A complex terrain-oriented distributed photovoltaic power generation space-time coordination prediction system

Abstract

This invention belongs to the field of spatiotemporal collaborative prediction technology. It discloses a spatiotemporal collaborative prediction system for distributed photovoltaic power generation in complex terrain. The system includes: a spatiotemporal correlation network construction and update module, a model training and collaborative prediction module, a risk quantification and assessment module, and a reinforcement learning decision strategy generation module. Based on spatiotemporal feature reports, the spatiotemporal correlation network is constructed and updated to obtain a dynamic graph structure. Based on the dynamic graph structure, a graph neural network model is trained and predicted to obtain a collaborative prediction report. Based on the collaborative prediction report, prediction uncertainty is quantified, and risk is assessed to obtain an uncertainty report. Based on the uncertainty report and system state, a strategy network is used to learn strategies, generating a strategy control report. Overall, this invention has significant advantages such as high accuracy of spatiotemporal collaborative prediction, strong system adaptability under changing environmental conditions, and good intelligent decision-making mechanism.

Description

Technical Field

[0001] This invention relates to the field of spatiotemporal collaborative prediction technology, and more specifically, to a spatiotemporal collaborative prediction system for distributed photovoltaic power generation in complex terrain. Background Technology

[0002] The distributed photovoltaic power generation spatiotemporal collaborative prediction system is an advanced energy management technology for areas with complex geographical environments (such as remote mountainous areas). By integrating meteorological, topographical and photovoltaic unit operation data, it uses artificial intelligence and pattern recognition algorithms to accurately predict the photovoltaic power generation within a specific spatiotemporal range in the future, and generates collaborative control instructions accordingly to achieve efficient, autonomous and stable operation of regional energy.

[0003] However, traditional distributed photovoltaic (PV) power generation spatiotemporal collaborative prediction systems suffer from several shortcomings during use. First, these systems often rely on single or limited data sources, making it difficult to effectively integrate heterogeneous data from meteorological satellites, ground weather stations, terrain elevation, and PV operation. In complex terrain conditions such as mountainous areas, this results in incomplete input features and generally higher prediction errors. Second, traditional systems typically construct the network of relationships between PV units based on fixed electrical connections or simple distance weights, neglecting the dynamic changes in spatiotemporal relationships caused by environmental factors. This makes it difficult for traditional systems to accurately reflect the interactions between PV units in complex environments, thus reducing the reliability of prediction results. Third, traditional systems are largely limited to providing prediction results and lack assessment of the uncertainty of those results. In the event of extreme weather or other uncertainties, the system may struggle to predict risks and adjust power supply strategies in a timely manner, potentially threatening grid security. Overall, effectively addressing the problems of low spatiotemporal collaborative prediction accuracy, weak adaptability to changing environments, and insufficient intelligence in decision-making mechanisms in traditional systems has become a challenge that current distributed PV power generation spatiotemporal collaborative prediction systems need to address.

[0004] In view of this, the present invention proposes a spatiotemporal collaborative prediction system for distributed photovoltaic power generation in complex terrain to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution, including: The multi-source data acquisition module is used to acquire raw data in real time based on sensor networks and to adjust the acquisition time window to obtain the raw dataset. Furthermore, the steps of acquiring raw data in real time based on sensor networks and adjusting the acquisition time window include: S1.1: Based on the sensor network, raw data is collected in real time to obtain the sensor dataset, which includes meteorological satellite data, ground meteorological station data, terrain elevation data and photovoltaic unit operation data; S1.2: Execute the data collection task in step S1.1 based on the preset data collection time; S1.3: Integrate the acquisition timestamps and spatial coordinates of all data items in the sensor dataset to obtain the original dataset; S1.4: Store the original dataset in the database and output it to the data preprocessing and fusion module; The data preprocessing and fusion module is used to preprocess, extract features, and perform spatiotemporal fusion based on the original dataset to obtain a spatiotemporal feature report. Furthermore, the steps of preprocessing, feature extraction, and spatiotemporal fusion based on the original dataset include: S2.1: Preprocessing is performed on the original dataset to obtain an aligned dataset. Preprocessing includes data cleaning, data normalization, and time alignment. The specific preprocessing method is as follows: The missing values ​​in the original dataset are identified and filled using linear interpolation to complete the data cleaning process. The original dataset after data cleaning is normalized using a normalization formula to complete the data normalization process. The original dataset after data normalization is unified to the same timestamp, and missing values ​​are filled using the neighbor interpolation method to complete time alignment; S2.2: Using a convolutional neural network, feature extraction is performed on the meteorological satellite data and topographic elevation data in the aligned dataset to obtain spatial features; Using a long short-term memory network, feature extraction is performed on ground meteorological station data and photovoltaic unit operation data in the aligned dataset to obtain time-series features; S2.3: Use an attention mechanism to perform weighted fusion of spatial and temporal features to obtain a spatiotemporal feature report; S2.4: Output the spatiotemporal feature report to the spatiotemporal correlation network construction and update module; The spatiotemporal correlation network construction and update module is used to construct and update the spatiotemporal correlation network based on the spatiotemporal feature report to obtain a dynamic graph structure; Furthermore, the steps for constructing and updating the spatiotemporal correlation network based on spatiotemporal feature reports include: S3.1: Based on the database, the initial spatiotemporal correlation network is retrieved, feature data is extracted according to the spatiotemporal feature report to obtain node features, and the node features are input into the spatiotemporal correlation network for parameter calculation to obtain the first spatiotemporal correlation network; S3.2: Adjust the edge weights of the first spatiotemporal association network based on real-time meteorological satellite data retrieved from the database, calculate the cloud impact factor based on cloud coverage data in the spatiotemporal feature report, and finally optimize the first spatiotemporal association network using graph structure learning technology to obtain a dynamic graph structure. S3.3: Output the dynamic graph structure to the model training and co-prediction module; The model training and co-prediction module is used to train and predict graph neural network models based on dynamic graph structures, and to obtain a co-prediction report. Furthermore, the steps for training and predicting graph neural network models based on dynamic graph structures include: S4.1: Retrieve the pre-trained graph neural network model based on the database and train it according to historical data to obtain the final graph neural network model. The historical data includes the model's historical predicted power generation value and the corresponding actual power generation value. S4.2: Input the dynamic graph structure and node features into the final graph neural network model, and output the predicted power generation value set; S4.3: Calculate the regression sum of squares based on the predicted power generation value set and the actual power generation value to obtain the model evaluation value; S4.4: Package the predicted power generation value set and the model evaluation value to obtain a collaborative prediction report; S4.5: Output the collaborative forecasting report to the risk quantification and assessment module; The risk quantification and assessment module is used to quantify the uncertainty of forecasts based on collaborative forecasting reports, assess the risks, and obtain an uncertainty report. Furthermore, based on the collaborative forecasting report, the steps for quantifying forecast uncertainty and assessing risk include: S5.1: Based on the historical collaborative prediction reports retrieved from the database, a collaborative prediction set is obtained, and the prediction mean, prediction variance, and confidence interval are calculated based on the collaborative prediction set. S5.2: Calculations are performed based on confidence intervals to obtain the risk assessment value and conditional risk value. The specific formula set for the calculation is as follows: Risk assessment values ​​were obtained respectively. Conditional risk value ,in, To predict the mean, Let be the confidence interval. To predict variance, This represents the confidence level of the confidence interval. Confidence level The risk assessment value, For integration variables The differential; S5.3: Package the forecast mean, forecast variance, confidence interval, risk assessment value, and conditional risk value to obtain an uncertainty report; S5.4: Output the uncertainty report to the reinforcement learning decision-making strategy generation module; The reinforcement learning decision-making strategy generation module is used to learn policies from the policy network based on uncertainty reports and system states, and generate policy regulation reports. Furthermore, the steps for generating a policy regulation report by performing policy learning on the policy network based on uncertainty reports and system states include: S6.1: Extract the prediction variance and risk assessment value based on the uncertainty report, extract the predicted power generation value set based on the collaborative prediction report, retrieve the state parameters based on the database, and package the prediction variance, risk assessment value, power generation value set and state parameters to obtain the state vector. The state parameters include energy storage capacity and load demand. S6.2: Retrieve control commands from the database as action vectors; S6.3: Based on the state vector and action vector, and using the near-end policy optimization algorithm, the policy network is trained. After reaching the preset number of iterations, the final policy network is output. S6.4: Input the state vector into the final policy network and output the policy control report; S6.5: Output the strategy control report to the system instruction distribution and execution module; The system instruction distribution and execution module is used to convert policy control reports into executable instructions, and to distribute and monitor their execution status. Furthermore, the steps of converting the policy control report into executable instructions, distributing them, and monitoring their execution status include: S7.1: Convert the control instructions in the strategy control report into a standard format and output them to the corresponding target device. At the same time, use the confirmation mechanism to trigger retransmission when the system does not receive a confirmation signal. S7.2: Monitor the execution of instructions on the target device. When the monitoring result indicates that the execution has failed, generate a warning report and record the execution failure log. The warning report indicates that the A0** command failed to execute on the target device, and requests that staff proceed to the target device location for inspection as soon as possible. S7.3: Package the instruction output status, monitoring results, instruction output timestamp, and monitoring result timestamp from step S7.1 to obtain the instruction execution report; S7.4: Output the instruction execution report to the reinforcement learning decision policy generation module for policy adjustment of the policy network; Furthermore, S1: Based on the sensor network, raw data is collected in real time, and the collection time window is adjusted to obtain the raw dataset; S2: Based on the original dataset, preprocessing, feature extraction, and spatiotemporal fusion are performed to obtain a spatiotemporal feature report; S3: Based on the spatiotemporal feature report, the spatiotemporal correlation network is constructed and updated to obtain a dynamic graph structure; S4: Train and predict the graph neural network model based on the dynamic graph structure to obtain a collaborative prediction report; S5: Based on the collaborative forecasting report, quantify the forecast uncertainty and assess the risk to obtain an uncertainty report; S6: Based on the uncertainty report and system state, the policy network is used to learn policies and generate a policy regulation report; S7: Convert policy control reports into executable instructions, and distribute and monitor their execution status.

[0006] The technical effects and advantages of the distributed photovoltaic power generation spatiotemporal collaborative prediction system for complex terrain proposed in this invention are as follows: This invention acquires raw data in real time using a sensor network and adjusts the acquisition time window to obtain a raw dataset. Based on this raw dataset, preprocessing, feature extraction, and spatiotemporal fusion are performed to obtain a spatiotemporal feature report. The spatiotemporal correlation network is constructed and updated based on this report to obtain a dynamic graph structure. A graph neural network model is trained and used to predict based on this dynamic graph structure to obtain a collaborative prediction report. Based on this collaborative prediction report, prediction uncertainty is quantified, and risk is assessed to obtain an uncertainty report. Based on the uncertainty report and system status, a policy network is used for policy learning to generate a policy control report. This policy control report is converted into executable instructions, distributed, and its execution status is monitored. This allows the system to achieve deep fusion of multi-dimensional environmental and operational data based on a multi-source data acquisition module and a data preprocessing and fusion module, and to extract key spatiotemporal features, maximizing... This invention significantly enhances the system's performance in complex terrain and meteorological environments. By combining the spatiotemporal correlation network construction and update module and the model training collaborative prediction module, the system is further able to capture the dynamic correlation between photovoltaic units, thereby effectively improving prediction accuracy and fundamentally reducing prediction errors. In addition, the establishment of the spatiotemporal correlation network construction and update module enables the system to maintain extremely high accuracy and reliability under changing environmental conditions. Finally, through the collaborative operation of the risk quantification assessment module, the reinforcement learning decision strategy generation module, and the system instruction distribution and execution module, the system possesses a proactive risk response capability, that is, it actively adjusts the equipment operation mode before extreme conditions occur, maximizing equipment safety. Overall, this invention has significant advantages such as high accuracy of spatiotemporal collaborative prediction, strong system adaptability under changing environmental conditions, and good intelligent decision-making mechanism. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of a distributed photovoltaic power generation spatiotemporal collaborative prediction system for complex terrain according to the present invention; Figure 2 This is a schematic diagram of a spatiotemporal collaborative prediction method for distributed photovoltaic power generation in complex terrain according to the present invention. Detailed Implementation

[0008] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0009] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” used in the embodiments of this invention are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0010] Depending on the context, the words “if” or “suppose” as used here can be interpreted as “when” or “in response to determination” or “in response to detection.” Similarly, depending on the context, the phrases “if determination” or “if detection (of the stated condition or event)” can be interpreted as “when determination” or “in response to determination” or “when detection (of the stated condition or event)” or “in response to detection (of the stated condition or event).”

[0011] Furthermore, the timing of the steps in the following method embodiments is merely an example and not a strict limitation.

[0012] In practice, the server-side equipment deployed in a distributed photovoltaic (PV) power generation spatiotemporal collaborative prediction system for complex terrain may consist of one or more devices. This system can be implemented as a business instance, a virtual machine, or hardware devices. For example, it can be implemented as a business instance deployed on one or more devices in a cloud node. Simply put, it can be understood as software deployed on a cloud node, providing spatiotemporal collaborative prediction services for complex terrain to various user terminals. Alternatively, it can be implemented as a virtual machine deployed on one or more devices in a cloud node, with application software installed to manage each user terminal. Or, it can also be implemented as a server-side system composed of numerous identical or different types of hardware devices, with one or more devices configured to provide spatiotemporal collaborative prediction services for complex terrain to various user terminals.

[0013] In terms of implementation, the spatiotemporal collaborative prediction system for distributed photovoltaic power generation in complex terrain and the user terminal are mutually adaptable. That is, if the spatiotemporal collaborative prediction system for distributed photovoltaic power generation in complex terrain is implemented as an application installed on a cloud service platform, then the user terminal is a client that establishes a communication connection with the application; or if the spatiotemporal collaborative prediction system for distributed photovoltaic power generation in complex terrain is implemented as a website, then the user terminal is implemented as a webpage; or if the spatiotemporal collaborative prediction system for distributed photovoltaic power generation in complex terrain is implemented as a cloud service platform, then the user terminal is implemented as a mini-program in an instant messaging application.

[0014] like Figure 1 The figure shown is a system architecture diagram of a distributed photovoltaic power generation spatiotemporal collaborative prediction system for complex terrain provided by an embodiment of the present invention.

[0015] The distributed photovoltaic power generation spatiotemporal collaborative prediction system for complex terrain described in this invention can be set up in a cloud server. In terms of implementation, it can be implemented as one or more service devices, or as an application installed in the cloud (e.g., a mobile service operator's server or server cluster), or it can be developed as a website. Depending on the functions implemented, the distributed photovoltaic power generation spatiotemporal collaborative prediction system for complex terrain may include a multi-source data acquisition module, a data preprocessing and fusion module, a spatiotemporal correlation network construction and update module, a model training and collaborative prediction module, a risk quantification and assessment module, a reinforcement learning decision strategy generation module, and a system instruction distribution and execution module. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device's processor and perform a fixed function, stored in the electronic device's memory.

[0016] In this embodiment of the invention, in the distributed photovoltaic power generation spatiotemporal collaborative prediction system for complex terrain, each of the above modules can be implemented independently and can call other modules. Here, "calling" can be understood as one module connecting to multiple modules of another type and providing corresponding services to those connected modules. For example, the system instruction distribution and execution module can call the same information acquisition module to obtain information collected by that module. Based on the above characteristics, in the distributed photovoltaic power generation spatiotemporal collaborative prediction system for complex terrain provided in this embodiment of the invention, the applicable scope of the system architecture can be adjusted by adding modules and directly calling them without modifying the program code, achieving cluster-based horizontal expansion to quickly and flexibly expand the distributed photovoltaic power generation spatiotemporal collaborative prediction system for complex terrain. In practical applications, the above modules can be set in the same device or different devices, or they can be set in virtual devices, such as service instances in a cloud server.

[0017] Example 1, please refer to Figure 1 As shown in this embodiment, a spatiotemporal collaborative prediction system for distributed photovoltaic power generation in complex terrain is described. The system includes: The multi-source data acquisition module is used to acquire raw data in real time based on the sensor network and to adjust the acquisition time window to obtain the raw dataset. Furthermore, the steps of acquiring raw data in real time based on sensor networks and adjusting the acquisition time window include: S1.1: Based on the sensor network, raw data is collected in real time to obtain the sensor dataset, which includes meteorological satellite data, ground meteorological station data, terrain elevation data and photovoltaic unit operation data; It should be explained that meteorological satellite data includes, but is not limited to, data on irradiance and cloud cover; ground meteorological station data includes, but is not limited to, data on ambient temperature and humidity, ambient wind speed and ambient wind direction, and each meteorological station has a unique identifier and location coordinates; terrain elevation data includes, but is not limited to, data on slope, aspect and altitude; photovoltaic unit operation data includes, but is not limited to, historical power generation and data on voltage, current and equipment temperature related to equipment operation, and each photovoltaic unit has a unique ID and location coordinates. S1.2: Execute the data collection task in step S1.1 based on the preset data collection time; It should be explained that the preset collection time is manually set and entered into the system, for example, the preset collection time is five minutes each time; S1.3: Integrate the acquisition timestamps and spatial coordinates of all data items in the sensor dataset to obtain the original dataset; S1.4: Store the original dataset in the database and output it to the data preprocessing and fusion module; The data preprocessing and fusion module is used to perform preprocessing, feature extraction and spatiotemporal fusion based on the original dataset to obtain a spatiotemporal feature report. Furthermore, the steps of preprocessing, feature extraction, and spatiotemporal fusion based on the original dataset include: S2.1: Preprocessing is performed on the original dataset to obtain an aligned dataset. Preprocessing includes data cleaning, data normalization, and time alignment. The specific preprocessing method is as follows: The missing values ​​in the original dataset are identified and filled using linear interpolation to complete the data cleaning process. The original dataset after data cleaning is normalized using a normalization formula to complete the data normalization process. It should be explained that data normalization is used to normalize data of different dimensions to the range [0, 1]. The original dataset after data normalization is unified to the same timestamp, and missing values ​​are filled using the neighbor interpolation method to complete time alignment; S2.2: Using a convolutional neural network, feature extraction is performed on the meteorological satellite data and topographic elevation data in the aligned dataset to obtain spatial features; Using a long short-term memory network, feature extraction is performed on ground meteorological station data and photovoltaic unit operation data in the aligned dataset to obtain time-series features; S2.3: Use an attention mechanism to perform weighted fusion of spatial and temporal features to obtain a spatiotemporal feature report; It should be explained that the spatiotemporal feature report contains fused feature vectors, with each vector corresponding to a photovoltaic unit; S2.4: Output the spatiotemporal feature report to the spatiotemporal correlation network construction and update module; The spatiotemporal correlation network construction and update module is used to construct and update the spatiotemporal correlation network based on the spatiotemporal feature report to obtain a dynamic graph structure; Furthermore, the steps for constructing and updating the spatiotemporal correlation network based on spatiotemporal feature reports include: S3.1: Based on the database, the initial spatiotemporal correlation network is retrieved, feature data is extracted according to the spatiotemporal feature report to obtain node features, and the node features are input into the spatiotemporal correlation network for parameter calculation to obtain the first spatiotemporal correlation network; It should be explained that the spatiotemporal correlation network is a graph structure; node features include, but are not limited to, data such as altitude, slope, aspect, and irradiance in the spatiotemporal feature report. S3.2: Adjust the edge weights of the first spatiotemporal association network based on real-time meteorological satellite data retrieved from the database, calculate the cloud impact factor based on cloud coverage data in the spatiotemporal feature report, and finally optimize the first spatiotemporal association network using graph structure learning technology to obtain a dynamic graph structure. It should be explained that the dynamic graph structure includes photovoltaic units, edge sets, and weight matrices; S3.3: Output the dynamic graph structure to the model training and co-prediction module; The model training collaborative prediction module is used to train and predict the graph neural network model based on the dynamic graph structure, and obtain a collaborative prediction report. Furthermore, the steps for training and predicting graph neural network models based on dynamic graph structures include: S4.1: Retrieve the pre-trained graph neural network model based on the database and train it according to historical data to obtain the final graph neural network model. The historical data includes the model's historical predicted power generation value and the corresponding actual power generation value. S4.2: Input the dynamic graph structure and node features into the final graph neural network model, and output the predicted power generation value set; S4.3: Calculate the regression sum of squares based on the predicted power generation value set and the actual power generation value to obtain the model evaluation value; S4.4: Package the predicted power generation value set and the model evaluation value to obtain a collaborative prediction report; S4.5: Output the collaborative forecasting report to the risk quantification and assessment module; The risk quantification and assessment module is used to quantify the uncertainty of forecasts based on the collaborative forecast report, assess the risk, and obtain an uncertainty report. Furthermore, based on the collaborative forecasting report, the steps to quantify forecast uncertainty and assess risk include: S5.1: Based on the historical collaborative prediction reports retrieved from the database, a collaborative prediction set is obtained, and the prediction mean, prediction variance, and confidence interval are calculated based on the collaborative prediction set. S5.2: Calculations are performed based on confidence intervals to obtain the risk assessment value and conditional risk value. The specific formula set for the calculation is as follows: Risk assessment values ​​were obtained respectively. Conditional risk value ,in, To predict the mean, Let be the confidence interval. To predict variance, This represents the confidence level of the confidence interval. Confidence level The risk assessment value below, For integration variables The derivative; S5.3: Package the predicted mean, predicted variance, confidence interval, risk assessment value and conditional risk value to obtain the uncertainty report; S5.4: Output the uncertainty report to the reinforcement learning decision-making strategy generation module; The reinforcement learning decision-making strategy generation module is used to learn policies from the policy network based on uncertainty reports and system states, and generate policy regulation reports. Furthermore, the steps for learning policies from the policy network based on uncertainty reports and system states to generate policy regulation reports include: S6.1: Extract the prediction variance and risk assessment value based on the uncertainty report, extract the predicted power generation value set based on the collaborative prediction report, retrieve the state parameters based on the database, and package the prediction variance, risk assessment value, power generation value set and state parameters to obtain the state vector. The state parameters include energy storage capacity and load demand. S6.2: Retrieve control commands from the database as action vectors; It should be explained that the control commands include, but are not limited to, power distribution setpoints and voltage regulation commands; S6.3: Based on the state vector and action vector, and using the near-end policy optimization algorithm, the policy network is trained. After reaching the preset number of iterations, the final policy network is output. S6.4: Input the state vector into the final policy network and output the policy control report; It should be explained that the strategy control report contains a series of control instructions; S6.5: Output the strategy control report to the system instruction distribution and execution module; The system instruction distribution and execution module is used to convert the policy control report into executable instructions, and to distribute and monitor the execution status. Furthermore, the steps of converting the policy control report into executable instructions, distributing them, and monitoring their execution status include: S7.1: Convert the control instructions in the strategy control report into a standard format and output them to the corresponding target device. At the same time, use the confirmation mechanism to trigger retransmission when the system does not receive a confirmation signal. S7.2: Monitor the execution of instructions on the target device. When the monitoring result indicates that the execution has failed, generate a warning report and record the execution failure log. The warning report indicates that the A0** command failed to execute on the target device, and requests that staff proceed to the target device location for inspection as soon as possible. S7.3: Package the instruction output status, monitoring results, instruction output timestamp, and monitoring result timestamp from step S7.1 to obtain the instruction execution report; S7.4: Output the instruction execution report to the reinforcement learning decision policy generation module for policy adjustment of the policy network; In this embodiment, the beneficial effects are achieved by acquiring raw data in real time based on a sensor network and adjusting the acquisition time window to obtain a raw dataset. Preprocessing, feature extraction, and spatiotemporal fusion are then performed on the raw dataset to obtain a spatiotemporal feature report. Based on the spatiotemporal feature report, a spatiotemporal correlation network is constructed and updated to obtain a dynamic graph structure. Based on the dynamic graph structure, a graph neural network model is trained and predicted to obtain a collaborative prediction report. Based on the collaborative prediction report, prediction uncertainty is quantified, and risk is assessed to obtain an uncertainty report. Based on the uncertainty report and system status, a policy network is used for policy learning to generate a policy control report. The policy control report is converted into executable instructions, which are then distributed and their execution status monitored. This enables the system to achieve deep fusion of multi-dimensional environmental and operational data based on a multi-source data acquisition module and a data preprocessing and fusion module, and to extract key spatiotemporal features. This invention maximizes the system's performance in complex terrain and meteorological environments. By combining the spatiotemporal correlation network construction and update module and the model training collaborative prediction module, the system can further capture the dynamic correlation between photovoltaic units, thereby effectively improving prediction accuracy and fundamentally reducing prediction errors. In addition, the spatiotemporal correlation network construction and update module enables the system to maintain extremely high accuracy and reliability under changing environmental conditions. Finally, through the collaborative operation of the risk quantification assessment module, the reinforcement learning decision strategy generation module, and the system instruction distribution and execution module, the system possesses a proactive risk response capability, that is, it actively adjusts the equipment operation mode before extreme conditions occur, maximizing equipment safety. Overall, this invention has significant advantages such as high accuracy of spatiotemporal collaborative prediction, strong adaptability to changing environmental conditions, and good intelligent decision-making mechanism.

[0018] Example 2, please refer to Figure 2 As shown, the parts not described in detail in this embodiment are described in Embodiment 1. A spatiotemporal collaborative prediction method for distributed photovoltaic power generation in complex terrain is provided. The method includes: S1: Real-time acquisition of raw data based on sensor network and adjustment of acquisition time window to obtain raw dataset; S2: Based on the original dataset, preprocessing, feature extraction, and spatiotemporal fusion are performed to obtain a spatiotemporal feature report; S3: Based on the spatiotemporal feature report, the spatiotemporal correlation network is constructed and updated to obtain a dynamic graph structure; S4: Train and predict the graph neural network model based on the dynamic graph structure to obtain a collaborative prediction report; S5: Based on the collaborative forecasting report, quantify the forecast uncertainty and assess the risk to obtain an uncertainty report; S6: Based on the uncertainty report and system state, the policy network is used to learn policies and generate a policy regulation report; S7: Convert policy control reports into executable instructions, and distribute and monitor their execution status.

[0019] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the present invention.

Claims

1. A spatiotemporal collaborative prediction system for distributed photovoltaic power generation in complex terrain, characterized in that, The system includes: a spatiotemporal correlation network construction and update module, a model training and collaborative prediction module, a risk quantification and assessment module, and a reinforcement learning decision-making strategy generation module, wherein: The spatiotemporal correlation network construction and update module is used to construct and update the spatiotemporal correlation network based on the spatiotemporal feature report to obtain a dynamic graph structure; The model training collaborative prediction module is used to train and predict the graph neural network model based on the dynamic graph structure, and obtain a collaborative prediction report. The risk quantification and assessment module is used to quantify the uncertainty of forecasts based on the collaborative forecast report, assess the risk, and obtain an uncertainty report. The reinforcement learning decision-making strategy generation module is used to learn policies from the policy network based on uncertainty reports and system states, and generate policy adjustment reports.

2. The distributed photovoltaic power generation spatiotemporal collaborative prediction system for complex terrain as described in claim 1, characterized in that, The system also includes: a multi-source data acquisition module, a data preprocessing and fusion module, and a system instruction distribution and execution module, wherein: The multi-source data acquisition module is used to acquire raw data in real time based on the sensor network and to adjust the acquisition time window to obtain the raw dataset. The data preprocessing and fusion module is used to perform preprocessing, feature extraction and spatiotemporal fusion based on the original dataset to obtain a spatiotemporal feature report. The system instruction distribution and execution module is used to convert the policy control report into executable instructions and to distribute and monitor the execution status.

3. The distributed photovoltaic power generation spatiotemporal collaborative prediction system for complex terrain as described in claim 2, characterized in that, The steps for real-time acquisition of raw data based on sensor networks and adjustment of the acquisition time window include: S1.1: Based on the sensor network, raw data is collected in real time to obtain the sensor dataset, which includes meteorological satellite data, ground meteorological station data, terrain elevation data and photovoltaic unit operation data; S1.2: Execute the data collection task in step S1.1 based on the preset data collection time; S1.3: Integrate the acquisition timestamps and spatial coordinates of all data items in the sensor dataset to obtain the original dataset; S1.4: Store the original dataset in the database and output it to the data preprocessing and fusion module.

4. The distributed photovoltaic power generation spatiotemporal collaborative prediction system for complex terrain according to claim 3, characterized in that, The steps for preprocessing, feature extraction, and spatiotemporal fusion based on the original dataset include: S2.1: Preprocessing is performed on the original dataset to obtain an aligned dataset. Preprocessing includes data cleaning, data normalization, and time alignment. The specific preprocessing method is as follows: The missing values ​​in the original dataset are identified and filled using linear interpolation to complete the data cleaning process. The original dataset after data cleaning is normalized using a normalization formula to complete the data normalization process. The original dataset after data normalization is unified to the same timestamp, and missing values ​​are filled using the neighbor interpolation method to complete time alignment; S2.2: Using a convolutional neural network, feature extraction is performed on the meteorological satellite data and topographic elevation data in the aligned dataset to obtain spatial features; Using a long short-term memory network, feature extraction is performed on ground meteorological station data and photovoltaic unit operation data in the aligned dataset to obtain time-series features; S2.3: Use an attention mechanism to perform weighted fusion of spatial and temporal features to obtain a spatiotemporal feature report; S2.4: Output the spatiotemporal feature report to the spatiotemporal correlation network construction and update module.

5. A distributed photovoltaic power generation spatiotemporal collaborative prediction system for complex terrain as described in claim 3, characterized in that, The steps for constructing and updating the spatiotemporal correlation network based on spatiotemporal feature reports include: S3.1: Based on the database, the initial spatiotemporal correlation network is retrieved, feature data is extracted according to the spatiotemporal feature report to obtain node features, and the node features are input into the spatiotemporal correlation network for parameter calculation to obtain the first spatiotemporal correlation network; S3.2: Adjust the edge weights of the first spatiotemporal association network based on real-time meteorological satellite data retrieved from the database, calculate the cloud impact factor based on cloud coverage data in the spatiotemporal feature report, and finally optimize the first spatiotemporal association network using graph structure learning technology to obtain a dynamic graph structure. S3.3: Output the dynamic graph structure to the model training and co-prediction module.

6. The distributed photovoltaic power generation spatiotemporal collaborative prediction system for complex terrain according to claim 3, characterized in that, The steps for training and predicting graph neural network models based on dynamic graph structures include: S4.1: Retrieve the pre-trained graph neural network model based on the database and train it according to historical data to obtain the final graph neural network model. The historical data includes the model's historical predicted power generation value and the corresponding actual power generation value. S4.2: Input the dynamic graph structure and node features into the final graph neural network model, and output the predicted power generation value set; S4.3: Calculate the regression sum of squares based on the predicted power generation value set and the actual power generation value to obtain the model evaluation value; S4.4: Package the predicted power generation value set and the model evaluation value to obtain a collaborative prediction report; S4.5: Output the collaborative forecasting report to the risk quantification assessment module.

7. A spatiotemporal collaborative prediction system for distributed photovoltaic power generation in complex terrain according to claim 3, characterized in that, Based on the collaborative forecasting report, the steps for quantifying forecast uncertainty and assessing risk include: S5.1: Based on the historical collaborative prediction reports retrieved from the database, a collaborative prediction set is obtained, and the prediction mean, prediction variance, and confidence interval are calculated based on the collaborative prediction set. S5.2: Calculate based on confidence intervals to obtain the risk assessment value and conditional risk value; S5.3: Package the forecast mean, forecast variance, confidence interval, risk assessment value, and conditional risk value to obtain an uncertainty report; S5.4: Output the uncertainty report to the reinforcement learning decision-making strategy generation module.

8. A spatiotemporal collaborative prediction system for distributed photovoltaic power generation in complex terrain according to claim 7, characterized in that, The steps for learning policies from policy networks based on uncertainty reports and system states to generate policy regulation reports include: S6.1: Extract the prediction variance and risk assessment value based on the uncertainty report, extract the predicted power generation value set based on the collaborative prediction report, retrieve the state parameters based on the database, and package the prediction variance, risk assessment value, power generation value set and state parameters to obtain the state vector. The state parameters include energy storage capacity and load demand. S6.2: Retrieve control commands from the database as action vectors; S6.3: Based on the state vector and action vector, and using the near-end policy optimization algorithm, the policy network is trained. After reaching the preset number of iterations, the final policy network is output. S6.4: Input the state vector into the final policy network and output the policy control report; S6.5: Output the strategy control report to the system instruction distribution and execution module.

9. A spatiotemporal collaborative prediction system for distributed photovoltaic power generation in complex terrain according to claim 8, characterized in that, The steps for converting policy control reports into executable instructions, distributing them, and monitoring their execution status include: S7.1: Convert the control instructions in the strategy control report into a standard format and output them to the corresponding target device. At the same time, use the confirmation mechanism to trigger retransmission when the system does not receive a confirmation signal. S7.2: Monitor the execution of instructions on the target device. When the monitoring result indicates that the execution has failed, generate a warning report and record the execution failure log. The warning report indicates that the A0** command failed to execute on the target device, and requests that staff proceed to the target device location for inspection as soon as possible. S7.3: Package the instruction output status, monitoring results, instruction output timestamp, and monitoring result timestamp from step S7.1 to obtain the instruction execution report; S7.4: Output the instruction execution report to the reinforcement learning decision policy generation module for policy adjustment of the policy network.

10. A spatiotemporal collaborative prediction method for distributed photovoltaic power generation in complex terrain. A spatiotemporal collaborative prediction system for distributed photovoltaic power generation in complex terrain, as described in any one of claims 1-9, is characterized in that... The work includes the following steps: S1: Based on the sensor network, raw data is collected in real time, and the collection time window is adjusted to obtain the raw dataset; S2: Based on the original dataset, preprocessing, feature extraction, and spatiotemporal fusion are performed to obtain a spatiotemporal feature report; S3: Based on the spatiotemporal feature report, the spatiotemporal correlation network is constructed and updated to obtain a dynamic graph structure; S4: Train and predict the graph neural network model based on the dynamic graph structure to obtain a collaborative prediction report; S5: Based on the collaborative forecasting report, quantify the forecast uncertainty and assess the risk to obtain an uncertainty report; S6: Based on the uncertainty report and system state, the policy network is used to learn policies and generate a policy regulation report; S7: Convert policy control reports into executable instructions, and distribute and monitor their execution status.

Images