Low-voltage distributed photovoltaic power output prediction and group control method based on big data model

Abstract

The present invention relates to the field of low-voltage distributed photovoltaic control, and aims to solve the problems caused by its high proportion of access to the distribution network. By collecting historical output, meteorological and geographic information data, after pre-processing, a power prediction model that integrates physics and machine learning is constructed, and the historical data is used to optimize the model hyperparameters to improve the prediction accuracy. Based on this, the photovoltaic output capacity is analyzed, and the power prediction and grid load demand are combined to formulate a control strategy, including the construction of a dynamic simulation model, early warning of reverse overload, etc. Through the group adjustment and group control scheme, the control is proportionally reduced according to the proportion of the user's rated capacity, and the inverter output is adjusted through the electricity consumption information collection system. The present invention can accurately predict photovoltaic output, effectively control, achieve full network power balance in special periods, improve grid stability and clean energy utilization efficiency, and promote the sustainable development of distributed photovoltaics in the power grid.

Description

Technical Field

[0001] This invention relates to the field of low-voltage distributed photovoltaic (PV) regulation technology, and in particular to a method for predicting and controlling low-voltage distributed PV power output based on a big data model. Background Technology

[0002] With the increasing global demand for clean energy, distributed photovoltaic (PV) power is seeing a continuous rise in its integration rate into power distribution networks due to its flexible installation methods and advantages of localized consumption. However, this high integration rate is a double-edged sword; while bringing clean energy benefits, it also brings a host of thorny problems.

[0003] On the one hand, low-voltage distributed photovoltaic systems are characterized by their large number, extremely dispersed distribution, and small installed capacity per household. This makes unified and efficient management and control extremely difficult. Traditional grid management models are ill-suited to this decentralized energy landscape, feeling powerless as if dealing with a disorganized mess.

[0004] On the other hand, it poses a series of severe challenges to the stable operation of the distribution network. The disorderly connection of a large number of distributed photovoltaic systems has led to frequent reverse overload phenomena in distribution transformers, which is like putting an excessive burden on the "heart" of the distribution network; the reverse power flow problem of 10kV has disrupted the originally orderly power flow, making the power flow distribution of the grid intricate; and the high voltage problem at the end of the distribution line is like a hidden danger that may affect the power supply quality at any time and threaten the safe and stable operation of electrical equipment.

[0005] In actual grid operation, special periods (such as peak electricity consumption and power emergency periods during extreme weather) place extremely high demands on the overall grid power balance. These periods are often accompanied by rapid load changes. If the output of low-voltage distributed photovoltaic power cannot be effectively managed, it will not only fail to leverage its clean energy advantages to help stabilize the grid, but will also further exacerbate the risk of grid imbalance.

[0006] Therefore, there is an urgent need for an innovative and effective method that can accurately grasp the output characteristics of low-voltage distributed photovoltaic power generation and implement intelligent group dispatch and control to perfectly match its load demand with the power grid, thereby ensuring the stable operation of the power grid under any operating conditions and achieving the key goal of power balance across the entire grid during special periods. Summary of the Invention

[0007] To address the aforementioned technical issues, this invention provides a low-voltage distributed photovoltaic power prediction and group control method based on a big data model. Through accurate power output prediction and effective group control strategies, it enables the stable operation of low-voltage distributed photovoltaic power in the power grid and allows it to participate in grid peak shaving, thereby achieving the goal of power balance across the entire grid during special periods.

[0008] The present invention provides a low-voltage distributed photovoltaic power output prediction and group control method based on a big data model, comprising the following steps:

[0009] Step 1: Data Collection and Analysis: Collect historical power output data, meteorological data, and geographic information data of low-voltage distributed photovoltaic systems. Preprocess the collected data, including cleaning and normalization operations.

[0010] Step 2: Photovoltaic power prediction model construction: Based on the physical principles of photovoltaic power generation, a photovoltaic power prediction fusion model integrating multiple machine learning regression methods is constructed, considering various influencing factors. The hyperparameters of the photovoltaic power prediction fusion model are fitted and corrected using historical data, including model research and selection, and model training process.

[0011] Step 3: Photovoltaic regulation strategy formulation: Conduct photovoltaic power output capacity analysis, power prediction and regulation demand analysis to determine the photovoltaic regulation strategy. The regulation strategy includes constructing a dynamic simulation model of power output capacity, predicting power using NWP data, analyzing regulation demand and time requirements, formulating a plan based on regional power output capacity indicators, and conducting photovoltaic reverse overload early warning.

[0012] Step 4: Implementation of the group control scheme: Generate a list of control users according to the photovoltaic control strategy, reduce the control according to the proportion of the user's rated capacity, send control commands to the acquisition terminal through the power consumption information acquisition system, then to the protocol converter, and finally adjust the inverter output to achieve group control.

[0013] Furthermore, the meteorological data includes solar radiation intensity, temperature, humidity, air pressure, wind speed, and wind direction; the geographic information data includes the latitude and longitude, altitude, and topography of the power station; the cleaning operation includes removing obviously erroneous data points, supplementing missing data using linear interpolation or mean interpolation methods; the normalization operation includes mapping the data to the [0,1] interval, establishing data interfaces with the distributed photovoltaic power station monitoring system, meteorological department, and geographic information system, and obtaining real-time data at time intervals of one hour or less.

[0014] Furthermore, in the training and optimization of the photovoltaic power prediction fusion model, after comparing various types of models, a hybrid model combining a physical model and machine learning methods was selected. Model training includes data partitioning, training, validation, and testing. During model training, cross-validation technology is used to select the optimal combination of model hyperparameters, and the model is retrained periodically based on newly acquired historical data. At least one of support vector machine, neural network, and decision tree is selected as the regression method in the fusion model. Based on the physical principles of photovoltaic power generation, a multi-parameter physical model is established, including angle, area, shading, irradiance, and temperature. The parameters in the physical model are estimated and optimized using time series analysis methods to construct the photovoltaic power prediction model.

[0015] Furthermore, in the training and optimization of the photovoltaic power prediction model, the introduced neural network models are transformer and linear models. The bagging ensemble learning method is used to integrate multiple base models. The base models are obtained by training different sub-training sets obtained by random sampling with replacement on the training set.

[0016] Furthermore, in the formulation of photovoltaic (PV) regulation strategies, machine learning algorithms are used to establish a model relating PV power output capacity to environmental factors and equipment status, and to calculate PV power output capacity indicators under different conditions in real time, thereby constructing a dynamic simulation model of power output capacity. Power is predicted using NWP data, specifically by using numerical weather prediction data and a trained PV power generation prediction fusion model to predict PV power output for future periods. Regulation demand and time requirements are analyzed, specifically by combining grid load demand forecasts to analyze the amount of regulation power and time requirements that PV power plants need to provide at different times. Schemes are formulated based on regional power output capacity indicators, specifically by decomposing regulation tasks to various regions and users, determining the number of users requiring regulation in each region, and the specific regulation values ​​for each user's equipment, including power reduction ratios and output adjustments. PV reverse overload early warning is implemented, i.e., real-time monitoring of power output during the entire regulation process, and early warning of potential reverse overload situations through analysis of power-related data. Regulation commands are sent to the equipment terminals of each distributed PV user through a communication system, which includes an electricity information acquisition system, acquisition terminals, and protocol converters.

[0017] Furthermore, in formulating photovoltaic control strategies, historical data and real-time monitoring data are used to dynamically assess the output capacity of photovoltaic power plants. The assessment formula is: P out = f(E,T,D,M), where P out Let E be the output capacity of the photovoltaic power station, T be the environmental factor vector, D be the equipment state vector, M be the power station design factor vector, and f be the maintenance factor vector. Let f be the function relationship constructed by the machine learning algorithm.

[0018] Furthermore, in formulating photovoltaic (PV) regulation strategies, the PV power output for future periods is predicted based on numerical weather forecast data and a trained power prediction model. The prediction formula is: P t =g(W t ,P t -1,P t -2,...), where P t Let W be the predicted photovoltaic power at time t. t Let P be the numerical weather forecast data at time t. t-1 ,P t-2 ,... represent the actual photovoltaic power values ​​at the previous time point, the second time point before, and so on, and g represents the functional relationship constructed by the power prediction model.

[0019] Furthermore, in the implementation of the group control and regulation scheme, the operating status and power output of the photovoltaic power station are monitored in real time during the regulation process. The regulation strategy is dynamically adjusted according to the actual situation to ensure that the regulation effect meets the grid operation requirements. The calculation formula for adjusting the regulation proportionally according to the user's rated capacity is as follows: Its P adj The value in P represents the power that the user needs to adjust. user R is the user's current actual power. adj As a regulatory indicator, R total This represents the total rated installed capacity for all users.

[0020] Furthermore, in the implementation of the group control scheme, the operating status and power output of the photovoltaic power station are monitored in real time, and the control strategy is dynamically adjusted according to the actual situation. The adjustment is based on the power deviation rate, which is calculated using the following formula: Where δ is the power deviation rate, P actual For actual power, P target The target power is set, and the control strategy is adjusted when the power deviation rate exceeds the set threshold.

[0021] Furthermore, in the implementation of the group control scheme, the electricity information acquisition system sends control commands to the acquisition terminal. The acquisition terminal and the protocol converter transmit data through wired or wireless communication. The protocol converter converts the control signals into instructions that the inverter can recognize, thereby adjusting the output of the inverter.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0023] 1. Accurate forecasting and scientific decision-making

[0024] By preprocessing multi-source data fusion, influencing factors are comprehensively considered, laying a solid foundation for prediction and improving accuracy. By constructing and optimizing a fusion model and training it based on historical data, a reliable basis for regulation strategies is provided.

[0025] 2. Intelligent regulation and control to stabilize the power grid.

[0026] By comprehensively analyzing multiple factors, power output capacity is assessed in real time, providing precise guidance for regulation; by combining forecasting with demand, precise regulation is achieved, and the communication system ensures the issuance of instructions, enabling group dispatch and control, stabilizing power grid operation, and balancing power during special periods.

[0027] 3. Clean and efficient, sustainable development

[0028] Solving the problem of photovoltaic grid connection, leveraging the advantages of clean energy, and promoting the transformation of the energy structure; optimizing operation, improving energy efficiency, reducing waste and emissions, achieving win-win results in multiple aspects, and contributing to sustainable development. Attached Figure Description

[0029] Figure 1 This is a flowchart of the low-voltage distributed photovoltaic power output prediction and group control method based on big data model of the present invention;

[0030] Figure 2 This is a flowchart of the photovoltaic power prediction model training and optimization process of the present invention;

[0031] Figure 3 This is a flowchart of the photovoltaic group control strategy execution and dynamic adjustment process of the present invention. Detailed Implementation

[0032] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0033] like Figures 1 to 3 As shown, this embodiment of the invention provides a method for low-voltage distributed photovoltaic power output prediction and group control based on a big data model, including the following steps:

[0034] Step 1: Data Collection and Preprocessing

[0035] 1. Establish data connection interfaces with the distributed photovoltaic power station monitoring system, meteorological departments, and geographic information systems to ensure the stability and real-time performance of data transmission. For example, establish a dedicated line connection with the local meteorological station to obtain the latest meteorological data every 30 minutes, including information such as solar radiation intensity, temperature, humidity, air pressure, wind speed, and wind direction; at the same time, communicate with the distributed photovoltaic power station monitoring system through an Ethernet interface to collect historical output data, voltage, current, and other operating parameters of the photovoltaic power station in real time, as well as obtain geographic information data such as the latitude, longitude, altitude, and topography of the power station from the geographic information system.

[0036] 2. During the data cleaning process, a power fluctuation threshold of ±10% of the rated power is set. When the collected power data exceeds this threshold, it is identified as an abnormal data point and removed. For missing solar radiation intensity data, if the data changes relatively smoothly at adjacent time points, linear interpolation is used to supplement it; if the data fluctuates significantly, mean interpolation is used, taking the average value of other normal data points within the same time period of the day as the supplement for missing values.

[0037] 3. During data normalization, the max-min normalization method is used to map various data types to the [0,1] interval. For example, for solar radiation intensity data, after normalization, data of different dimensions can be analyzed and processed on the same scale, providing a good data foundation for subsequent model training. In the field of data processing and analysis, especially in machine learning and statistical analysis, normalization to the [0,1] interval is a common data preprocessing operation. For professionals working in this field, this is a basic concept and a commonly used technique, and they have a certain understanding of its principles and functions. In this sense, it can be considered common knowledge for those in the technical field.

[0038] Step 2: Training and Optimization of Photovoltaic Power Prediction Model

[0039] 1. In the model research phase, a detailed analysis was conducted on the characteristics of traditional time series models (such as ARIMA, STL, MLP), photovoltaic power generation physics models (such as pvlib, module models, inverter models), photovoltaic power generation Transformer-based models (such as TFT, Autoformer, Informer, Flowformer, etc.), and multi-task learning models. Taking a distributed photovoltaic power station with an installed capacity of 100kW as an example, it was found that its output data exhibits certain seasonality and periodicity over time, and is closely related to meteorological factors such as solar radiation intensity and temperature. After comparative testing, a hybrid model was constructed by selecting a photovoltaic power generation integrated physics model as the base model and introducing transformer and linear neural network models as supplements.

[0040] 2. When constructing the integrated physical model, based on the equivalent circuit model of photovoltaic cells, physical parameters such as series resistance, parallel resistance, and diode characteristics of the modules are considered. Combined with actual measured information such as the power station tilt angle, azimuth angle, and module area, a multi-parameter physical model is established. Through time series analysis of historical data, the influence weights of parameters such as irradiance and temperature on power output under different seasons and weather conditions are estimated, and the parameters in the physical model are optimized. For example, during the high-temperature period in summer, temperature has a significant impact on power output; by determining the correction coefficient for the temperature parameter through time series analysis, the prediction accuracy of the model during this period is improved.

[0041] 3. During model training, the historical data (after preprocessing) collected from the power station over the past year was divided into training, validation, and test sets at a ratio of 70%, 20%, and 10%, respectively. The fusion model was trained using the training set, employing stochastic gradient descent to optimize model parameters. The initial learning rate was set to 0.01, and the regularization coefficient was 0.001. After each training cycle, the model performance was evaluated using the validation set, measuring the deviation between the model's predicted and actual values ​​by calculating the root mean square error (RMSE) and mean absolute error (MAE). Based on the validation results, cross-validation techniques were used to adjust the model's hyperparameters. If overfitting was observed, the regularization coefficient was appropriately increased or the learning rate decreased. After multiple iterations of training, when the model's performance on the validation set stabilized and met the preset accuracy requirements (e.g., RMSE less than 5% of rated power), the test set was used for final evaluation. If the performance on the test set also met the requirements, model training was complete; otherwise, the hyperparameters were adjusted, and the model was retrained. New historical data is acquired periodically (e.g., monthly) and added to the training set to retrain the model to adapt to changes in the power plant's operating status and environmental conditions.

[0042] Step 3: Formulation of Photovoltaic Regulation Strategies

[0043] 1. Utilizing historical and real-time monitoring data from the distributed photovoltaic (PV) power station over the past three years, a neural network algorithm was used to establish a relationship model between PV power output and environmental factors (such as different weather types like sunny, cloudy, and rainy days, as well as different seasons and time periods), equipment status (e.g., judging the aging degree of modules by monitoring changes in output current and voltage, and analyzing inverter efficiency based on the inverter's efficiency curve), power station design factors (module tilt angle of 30 degrees, azimuth direction due south, module material of monocrystalline silicon, etc.), and maintenance factors (monthly regular maintenance). For example, on a cloudy and cold winter morning, the model calculated that the PV power station's output capacity was approximately 30% of its rated power.

[0044] 2. Based on numerical weather forecast data (hourly solar radiation intensity and temperature forecasts for the next 72 hours) provided by the meteorological department and a trained power prediction model, the photovoltaic power output for future periods is predicted. Taking a certain day as an example, the predicted solar radiation intensity will gradually increase and the temperature will slowly rise between 9:00 AM and 10:00 AM. Combined with model calculations, the photovoltaic power output during this period is expected to gradually increase from 50kW to 70kW. Considering the grid load demand forecast (80kW for this period), the analysis indicates that the photovoltaic power station needs to provide 10kW of controlled power (80kW - 70kW) during this period, with the control time requirement being real-time adjustment based on power changes between 9:00 AM and 10:00 AM.

[0045] 3. Based on the distributed photovoltaic (PV) distribution in the area where the PV power station is located (there are 50 distributed PV users in the area), user types (residential users, commercial users, etc.), and equipment performance (inverter maximum power point tracking range, efficiency, etc.), the control task is decomposed to each user. For example, if the number of users requiring control in the area is determined to be 20, including 15 residential users and 5 commercial users, for residential users, the power reduction ratio for each user's equipment (inverter) is calculated based on their rated capacity ratio and control targets. For example, if residential user A has a rated capacity of 5kW, the total rated installed capacity of all users participating in the control is 80kW, and the control target is to reduce power by 10kW, then the power reduction required by residential user A is... Power reduction ratio is The electricity consumption information collection system accurately sends control commands to the equipment terminals of each distributed photovoltaic user, enabling group control and dispatch of photovoltaic power stations.

[0046] Step 4: Implementation of the Group Control and Dispatch Scheme

[0047] 1. Based on the control scope (e.g., distributed photovoltaic users within a specific area), control targets (e.g., a reduction of 20kW in total power) and user type profiles (including user rated capacity, industry, etc.) in the control strategy plan, an automatic control user list is generated. For example, in an area containing 100 distributed photovoltaic users, 30 eligible users are selected for the control list based on the control requirements, including 5 industrial users, 10 commercial users, and 15 residential users.

[0048] 2. Based on the user's rated capacity percentage and the ratio of the control target to the rated installed capacity, the power adjustment for each user will be proportionally reduced. Taking commercial user B as an example, its rated capacity is 8kW, and the total rated installed capacity of all users is 120kW. The control target is to reduce power by 20kW. Therefore, the power adjustment required for commercial user B is: The power consumption information collection system sends control commands to the collection terminal. The collection terminal and the protocol converter transmit data wirelessly (such as 4G network) to ensure the timeliness and stability of data transmission.

[0049] 3. The protocol converter accurately converts the received control signals into instructions that the inverter can recognize, thereby regulating the inverter's output. For example, when the protocol converter receives an instruction to reduce power by 1.33kW, it converts it into a power adjustment signal for the inverter, causing the inverter to reduce its output power. During the regulation process, the operating status and power output of the photovoltaic power station are monitored in real time, and the power deviation rate is calculated every 5 minutes. If the calculated power deviation rate is 8% at a certain moment (the set threshold is 5%), exceeding the preset threshold, the regulation strategy is adjusted in a timely manner according to the deviation. This may involve increasing or decreasing the number of users participating in the regulation, or recalculating the regulation value for each user, to ensure that the regulation effect always meets the grid operation requirements and achieves overall grid power balance.

[0050] The low-voltage distributed photovoltaic power output prediction and group control method based on big data model of the present invention can achieve the following technical effects:

[0051] 1. Multi-source data fusion for comprehensive consideration of influencing factors: Data collection includes historical output data, various meteorological data (such as solar radiation intensity, temperature, humidity, air pressure, wind speed, and wind direction), and geographic information data (power station latitude and longitude, altitude, and topography). This comprehensive data reflects various conditions affecting photovoltaic output. For example, different meteorological conditions directly influence photovoltaic power generation efficiency, geographical factors affect sunlight reception and temperature conditions, and historical data reveals the changing patterns of photovoltaic output. This provides a rich and comprehensive information foundation for accurate prediction, avoiding prediction errors caused by missing or incomplete data.

[0052] Advanced model construction enhances prediction accuracy: A hybrid model integrating physical models and machine learning methods is constructed, leveraging the advantages of the physical model's strong interpretability based on principles (relying on geographical, meteorological, and equipment information) and the machine learning model's superior ability to handle complex nonlinear data. Based on a photovoltaic power generation integrated physical model, considering numerous physical parameters (angle, area, shading, irradiance, temperature, etc.), and supplemented and optimized with the expertise of time-series data processing by transformer and linear neural network models, this hybrid model can deeply explore the intrinsic relationships within the data, accurately capture photovoltaic power variation trends, effectively cope with various complex operating conditions, and provide a highly reliable basis for the scientific formulation of subsequent control strategies.

[0053] 2. Precise strategy planning and execution to ensure dynamic balance between supply and demand.

[0054] Real-time dynamic assessment of photovoltaic (PV) power output capacity: Utilizing machine learning algorithms, a multi-factor model is constructed to assess the relationship between PV power output capacity and environmental factors (weather, time, season, etc.), equipment (module aging, inverter efficiency, etc.), power plant design (module orientation, tilt angle, materials, etc.), and maintenance (equipment maintenance frequency, etc.). This model calculates precise PV power output capacity indicators under different conditions in real time. Regardless of environmental changes or equipment fluctuations, the actual power output level of the PV power plant can be accurately grasped, providing crucial data for the precise formulation of control strategies, preventing blind control, and ensuring appropriate and effective regulation.

[0055] Based on supply and demand forecasting, precise matching and control are achieved: Using numerical weather forecasts and trained power prediction models, combined with grid load demand forecasts, the required control power and timing for photovoltaic power plants at different times are meticulously analyzed. During peak and off-peak electricity consumption, targeted control plans can be anticipated and formulated in advance, determining when, where, and which regions and users to control. This ensures dynamic and precise matching between photovoltaic output and grid load demand, maintaining stable grid operation, preventing grid failures caused by supply and demand imbalances, and guaranteeing stable and reliable power supply quality.

[0056] Real-time monitoring and dynamic adjustment optimize control effectiveness and adaptability: Throughout the control process, the operating status and power output of the photovoltaic power station are monitored in real time at high frequency (e.g., every 5 minutes), and the power deviation rate is calculated promptly. Once the power deviation exceeds the set threshold, the control strategy adjustment mechanism is immediately triggered. By flexibly adjusting the number of users participating in control and recalculating user control values ​​(power reduction ratio, output adjustment, etc.), deviations are quickly corrected, ensuring that the control effect always closely matches the actual needs of the power grid operation. This dynamic adjustment mechanism greatly enhances the grid's ability to accommodate distributed photovoltaic power, enabling it to better integrate into the grid operation system, effectively cope with various complex and changing operating conditions, and ensure the safe and stable operation of the power grid.

[0057] 3. Balance grid supply and demand, and enhance system stability and reliability: Effectively solve a series of grid problems caused by the large-scale integration of low-voltage distributed photovoltaic power, such as reverse overload of distribution transformers, reverse power flow at 10kV, and high voltage at the end of distribution lines. During peak electricity consumption periods and extreme weather, photovoltaic output can be precisely controlled to ensure the balance of power supply and demand across the entire grid, prevent the grid from being impacted by fluctuations in photovoltaic output, enhance the stability and reliability of grid operation, and provide a continuous and stable power guarantee for social production and life.

[0058] Optimize clean energy allocation to promote energy structure upgrading and sustainable development: Enable low-voltage distributed photovoltaic power to play a positive role in grid peak shaving, rationally adjust output according to grid load changes, reduce energy waste, and improve the efficiency of clean energy utilization. Promote the widespread application and deep integration of clean energy in the power grid, optimize and upgrade the energy structure towards a clean and low-carbon direction, support the implementation of environmental protection and sustainable development strategies, and contribute to the construction of a green and low-carbon energy system.

[0059] Feasibility analysis of the scalability and sustainable development of the technical solution of the present invention, which is based on a big data model for low-voltage distributed photovoltaic power output prediction and group dispatch and control:

[0060] I. Scalability

[0061] (1) Model level

[0062] Flexible and easily scalable architecture: The hybrid model architecture, which integrates physics and machine learning, can flexibly incorporate new physical factors or advanced neural network structures. For example, as research progresses or new technologies emerge, the model can be quickly adjusted to adapt to low-voltage distributed photovoltaic systems of different scales and scenarios, ensuring the accuracy of power prediction.

[0063] Data-driven optimization: Models trained on and regularly updated from a large amount of historical data can automatically adapt to changes in power plant operation and the environment. For example, when the system scales up or equipment is upgraded, new data is incorporated into the training set, prompting the model to adaptively adjust parameters, maintain predictive ability, and reduce expansion costs.

[0064] (2) System Architecture Level

[0065] A universal data interface facilitates scalability: Data interfaces established with multiple systems are based on common standards, making it easy to access more data sources. Whether adding new meteorological stations or integrating data from different power plants, this can be achieved through extended interfaces, supporting the diversification of system geographical coverage and data sources.

[0066] Distributed management of control strategies: In group control schemes, control strategies can be implemented in a layered and distributed manner. When the scale expands, control centers can be added or user groups can be subdivided. For example, after the development of urban distributed photovoltaic power, sub-control centers can be established in different regions to work together and ensure efficient operation.

[0067] II. Optimization of Sustainable Development

[0068] (1) Energy and Environment

[0069] Power forecasting optimization improves energy efficiency: Continuously optimizes the power forecasting model to accurately match power generation and load demand, reduce curtailment of solar power, and improve energy utilization. Furthermore, it can automatically adjust forecasting parameters based on factors such as climate change and component aging to adapt to environmental changes and achieve long-term high-efficiency power generation.

[0070] Regulation strategies adapt to environmental changes: Regulation strategies take environmental factors into account and can be further optimized based on climate research and environmental monitoring data. For example, seasonal strategies can be developed to adjust output in advance to cope with extreme weather, ensuring system-environment coordination, reducing negative impacts, and promoting sustainable development.

[0071] (2) Technological and industry adaptability

[0072] Keep pace with technological trends and upgrades: Involving multiple advanced technologies, the system can be upgraded as technology advances. For example, by leveraging advancements in big data processing and communication technologies, the depth of data mining and the efficiency of control command transmission can be improved, maintaining the system's advanced nature and meeting the future needs of the energy industry.

[0073] Integrating Industry Standards and Specifications: The system is optimized based on standards and specifications in the development of the distributed photovoltaic industry. This includes adhering to standards related to data interfaces, power prediction accuracy, and the security of control strategies to ensure compatibility and compliance, promoting healthy industry development and system interconnectivity.

[0074] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting and controlling low-voltage distributed photovoltaic power output based on a big data model, characterized in that, Includes the following steps: Step 1: Data Collection and Analysis: Collect historical power output data, meteorological data, and geographic information data of low-voltage distributed photovoltaic systems. Preprocess the collected data, including cleaning and normalization operations. Step 2: Photovoltaic power prediction model construction: Based on the physical principles of photovoltaic power generation, a photovoltaic power prediction fusion model integrating multiple machine learning regression methods is constructed, considering various influencing factors. The hyperparameters of the photovoltaic power prediction fusion model are fitted and corrected using historical data, including model research and selection, and model training process. Step 3: Photovoltaic regulation strategy formulation: Conduct photovoltaic power output capacity analysis, power prediction and regulation demand analysis to determine the photovoltaic regulation strategy. The regulation strategy includes constructing a dynamic simulation model of power output capacity, predicting power using NWP data, analyzing regulation demand and time requirements, formulating a plan based on regional power output capacity indicators, and conducting photovoltaic reverse overload early warning. Step 4: Implementation of the group control scheme: Generate a list of control users according to the photovoltaic control strategy, reduce the control according to the proportion of the user's rated capacity, send control commands to the acquisition terminal through the power consumption information acquisition system, then to the protocol converter, and finally adjust the inverter output to achieve group control. In the implementation of the group control and regulation scheme, the operating status and power output of the photovoltaic power station are monitored in real time during the regulation process. The regulation strategy is dynamically adjusted according to the actual situation to ensure that the regulation effect meets the grid operation requirements. The calculation formula for adjusting the regulation proportionally according to the user's rated capacity is as follows: ,That The value in the middle represents the power that the user needs to adjust. The user's current actual power. As a regulatory indicator, The sum of the rated installed capacity for all users; In the implementation of the group control and regulation scheme, the operating status and power output of the photovoltaic power station are monitored in real time, and the control strategy is dynamically adjusted according to the actual situation. The adjustment is based on the power deviation rate, and the formula for calculating the power deviation rate is: ,That The middle value represents the power deviation rate. This is the actual power. The target power is set, and the control strategy is adjusted when the power deviation rate exceeds the set threshold.

2. The method for low-voltage distributed photovoltaic power output prediction and group control based on a big data model as described in claim 1, characterized in that, The meteorological data includes solar radiation intensity, temperature, humidity, air pressure, wind speed, and wind direction. The geographic information data includes the latitude and longitude, altitude, and topography of the power station. The cleaning operation includes removing obviously erroneous data points and supplementing missing data using linear interpolation or mean interpolation. The normalization operation includes mapping the data to the [0,1] interval, establishing data interfaces with the distributed photovoltaic power station monitoring system, meteorological department, and geographic information system, and obtaining real-time data at time intervals of one hour or less.

3. The method for low-voltage distributed photovoltaic power output prediction and group control based on a big data model as described in claim 1, characterized in that, In the training and optimization of the photovoltaic power prediction fusion model, after comparing various types of models, a hybrid model combining a physical model and machine learning methods was selected. Model training includes data partitioning, training, validation, and testing. During model training, cross-validation technology is used to select the optimal combination of model hyperparameters, and the model is retrained periodically based on newly acquired historical data. At least one of support vector machine, neural network, and decision tree is selected as the regression method in the fusion model. Based on the physical principles of photovoltaic power generation, a multi-parameter physical model is established, including angle, area, shading, irradiance, and temperature. Time series analysis is used to estimate and optimize the parameters in the physical model to construct the photovoltaic power prediction model.

4. The method for low-voltage distributed photovoltaic power output prediction and group dispatch and control based on a big data model as described in claim 3, characterized in that, In the training and optimization of the photovoltaic power prediction model, the introduced neural network models are transformer and linear models. The bagging ensemble learning method is used to integrate multiple base models. The base models are obtained by training different sub-training sets obtained by random sampling with replacement on the training set.

5. The method for low-voltage distributed photovoltaic power output prediction and group control based on a big data model as described in claim 1, characterized in that, In formulating photovoltaic (PV) regulation strategies, machine learning algorithms are used to establish a model relating PV power output capacity to environmental factors and equipment status, and to calculate PV power output capacity indicators under different conditions in real time, thereby constructing a dynamic simulation model of power output capacity. Power is predicted using NWP data, specifically by using numerical weather forecast data and a trained PV power generation prediction fusion model to predict PV power output for future periods. Regulation demand and time requirements are analyzed, specifically by combining grid load demand forecasts to analyze the amount of regulation power and regulation time requirements that PV power plants need to provide in different time periods. The plan is formulated based on regional power output capacity indicators. Specifically, the control tasks are decomposed to various regions and users, the number of users in each region that need to participate in the control is determined, and the specific control values ​​for each user's equipment are determined. The control values ​​include the power reduction ratio and the power output adjustment amount. Photovoltaic reverse overload early warning is carried out, which means that the power output is monitored in real time throughout the entire control process, and the potential reverse overload situation is warned through the analysis of power-related data. Control commands are sent to the equipment terminals of each distributed photovoltaic user through a communication system, which includes an electricity information acquisition system, acquisition terminals, and protocol converters.

6. The method for low-voltage distributed photovoltaic power output prediction and group dispatch and control based on a big data model as described in claim 5, characterized in that, In formulating photovoltaic (PV) regulation strategies, historical data and real-time monitoring data are used to dynamically assess the output capacity of PV power plants. The assessment formula is as follows: ,in To enhance the power output capacity of photovoltaic power plants For environmental factors vectors, This is the device state vector. For power plant design factor vectors, To maintain the factor vector, Functional relationships constructed for machine learning algorithms.

7. The method for low-voltage distributed photovoltaic power output prediction and group dispatch and control based on a big data model as described in claim 6, characterized in that, In formulating photovoltaic (PV) regulation strategies, the PV power output for future periods is predicted based on numerical weather forecast data and a trained power prediction model. The prediction formula is as follows: ,in Let be the predicted photovoltaic power at time t. The numerical weather forecast data is at time t. This represents the actual photovoltaic power values ​​for the previous moment, the second moment before, and so on. The functional relationships constructed for the power prediction model.

8. The method for low-voltage distributed photovoltaic power output prediction and group control based on big data model as described in claim 1, characterized in that, In the implementation of the group control scheme, the power information acquisition system sends control commands to the acquisition terminal. The acquisition terminal and the protocol converter transmit data through wired or wireless communication. The protocol converter converts the control signals into instructions that the inverter can recognize, thereby regulating the output of the inverter.

Images