervised learning and supervised learning. i and b i , improve the accuracy and reliability of fault diagnosis. When the equipment operation data is monitored to be abnormal, the DBN model can quickly and accurately determine the fault type and fault location, issue an early warning, and provide corresponding fault solution suggestions. When calculating the fault probability, the characteristic vector H output by the DBN model can be used. k , through the logistic regression model:

[0104] P(fault)=sigmoid(β0+β1H k1 +…+β n H kn )

[0105] Among them, β0, β1,…, β n is the regression coefficient;

[0106] The blockchain service application: introduces blockchain services, uses the decentralized, tamper-proof and traceable characteristics of blockchain to achieve trusted sharing and traceability of data. In the data management of distributed photovoltaic power stations, key data is stored and shared in the form of blockchain. Each data block contains the hash value of the previous data block, forming a chain structure to ensure the integrity and consistency of the data. Any modification to the data needs to be verified by consensus of multiple nodes, and all operation records are permanently saved, which is convenient for regulatory authorities and users to trace and audit data, and enhance the credibility and security of data. The hash value of the blockchain data block is Hash. block calculate:

[0107] Hash block =H(data+Hash prev +timestamp+nonce),

[0108] Among them, H is the hash function, data is the data in the data block, Hash prev It is the hash value of the previous data block, timestamp is the timestamp, and nonce is a random number;

[0109] The intelligent decision-making engine is constructed as follows: an intelligent decision-making engine is constructed, which combines operations research algorithms (such as linear programming and integer programming) with artificial intelligence technology to formulate optimal power generation and operation and maintenance strategies based on multiple sources of information such as power plant operation data, market electricity price information, and weather forecast data. For example, when formulating a power generation plan, the difference in electricity prices in different time periods and the impact of weather changes on power generation are taken into account. Assuming that there are m power generation time periods, the electricity price of each time period is pi, the predicted power generation is ei, and the power generation cost is ci, the objective function for maximizing power generation revenue can be expressed as:

[0110]

[0111] At the same time, a series of constraints are met, such as equipment power generation capacity constraints, power balance constraints, etc. The objective function is solved by linear programming algorithm to obtain the optimal power distribution plan and maximize the power generation benefits. In terms of operation and maintenance management, the integer programming algorithm is used to reasonably arrange the equipment maintenance plan, comprehensively considering factors such as equipment service life, maintenance cost, and failure risk. Under the premise of ensuring the normal operation of the equipment, the operation and maintenance cost is minimized. At the same time, the reinforcement learning algorithm is introduced to enable the decision engine to continuously optimize the decision strategy based on real-time feedback and adapt to the complex and changeable power station operation environment. In the equipment maintenance plan, the equipment maintenance cost C is set. maintenance , equipment failure loss cost C failure , maintenance time t, then the total operation and maintenance cost C total It can be expressed as:

[0112]

[0113] Solving the optimal maintenance time t by integer programming i .

[0114] Furthermore, the method also includes the improvement of the intelligent operation and maintenance APP: the intelligent operation and maintenance APP developed for power station operation and maintenance personnel has been further improved on the basis of the original functions. In addition to real-time equipment status monitoring, fault warning push, maintenance work order generation, maintenance knowledge base query, intelligent operation and maintenance assistant and inspection route planning, it has added equipment performance analysis and comparison functions. By setting a series of key performance indicators (KPIs) to quantitatively evaluate equipment performance, for example, calculating the average power generation efficiency η of the equipment avg , the formula is:

[0115]

[0116] Among them, n is the number of data points in the statistical time period, P out,i is the output power corresponding to the i-th data point, P in,i is the input power at the same moment (such as the equivalent value of light power);

[0117] For performance comparison in different time periods, the time series analysis method is used to arrange historical data in chronological order. The moving average method is used to eliminate short-term fluctuations in the data and highlight long-term trends. Assuming the moving average period is m, the moving average power generation efficiency is:

[0118]

[0119] By comparing the moving average power generation efficiency in different time periods, we can clearly see the changing trend of the equipment power generation efficiency. If the moving average power generation efficiency in the current time period is Compared with the average power generation efficiency in the same period of history Drops below a certain threshold Δη threshold ,Right now:

[0120]

[0121] If the device performance is abnormal, the APP will automatically push a detailed performance analysis report to the operation and maintenance personnel. The report includes information such as the time period of performance degradation and possible causes.

[0122] When abnormal equipment performance is detected, the app automatically associates the equipment's historical maintenance records and fault data, and uses fault tree analysis (FTA) to graphically display the cause and effect of the fault, helping operation and maintenance personnel quickly locate the root cause of the fault. At the same time, the app is linked with the intelligent spare parts management system to automatically recommend the required spare parts list based on the equipment fault type and historical maintenance data, and display the spare parts inventory location and estimated arrival time, thereby improving fault repair efficiency.

[0123] By those skilled in the art, the method in this case is operated in sequence. The specific method and operation sequence should refer to the following working principle. The detailed connection means are well-known technologies in the field. The following mainly introduces the working principle and process.

[0124] Embodiment: comprising the following steps:

[0125] S1. Data collection at the perception layer mainly involves intelligent sensors adaptively adjusting the collection frequency based on data fluctuations.

[0126] S2. Transport layer data transmission, selects the optimal link to transmit data, and uses quantum encryption algorithm to ensure data transmission security.

[0127] S3. Initial processing at the edge computing layer: Use the DHT algorithm to quickly classify and filter the massive amount of raw data collected by the perception layer.

[0128] S4. Platform-level data fusion and storage: Utilize CNN-based deep learning algorithms for data fusion, optimize weights through training, and improve data accuracy and reliability.

[0129] S5.PaaS platform function implementation: build a machine learning platform and integrate multiple algorithm frameworks.

[0130] S6.SaaS service application: An intelligent operation and maintenance APP for operation and maintenance personnel that integrates multiple functions.

[0131] Among them, S1 is specifically:

[0132] Various high-precision sensors are deployed on various components of distributed photovoltaic power stations, including sensors for light intensity, temperature, voltage, current, stress, cleanliness, etc., to fully and real-timely perceive the power station's operating status and environmental factors. Intelligent sensors adaptively adjust the collection frequency according to data fluctuations. The degree of data fluctuation is measured by the standard deviation σ, and the preset dynamic threshold is σ threshold , when σ>σ threshold When σ≤σ, the data is collected 10 times per second, and the acquisition frequency is f=10Hz; when σ≤σ threshold The collection frequency is 1 time per minute. To accurately capture device status changes while reducing data transmission and processing pressure, the frequency adjustment formula is:

[0133] S2 is specifically:

[0134] The transmission network is built using SDN technology. The backbone network uses optical fiber to ensure high-speed and stable transmission, while the access layer in remote areas utilizes wireless technologies such as NB-IoT and 5G. The SDN controller monitors link status in real time, calculates a comprehensive link score based on factors such as bandwidth, latency, packet loss rate, and reliability, and selects the optimal link for data transmission. Quantum encryption algorithms are also used to ensure data security.

[0135] S3 specifically:

[0136] ARM-based microservers are deployed at edge nodes near power plants, using the Distributed Hierarchy Transport (DHT) algorithm to rapidly classify and filter the massive amounts of raw data collected by the perception layer. Real-time control data and critical data are prioritized for processing and transmission to the platform layer. General historical data is cached locally and uploaded in batches every hour, reducing transmission pressure on the core network and improving data processing timeliness.

[0137] S4 is specifically:

[0138] Based on big data technology, a data lake is built, which uses a combination of distributed file system and column storage database to store massive structured, unstructured and semi-structured data. Assume that there are n different types of data, and the feature vectors extracted by CNN for each data are F1, F2, ..., F n , the fused data feature vector F fusion It can be calculated by the following formula:

[0139]

[0140] Among them, wi is the weight of the i-th data feature vector, and The weights are continuously optimized through model training to achieve the best data fusion effect, improve data accuracy and reliability, and provide a data basis for subsequent analysis and decision-making.

[0141] S5 is specifically:

[0142] Build a machine learning platform that integrates multiple algorithm frameworks. Leverage DBN for fault diagnosis and calculate fault probabilities through multi-layer RBM feature extraction. Introduce blockchain services to ensure trusted data sharing and traceability. Build an intelligent decision-making engine that combines operations research and AI technology to formulate power generation and operation and maintenance strategies based on multi-source information. For example, use linear programming to maximize power generation revenue and integer programming to schedule equipment maintenance.

[0143] S6 specifically:

[0144] The intelligent operation and maintenance APP for operation and maintenance personnel integrates multiple functions and calculates the average power generation efficiency η of the equipment. avg , the formula is:

[0145]

[0146] For performance comparison in different time periods, the time series analysis method is used to arrange historical data in chronological order, and the moving average method is used to eliminate short-term fluctuations in the data and highlight long-term trends. Assuming the moving average period is m, the moving average power generation efficiency is: By comparing the moving average power generation efficiency in different time periods, the changing trend of the equipment power generation efficiency can be clearly seen. Compared with the average power generation efficiency in the same period of history Drops below a certain threshold Δη threshold ,Right now:

[0147]

[0148] If the equipment performance is abnormal, the fault tree analysis method is used to locate the fault, and the fault repair efficiency is improved by linkage with the intelligent spare parts management system. The decision support system for managers uses simulation technology and Monte Carlo simulation method to simulate strategic planning and risk assessment, and uses time series analysis algorithms (such as ARIMA model) to predict power generation. The formula is:

[0149]

[0150] Combined with other data, it provides decision-making recommendations to assist in scientific decision-making.

[0151] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include," "comprise," or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations. The phrase "includes an element defined by..." does not exclude the presence of other identical elements in the process, method, article, or device that includes the element.

[0152] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A regional distributed photovoltaic power station intelligent management method based on cloud-fog-edge architecture, characterized by: The method steps include building a layered architecture: The layered architecture includes: perception layer, transport layer, edge computing layer, platform layer and application layer; The perception layer consists of densely deploying various types of high-precision sensors on and within each photovoltaic panel, inverter, and combiner box in a distributed photovoltaic power station, based on different monitoring requirements. In addition to conventional sensors for collecting basic data such as light intensity, temperature, voltage, and current, these sensors also include optical sensors for monitoring the cleanliness of photovoltaic panel surfaces, stress sensors for detecting stress changes in key inverter components, and sensors for monitoring meteorological parameters such as ambient humidity and wind speed. These sensors work together to comprehensively perceive the power station's operating status and environmental factors from multiple dimensions. The transport layer uses software-defined networking (SDN) technology as its core architecture to build a highly flexible and intelligent transmission network. In the backbone network, high-fiber, high-bandwidth single-mode optical fibers are used to establish high-speed and stable data transmission channels. This ensures that data can be transmitted over long distances within the region at ultra-high speeds and low latency, meeting the needs of large-scale real-time data transmission. The edge computing layer consists of deploying high-performance micro servers based on the advanced ARM architecture at edge nodes close to distributed photovoltaic power plants. These servers are compact, energy-efficient, and have powerful computing capabilities and local storage capabilities. They can cache raw data and preliminary processing results from the past 24 hours. The platform layer builds an ultra-large-scale data lake based on big data technology, using an innovative storage architecture that combines a distributed file system with a columnar storage database to achieve efficient storage and rapid retrieval of massive power plant operation data. The data lake can store not only structured numerical data, but also unstructured text data and semi-structured JSON data. This comprehensive data storage method provides a rich data source for subsequent data mining and analysis. The application layer involves developing diversified application services for different user groups. For power plant operators, a powerful intelligent operation and maintenance app integrates real-time equipment status monitoring, fault warning notifications, maintenance work order generation, and maintenance knowledge base querying. Furthermore, an intelligent operation and maintenance assistant is introduced. Leveraging natural language processing technology, operators can query equipment information and obtain maintenance recommendations through voice input, and even engage in conversations with the intelligent operation and maintenance assistant to quickly resolve issues encountered during operation and maintenance. Furthermore, the app features equipment inspection route planning. Based on the distribution and operating status of power plant equipment, a path planning algorithm is used to plan optimal inspection routes for operators, improving inspection efficiency. For managers, a decision support system is built, providing visual power plant operation reports, key performance indicator (KPI) analysis, and cost-benefit analysis. Through intuitive data visualization charts and intelligent analysis reports, managers are assisted in making scientific and rational strategic decisions. The system also incorporates intelligent forecasting and risk assessment capabilities, leveraging time series analysis and machine learning algorithms to predict key indicators such as future power generation, equipment failure rates, and cost expenditures.

2. The method for intelligent management of regional distributed photovoltaic power stations based on cloud-fog-edge architecture according to claim 1 is characterized in that: The perception layer also needs to introduce intelligent sensors with advanced adaptive acquisition frequency adjustment functions. The sensors have built-in high-performance microprocessors and data analysis algorithms based on deep learning. By analyzing the fluctuations of the collected data in real time, when the data fluctuations exceed the preset dynamic threshold, it is judged that the operating status of the power station has changed drastically. The intelligent sensor automatically increases the acquisition frequency to a maximum of 10 times per second to capture equipment status changes more timely and accurately. When the operating status is stable, the acquisition frequency is reduced to 1 time per minute. While ensuring data integrity, it effectively reduces the amount of data transmitted and processing pressure, thereby reducing system energy consumption. When σ≤σ threshold When the lower acquisition frequency f low , the acquisition frequency formula can be expressed as: The degree of data fluctuation is measured by the standard deviation σ. When σ>σ threshold When the acquisition frequency f high .

3. The method for intelligent management of regional distributed photovoltaic power stations based on cloud-fog-edge architecture according to claim 1 is characterized in that: The transmission layer also needs to target distributed photovoltaic power stations in remote areas or with complex terrain and difficult wiring. The access layer adopts wireless communication technologies such as NB-IoT (Narrowband Internet of Things) and 5G, and takes advantage of the low power consumption, wide coverage and low connection cost of NB-IoT technology to achieve stable connection and data collection for small distributed photovoltaic equipment. For scenarios with large data volumes and extremely high real-time requirements, such as real-time operation data transmission of inverters, the high speed and low latency characteristics of the 5G network are used to ensure that data can be transmitted quickly and accurately to the upper node. A powerful SDN controller is deployed to continuously monitor the key performance indicators of each transmission link in real time. Including bandwidth usage, latency, and packet loss rate. When a link fails, such as a fiber break, wireless signal interruption, or link congestion causing excessive latency, the SDN controller automatically switches data traffic to the optimal backup link within milliseconds based on a preset complex link switching strategy. At the same time, to ensure the security and confidentiality of data during transmission, advanced quantum encryption algorithms are used to encrypt the transmitted data. The untapped and unbreakable characteristics of quantum key distribution are used to ensure that data cannot be stolen or tampered with during transmission, providing a solid guarantee for the secure transmission of power station data. When selecting a link, the formula for calculating the link comprehensive score is as follows: Where B is bandwidth, D is delay, L is packet loss rate, and R is reliability. w1, w2, w3, and w4 are the weights of each factor, and w1+w2+w3+w4=1 selects the link with the highest score for data transmission.

4. The method for intelligent management of regional distributed photovoltaic power stations based on cloud-fog-edge architecture according to claim 1 is characterized in that: The edge computing layer also includes the use of a distributed hashing (DHT) algorithm to quickly classify and filter the massive amount of raw data collected by the perception layer, and map the data to different storage buckets based on multiple attributes such as data type, timestamp, and device identification. For real-time control data, such as the power adjustment instructions of the inverter, and important data closely related to the safe operation of the power station, such as abnormal current and voltage data, after preliminary analysis and processing by the edge computing device, they are preferentially sent to the platform layer through the transport layer; for general historical operation data, they are locally cached and uploaded to the platform layer in batches at a time period of once an hour, which greatly reduces the data transmission pressure of the core network and significantly improves the timeliness of data processing and the overall operation efficiency of the system. The location h of the data mapped to the storage bucket is calculated by the hash function H: h=H(data_type, timestamp, device_id) Among them, data_type is the data type, timestamp is the timestamp, and device_id is the device number.

5. The method for intelligent management of regional distributed photovoltaic power stations based on cloud-fog-edge architecture according to claim 1 is characterized in that: The platform layer also uses a deep learning algorithm based on convolutional neural network (CNN) for data fusion. Assuming that there are n different types of data, the feature vectors of each data extracted by CNN are F1, F2, ..., F n , the fused data feature vector F fusion It can be calculated by the following formula: Among them, w i is the weight of the eigenvector of the i-th data, and The weights are continuously optimized through model training to achieve the best data fusion effect, effectively improve the accuracy and reliability of the data, and provide a solid data foundation for subsequent analysis and decision-making.

6. The method for intelligent management of regional distributed photovoltaic power stations based on cloud-fog-edge architecture according to claim 1, characterized in that: The application layer also includes a risk assessment model to quantitatively assess the market risks, technical risks, and policy risks that may be faced during the operation of the power station, providing managers with a comprehensive decision-making reference. When predicting power generation, the ARIMA (p, d, q) model is used, and its formula is: Among them, y t is the observed value of power generation at time t, B is the backward operator, Φ p (B) is an autoregressive polynomial, is a d-order difference operator, Θ q (B) is the moving average polynomial, t is a white noise sequence.

7. The method for intelligent management of regional distributed photovoltaic power stations based on cloud-fog-edge architecture according to claim 1, characterized in that: The method also includes IaaS resource allocation, which is divided into computing resource allocation, storage resource management and network resource optimization: The computing resource allocation described above: uses container orchestration technology to achieve dynamic allocation of computing resources. Physical server resources are divided into multiple independent containerized service instances, each of which has an independent operating environment and resource configuration. According to the real-time changes in the power station's business load, Kubernetes automatically adjusts the number of container instances and resource allocation through built-in resource monitoring and scheduling algorithms. For example, during periods of sufficient sunlight and peak power generation, when the amount of data processing and analysis tasks increases dramatically, Kubernetes automatically and quickly increases the number of container instances related to data processing and analysis, and allocates more CPU cores and memory resources to them; at night or during periods of low power generation, when the business load is reduced, Kubernetes automatically reduces unnecessary container instances, releases idle resources, and reduces energy consumption and costs; The storage resource management described above: Storage uses distributed object storage (such as Ceph), which stores data in a dispersed manner on multiple storage nodes. Redundant backup and fault-tolerant processing of data are achieved through a copy mechanism and erasure coding technology. Each storage node has independent read and write capabilities and can process data requests in parallel, greatly improving the data read and write speed. Ceph provides a unified storage interface to facilitate docking with upper-layer applications and platforms. At the same time, based on the access frequency and importance of data, a hierarchical storage strategy is adopted to store frequently accessed hot data on high-performance solid-state drives (SSDs), while storing less frequently accessed cold data on large-capacity mechanical hard drives (HDDs). While ensuring data access performance, it effectively reduces storage costs. The data storage location decision formula is: Among them, f a is the data access frequency, f threshold is the access frequency threshold; Network resource optimization: Software-Defined Wide Area Network (SD-WAN) technology enables flexible cross-regional network configuration and optimization. The SD-WAN controller dynamically adjusts network traffic routing strategies based on the network requirements and link status of power plants in different regions. For example, for monitoring data transmission with high real-time requirements, low-latency links are prioritized; for batch data transmission tasks such as historical data backup and software upgrade package downloads, lower-cost links are selected to achieve optimal utilization of network resources. SD-WAN technology also enables network slicing management, allocating independent network resources to different business applications, ensuring network quality and security for critical services.

8. The method for intelligent management of regional distributed photovoltaic power stations based on cloud-fog-edge architecture according to claim 1, characterized in that: The method also includes PaaS platform construction, including machine learning platform construction, fault diagnosis and prediction, blockchain service application and intelligent decision-making engine construction: The machine learning platform construction: Build a feature-rich, easily scalable machine learning platform that integrates multiple mainstream deep learning frameworks and traditional machine learning algorithm libraries to provide powerful data analysis and model training capabilities for power plant operation and management. The platform supports the import and preprocessing of multiple data formats and has a visual model training and evaluation interface, allowing users to quickly build and optimize machine learning models. The fault diagnosis and prediction: Deep Belief Network (DBN) is used for fault diagnosis. The DBN model is composed of multiple restricted Boltzmann machines (RBMs). Assuming that the input equipment operation data is \(X\), the feature vector obtained after feature extraction of \(k\) layers of RBM is H k , it is calculated using the following formula: H1=sigmoid(W1X+b1) H2=sigmoid(W2H1+b2) … H k =sigmoid(W k H k-1 +b k ) Among them, W i is the weight matrix of layer i, b i is the bias vector of the i-th layer, sigmoid is the activation function, and the model parameters W are continuously optimized by combining unsupervised learning and supervised learning. i and b i , improve the accuracy and reliability of fault diagnosis. When the equipment operation data is monitored to be abnormal, the DBN model can quickly and accurately determine the fault type and fault location, issue an early warning, and provide corresponding fault solution suggestions. When calculating the fault probability, the characteristic vector H output by the DBN model can be used. k , through the logistic regression model: P(fault)=sigmoid(β0+β1H k1 +…+b n H kn ) Among them, β0, β1,…, β n is the regression coefficient; The blockchain service application: introduces blockchain services, uses the decentralized, tamper-proof and traceable characteristics of blockchain to achieve trusted sharing and traceability of data. In the data management of distributed photovoltaic power stations, key data is stored and shared in the form of blockchain. Each data block contains the hash value of the previous data block, forming a chain structure to ensure the integrity and consistency of the data. Any modification to the data needs to be verified by consensus of multiple nodes, and all operation records are permanently saved, which is convenient for regulatory authorities and users to trace and audit data, and enhance the credibility and security of data. The hash value of the blockchain data block is Hash. block calculate: Hash block =H(data+Hash prev +timestamp+nonce), Among them, H is the hash function, data is the data in the data block, Hash prev It is the hash value of the previous data block, timestamp is the timestamp, and nonce is a random number; The intelligent decision-making engine is constructed: an intelligent decision-making engine is constructed, which combines operations research algorithms (such as linear programming, integer programming) and artificial intelligence technology to formulate the optimal power generation and operation and maintenance strategy based on multi-source information such as power station operation data, market electricity price information, weather forecast data, etc. For example, when formulating a power generation plan, the difference in electricity prices in different time periods and the impact of weather changes on power generation are taken into account. Assume that there are m power generation time periods in total, and the electricity price of each time period is p i , the predicted power generation is e i , the power generation cost is c i , then the objective function of maximizing power generation revenue can be expressed as: At the same time, a series of constraints are met, such as equipment power generation capacity constraints, power balance constraints, etc. The objective function is solved by linear programming algorithm to obtain the optimal power distribution plan and maximize the power generation benefits. In terms of operation and maintenance management, the integer programming algorithm is used to reasonably arrange the equipment maintenance plan, comprehensively considering factors such as equipment service life, maintenance cost, and failure risk. Under the premise of ensuring the normal operation of the equipment, the operation and maintenance cost is minimized. At the same time, the reinforcement learning algorithm is introduced to enable the decision engine to continuously optimize the decision strategy based on real-time feedback and adapt to the complex and changeable power station operation environment. In the equipment maintenance plan, the equipment maintenance cost C is set. maintenance , equipment failure loss cost C failure , maintenance time t, then the total operation and maintenance cost C total It can be expressed as: Solving the optimal maintenance time t by integer programming i .

9. The method for intelligent management of regional distributed photovoltaic power stations based on cloud-fog-edge architecture according to claim 1, characterized in that: The method also includes improving the intelligent operation and maintenance APP: the intelligent operation and maintenance APP developed for power plant operation and maintenance personnel has been further improved on the basis of the original functions. In addition to real-time equipment status monitoring, fault warning push, maintenance work order generation, maintenance knowledge base query, intelligent operation and maintenance assistant and inspection route planning, it has also added equipment performance analysis and comparison functions. By setting a series of key performance indicators (KPIs) to quantitatively evaluate equipment performance, for example, calculating the average power generation efficiency η of the equipment avg , the formula is: Among them, n is the number of data points in the statistical time period, P out,i is the output power corresponding to the i-th data point, P in,i is the input power at the same moment (such as the equivalent value of light power); For performance comparison in different time periods, the time series analysis method is used to arrange historical data in chronological order. The moving average method is used to eliminate short-term fluctuations in the data and highlight long-term trends. Assuming the moving average period is m, the moving average power generation efficiency is: By comparing the moving average power generation efficiency in different time periods, we can clearly see the changing trend of the equipment power generation efficiency. If the moving average power generation efficiency in the current time period is Compared with the average power generation efficiency in the same period of history Drops below a certain threshold Δη threshold ,Right now: If the device performance is abnormal, the APP will automatically push a detailed performance analysis report to the operation and maintenance personnel. The report includes information such as the time period of performance degradation and possible causes. When abnormal equipment performance is detected, the app automatically associates the equipment's historical maintenance records and fault data, and uses fault tree analysis (FTA) to graphically display the cause and effect of the fault, helping operation and maintenance personnel quickly locate the root cause of the fault. At the same time, the app is linked with the intelligent spare parts management system to automatically recommend the required spare parts list based on the equipment fault type and historical maintenance data, and display the spare parts inventory location and estimated arrival time, thereby improving fault repair efficiency.

Images