Photovoltaic solar energy generating capacity analysis algorithm

Abstract

The invention relates to the technical field of generating capacity analysis, and discloses a photovoltaic solar generating capacity analysis algorithm. According to the method, through a multi-dimensional feature fusion and screening mechanism, key meteorological, geographical and technical parameters influencing photovoltaic power generation are effectively integrated and screened out, dependence on a large amount of historical data is reduced, and adaptability of the model to different data distributions and new conditions is improved. Meanwhile, a space-time dynamic modeling and self-adaptive learning mechanism is adopted, the dynamic change rule of the photovoltaic power generation capacity along with time and space is captured, and the accuracy and stability of the prediction model on different time scales and space distribution are ensured. In addition, due to application of deep learning model integration and optimization strategies, the robustness and accuracy of prediction results are further improved, computing resource consumption is reduced, and the efficiency and practicability of the model in actual deployment are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power generation analysis, and in particular to a photovoltaic solar power generation analysis algorithm. Background Art

[0002] Photovoltaic power generation is affected by many factors such as solar radiation intensity and meteorological conditions, and is highly random, volatile and intermittent, which poses great challenges to the grid connection and scheduling of photovoltaic power generation. Therefore, accurate prediction of photovoltaic power generation is of great significance for the stable operation and efficient management of the power system.

[0003] Most existing photovoltaic power generation prediction algorithms rely on a large amount of historical data to build models. In practical applications, obtaining sufficient and high-quality historical data often faces many difficulties, such as the lack of data due to the short operation time of photovoltaic power stations in some areas and the failure of data recording equipment. Even if the data is complete, when the data distribution characteristics change (such as the expansion of photovoltaic power stations, aging of components, etc., which lead to changes in power generation characteristics) or new influencing factors (such as the application of new component technology), the model is difficult to adapt quickly and the prediction accuracy drops significantly. For example, due to changes in the surrounding environment of photovoltaic power stations in a certain area, the shadow occlusion situation has increased. The original model trained based on historical data cannot effectively cope with it, and the prediction error has increased significantly. The present invention aims to reduce the dependence on a large amount of historical data and improve the adaptability of the model to different data distributions and new situations by optimizing the data processing and feature selection mechanism. To this end, a photovoltaic solar power generation analysis algorithm is proposed. Summary of the invention

[0004] In view of the deficiencies of the prior art, the present invention provides a photovoltaic solar power generation analysis algorithm to solve the background technical problems.

[0005] To achieve the above object, the present invention provides the following technical solution: a photovoltaic solar power generation analysis algorithm, comprising the following steps: 1. Data collection and preprocessing Step 1.1: Use the API or database provided by Meteonorm and Solargis weather stations to obtain historical meteorological data from the project site and nearby weather stations, including but not limited to the following meteorological and environmental parameters: longitude and latitude, altitude, temperature, turbidity, direct horizontal irradiation, diffuse horizontal irradiation, global horizontal irradiation, humidity, sunshine, climate type and air pressure. At the same time, connect the meteorological sensors at the photovoltaic power station site through the Internet of Things technology to obtain more accurate and high-frequency meteorological data in real time, including but not limited to the following: wind speed, wind direction and cloud thickness.

[0006] Step 1.2: Collect detailed technical parameters of the PV power station, including but not limited to whether the module is bifacial, module array length, module height from the ground, module row spacing, inverter functional factors, inverter temperature range, inverter rated power / installed power ratio, inverter system utilization, inverter system maximum voltage, and reflectivity.

[0007] Step 1.3: Determine other factors that affect power generation, including but not limited to DC cable and AC cable loss coefficients, module dirt loss coefficients, module aging coefficients, module LID (light attenuation) coefficients, module power tolerances, module mismatch rates, module back ventilation parameters, module current safety factors, module transmission coefficients, and module back mismatch rates.

[0008] 2. Feature selection and construction Step 2.1: Based on domain knowledge and correlation analysis, select characteristic variables that have a significant impact on power generation forecasting.

[0009] Step 2.2: Construct feature engineering to generate new features through mathematical transformation or combination to improve the predictive ability of the model.

[0010] 3. Model training and optimization Step 3.1: Select a machine learning or deep learning algorithm to build a power generation prediction model.

[0011] Step 3.2: Divide the preprocessed data into training set and test set, and use the training set data to train the model.

[0012] Step 3.3: Optimize the model parameters through cross-validation and grid search techniques to improve the generalization ability and prediction accuracy of the model.

[0013] Step 3.4: Introduce regularization and ensemble learning techniques to reduce the risk of model overfitting.

[0014] 4. Power generation forecast and result analysis Step 4.1: Use the optimized model to predict the power generation of the project site in the future, with the prediction results accurate to the hourly level.

[0015] Step 4.2: Based on the forecast results, provide a scientific basis for investment and financing decisions of photovoltaic projects, including but not limited to power generation forecast, cost-benefit analysis and risk assessment.

[0016] Preferably, in step 1.1, missing values ​​in the meteorological data are filled using a time series-based interpolation method. Specifically, the missing values ​​are estimated based on the linear relationship or weighted average of the data at previous and subsequent moments. To ensure data quality, meteorological data from different sources are cross-validated and quality assessed.

[0017] Preferably, in step 2.1, the Pearson correlation coefficient and mutual information statistical method are specifically used to screen the characteristic variables that have a significant impact on the power generation prediction.

[0018] Preferably, in step 2.2, feature engineering includes mathematical transformation and feature combination, and constructs composite features, which can better reflect the changes in power generation performance of the components in actual operation, and expands the polynomial features to capture possible nonlinear relationships. By comparing the model performance of different feature engineering methods on the training set and the test set, the optimal feature engineering solution is selected.

[0019] Preferably, in step 3.2, the preprocessed data is specifically divided into a training set and a test set in a ratio of 70%:30% to ensure that the training set and the test set have similarity in data distribution. During the training process, an early stopping method is used to prevent the model from overfitting, that is, when the performance of the model on the validation set does not improve for several consecutive rounds, the training is stopped.

[0020] Preferably, in step 3.4, for the neural network model, L1 and L2 regularization methods are used, regularization terms are added to the loss function, the size of the model parameters are constrained, and at the same time, ensemble learning technology is used to train multiple base models in parallel, and their prediction results are averaged or voted to reduce the model variance.

[0021] Compared with the prior art, the present invention has the following beneficial effects: The present invention effectively integrates and screens out the key meteorological, geographical and technical parameters that affect photovoltaic power generation through a multi-dimensional feature fusion and screening mechanism, reduces the reliance on a large amount of historical data, and improves the adaptability of the model to different data distributions and new situations. At the same time, spatiotemporal dynamic modeling and adaptive learning mechanisms are used to capture the dynamic changes in photovoltaic power generation over time and space, ensuring the accuracy and stability of the prediction model at different time scales and spatial distributions. In addition, the application of deep learning model integration and optimization strategies further improves the robustness and accuracy of the prediction results, reduces computing resource consumption, and improves the efficiency and practicality of the model in actual deployment. In summary, the present invention can significantly improve the accuracy and adaptability of photovoltaic power generation prediction, provide a scientific basis for investment and financing decisions of photovoltaic projects, and help promote the sustainable development of the photovoltaic power generation industry.

[0022] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flowchart of the photovoltaic solar power generation analysis algorithm of the present invention. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this technical field without creative work are within the scope of protection of the present invention.

[0025] See also Figure 1 , a photovoltaic solar power generation analysis algorithm, comprising the following steps: 1. Data collection and preprocessing Step 1.1: Use the API or database provided by Meteonorm and Solargis weather stations to obtain historical meteorological data of the project site and nearby meteorological stations, including latitude and longitude, altitude, temperature, turbidity, direct horizontal irradiation, diffuse horizontal irradiation, global horizontal irradiation, humidity, sunshine, climate type and air pressure. At the same time, connect the meteorological sensors at the photovoltaic power station site through the Internet of Things technology to obtain more accurate and high-frequency meteorological data in real time, such as wind speed, wind direction, cloud thickness, etc. Specifically: For missing values ​​in meteorological data, we use time series-based interpolation methods to fill in, such as estimating based on the linear relationship or weighted average of the data before and after; to ensure data quality, we cross-validate and evaluate the quality of meteorological data from different sources. For example, we compare the difference in irradiation data between Meteonorm and Solargis at the same time and place. If the difference exceeds a certain threshold, we further analyze the cause and perform data correction. We monitor and preprocess the data collected in real time to remove obviously erroneous data points.

[0026] Step 1.2: Collect detailed technical parameters of the PV power station to determine whether the module is bifacial, the module array length, the module height from the ground, the module row spacing, the inverter function factors, the inverter temperature range, the inverter rated power / installed power ratio, the inverter system utilization, the inverter system maximum voltage, reflectivity, etc. Specifically: Establish a PV power station technical parameter database, classify and manage the parameters of different types of components and inverters, and facilitate data query and update. At the same time, maintain close cooperation with manufacturers to obtain the technical parameters of new equipment in a timely manner to ensure the timeliness of data.

[0027] Step 1.3: Determine other factors that affect power generation, including DC cable and AC cable loss coefficient, component dirt loss coefficient, component aging coefficient, component LID (light attenuation) coefficient, component power tolerance, component mismatch rate, component back ventilation parameters, component current safety factor, component transmission coefficient, and component back mismatch rate, etc. Specifically: Regularly inspect and perform performance tests on components and cables, and update relevant parameters based on the test results. For example, by measuring the output power of components under different lighting conditions, the actual change in component power tolerance can be calculated; thermal imaging technology can be used to detect cable temperature distribution and evaluate changes in cable loss coefficient.

[0028] The conditions required for the photovoltaic power generation prediction model algorithm are shown in Table 1: Table 1 2. Feature selection and construction Step 2.1: Based on domain knowledge and correlation analysis, select the characteristic variables that have a significant impact on power generation forecast. Specifically: Statistical methods such as Pearson correlation coefficient and mutual information are used to screen characteristic variables that have a significant impact on power generation prediction. The Pearson correlation coefficient between each feature and power generation is calculated. Features with an absolute value greater than 0.5 are considered highly correlated and retained for subsequent analysis. At the same time, the mutual information value is calculated to measure the nonlinear dependency between the feature and power generation, and features with larger mutual information values ​​are selected. For example, the analysis found that the Pearson correlation coefficient between direct horizontal irradiation and power generation is 0.8, and the mutual information value is also high, indicating that this feature is crucial to power generation prediction.

[0029] Step 2.2: Construct feature engineering to generate new features through mathematical transformation or combination to improve the predictive ability of the model. Specifically: Feature engineering includes mathematical transformation (logarithmic transformation, Box-Cox transformation, etc.) and feature combination (constructing composite features, polynomial features, etc.). For example, logarithmic transformation of radiation data can make the data closer to normal distribution and improve the model's ability to fit the data. Construct composite features, such as multiplying the component temperature by the radiation to obtain the "temperature-radiation" composite feature, which can better reflect the changes in the power generation performance of the component in actual operation. Perform polynomial feature expansion, such as adding the square of the height of the component from the ground as a new feature to the model to capture possible nonlinear relationships. By comparing the model performance (such as accuracy, mean square error, etc.) of different feature engineering methods on the training set and test set, the optimal feature engineering solution is selected.

[0030] 3. Model training and optimization Step 3.1: Select a machine learning or deep learning algorithm (such as random forest, gradient boosting tree, neural network, etc.) to build a power generation prediction model. Specifically: The random forest algorithm has the advantages of being less prone to overfitting and highly interpretable when processing high-dimensional feature data, and is suitable for preliminary analysis of feature importance; the gradient boosting tree performs well in improving model accuracy and can optimize the random forest model; neural networks are good at capturing complex nonlinear relationships and have potential for processing complex mapping relationships between photovoltaic power generation and multiple factors.

[0031] Step 3.2: Divide the preprocessed data into training set and test set, and use the training set data to train the model. Specifically: The preprocessed data is divided into a training set and a test set in a ratio of 70%:30%, ensuring that the training set and the test set have similar data distribution. Use the training set data to train the model. For the random forest and gradient boosting tree models, set reasonable parameters such as the number of trees and tree depth; for the neural network model, set parameters such as the number of training rounds and batch size. During the training process, the Early Stopping method is used to prevent the model from overfitting, that is, when the performance of the model on the validation set has not improved for several consecutive rounds, stop training. At the same time, record the model performance indicators (such as accuracy, loss value, etc.) during the training process, draw a learning curve, and observe the stability and convergence of the model learning process.

[0032] Step 3.3: Optimize model parameters through cross-validation, grid search and other techniques to improve the generalization ability and prediction accuracy of the model. Specifically: The model parameters are optimized through cross-validation (such as five-fold cross-validation) and grid search technology. For the random forest model, the optimal combination of parameters such as the number of trees, tree depth, and feature sampling ratio is searched; for the neural network model, parameters such as learning rate, number of hidden layer nodes, and regularization coefficient are searched. During the grid search process, the performance indicators of the model on the validation set (such as accuracy, F1 value, etc.) are calculated under different parameter combinations, and the parameter combination with the best performance is selected as the final model parameters.

[0033] Step 3.4: Introduce regularization, ensemble learning and other techniques to reduce the risk of model overfitting. Specifically: Regularization technology is introduced to reduce the risk of model overfitting. For neural network models, L1 and L2 regularization methods are used to add regularization terms to the loss function to constrain the size of model parameters. At the same time, ensemble learning techniques such as Bagging and Boosting are used. Bagging trains multiple base models in parallel (such as multiple decision trees in a random forest) and averages or votes on their prediction results to reduce model variance; Boosting iteratively trains multiple weak classifiers (such as decision trees in a gradient boosting tree), adjusts sample weights according to the prediction error of the previous model, and gradually improves model accuracy. Through experimental comparison, select appropriate ensemble learning methods and parameter settings to improve the robustness and prediction accuracy of the model.

[0034] 4. Power generation forecast and result analysis Step 4.1: Use the optimized model to predict the power generation of the project site in the future. The prediction results are accurate to the hourly level. Specifically: During the prediction process, the latest meteorological data and photovoltaic power station operating parameters are obtained in real time and input into the model for dynamic prediction. The power generation is predicted for different seasons, different weather conditions (such as sunny, cloudy, overcast, rainy, etc.) and different types of photovoltaic power stations (such as ground centralized power stations and rooftop distributed power stations), and the stability and accuracy of the prediction results are analyzed. By comparing the prediction errors under different scenarios, the adaptability and reliability of the model are evaluated.

[0035] Step 4.2: Based on the forecast results, provide a scientific basis for investment and financing decisions of photovoltaic projects, including power generation forecast, cost-benefit analysis, risk assessment, etc. Specifically: Based on the forecast results, a detailed cost-benefit analysis is conducted. Consider the initial investment cost (including component procurement, installation and commissioning, inverter purchase, etc.), operation and maintenance costs (such as regular inspections, component cleaning, equipment maintenance, etc.) and power generation revenue (calculated based on the forecasted power generation and local real-time electricity prices). A cost-benefit evaluation model is established to calculate the economic benefit indicators of the project in different time periods.

[0036] The present invention has the following effects: 1. Multi-dimensional feature fusion and intelligent screening The present invention proposes a multi-dimensional feature fusion and intelligent screening mechanism, which can automatically integrate and screen the key meteorological, geographical and technical parameters that affect photovoltaic power generation. The mechanism uses advanced feature selection algorithms, such as recursive feature elimination (RFE), model-based feature selection, etc. In the process of training the model, the algorithm repeatedly builds the model and evaluates the importance of each feature, gradually eliminating the least important features until the predetermined number of features or performance indicators are reached. For example, starting from the initial dozens of features, one or more features that contribute the least to the model are deleted in each iteration, and the accuracy of the model on the validation set is observed at the same time. When the accuracy no longer improves or begins to decline, the iteration is stopped and the feature subset at this time is retained. In practical applications, the important features are preliminarily screened out using statistical methods such as the Pearson correlation coefficient and mutual information, and then the feature subset is further optimized by the RFE algorithm to ensure that the selected features can effectively represent the factors affecting photovoltaic power generation and reduce the complexity of the model.

[0037] 2. Spatiotemporal dynamic modeling and adaptive learning In view of the fact that photovoltaic power generation is affected by both time (such as day and night changes, seasonal changes) and space (such as geographical location, topography), the present invention designs a spatiotemporal dynamic modeling method. This method combines time series analysis and spatial interpolation technology, and can capture the dynamic changes of photovoltaic power generation over time and space. In terms of time series analysis, the autoregressive moving average (ARIMA) model or seasonal decomposition model is used to analyze the historical data of photovoltaic power generation, extract information such as time trends and seasonal cycles, and predict the power generation trend at future time points. In terms of spatial interpolation technology, according to the geographical location distribution and meteorological data of the project site and surrounding meteorological stations, the meteorological data (such as irradiation, temperature, etc.) of the unmonitored area are interpolated and calculated to obtain a spatially continuous meteorological data field. The irradiation in the area is estimated, thereby providing accurate meteorological input data for photovoltaic power stations in different geographical locations, and realizing modeling in the spatial dimension.

[0038] At the same time, through the adaptive learning mechanism, the model can automatically adjust parameters according to changes in historical data, ensuring the accuracy and stability of the prediction model at different time scales and spatial distributions. When testing photovoltaic power stations in multiple different regions, whether they are large centralized power stations in plains or small distributed power stations in mountainous areas, the model can maintain high prediction accuracy in different seasons and weather conditions.

[0039] 3. Deep learning model integration and optimization The present invention adopts a deep learning model integration strategy to integrate the output results of multiple deep learning models (such as convolutional neural network CNN, long short-term memory network LSTM, attention mechanism network Attention-based Networks, etc.), and improves the robustness and accuracy of the prediction results by weighted average, voting or stacking. In the voting integration method, for classification problems (such as judging the impact of weather types on power generation), the majority voting rule is adopted, that is, the category with the most appearances in the prediction results of multiple models is taken as the final result; for regression problems (such as power generation prediction values), average voting or weighted average voting can be adopted. In the stacking integration method, the outputs of multiple base models are input as new features into a meta-model (such as a simple linear regression model or a neural network) for re-training, and the meta-model learns the relationship between the outputs of the base models to obtain the final prediction results.

[0040] In addition, model optimization techniques such as hyperparameter tuning, model pruning, and quantization are introduced. Hyperparameter tuning techniques such as grid search, random search, or Bayesian optimization methods are introduced to find the optimal hyperparameter combination. Taking the LSTM model as an example, the hyperparameters include the number of hidden layers, learning rate, time step, etc. All possible hyperparameter combinations are traversed through grid search, and the model performance is evaluated on the validation set to find the hyperparameter settings that optimize the model performance. Model pruning technology is used to remove unimportant connections or neurons in the neural network to reduce the complexity of the model. For example, by calculating the importance indicators of neurons (such as the absolute value of weights, activation value distribution, etc.), neurons with low importance and their connections are deleted. Without significantly affecting the accuracy of the model, model pruning can reduce the model size by 30%-50%, reduce computing resource consumption, and improve the model's deployment capabilities in resource-constrained environments. At the same time, quantization technology is used to compress model parameters into low-precision data types (such as from 32-bit floating point numbers to 8-bit integers), further reducing the model storage space and calculation amount, accelerating the model inference process, reducing computing resource consumption, and improving the efficiency and practicality of the model in actual deployment.

Claims

1. A photovoltaic solar power generation analysis algorithm, characterized in that: The following steps are involved:

1. Data collection and preprocessing Step 1.1: Use the API or database provided by Meteonorm and Solargis weather stations to obtain historical meteorological data of the project site and nearby meteorological stations, including but not limited to the following meteorological and environmental parameters: longitude and latitude, altitude, temperature, turbidity, direct horizontal irradiation, diffuse horizontal irradiation, global horizontal irradiation, humidity, sunshine, climate type and air pressure. At the same time, connect the meteorological sensors at the photovoltaic power station site through the Internet of Things technology to obtain more accurate and high-frequency meteorological data in real time, including but not limited to the following: wind speed, wind direction and cloud thickness; Step 1.2: Collect detailed technical parameters of the PV power station, including but not limited to whether the module is bifacial, module array length, module height from the ground, module row spacing, inverter function factors, inverter temperature range, inverter rated power / installed power ratio, inverter system utilization, inverter system maximum voltage, and reflectivity; Step 1.3: Determine other factors that affect power generation, including but not limited to loss coefficients of DC cables and AC cables, module dirt loss coefficients, module aging coefficients, module LID (light attenuation) coefficients, module power tolerances, module mismatch rates, module back ventilation parameters, module current safety factors, module transmission coefficients, and module back mismatch rates; 2. Feature selection and construction Step 2.1: Based on domain knowledge and correlation analysis, select characteristic variables that have a significant impact on power generation forecast; Step 2.2: Construct feature engineering to generate new features through mathematical transformation or combination to improve the predictive ability of the model; 3. Model training and optimization Step 3.1: Select machine learning or deep learning algorithm to build a power generation prediction model; Step 3.2: Divide the preprocessed data into training set and test set, and use the training set data to train the model; Step 3.3: Optimize model parameters through cross-validation and grid search techniques to improve the generalization ability and prediction accuracy of the model; Step 3.4: Introduce regularization and ensemble learning techniques to reduce the risk of model overfitting; 4. Power generation forecast and result analysis Step 4.1: Use the optimized model to predict the power generation of the project site in the future, with the prediction results accurate to the hourly level; Step 4.2: Based on the forecast results, provide a scientific basis for investment and financing decisions of photovoltaic projects, including but not limited to power generation forecast, cost-benefit analysis and risk assessment.

2. A photovoltaic solar power generation analysis algorithm according to claim 1, characterized in that: In step 1.1, missing values ​​in meteorological data are filled by time series-based interpolation. Specifically, they are estimated based on the linear relationship or weighted average of the data at previous and subsequent moments. To ensure data quality, meteorological data from different sources are cross-validated and quality assessed.

3. A photovoltaic solar power generation analysis algorithm according to claim 1, characterized in that: In step 2.1, the Pearson correlation coefficient and mutual information statistical method are specifically used to screen the characteristic variables that have a significant impact on the power generation prediction.

4. A photovoltaic solar power generation analysis algorithm according to claim 1, characterized in that: In step 2.2, feature engineering includes mathematical transformation and feature combination, and constructs composite features, which can better reflect the changes in the power generation performance of the components in actual operation, and expand the polynomial features to capture possible nonlinear relationships. By comparing the model performance of different feature engineering methods on the training set and the test set, the optimal feature engineering solution is selected.

5. The photovoltaic solar power generation analysis algorithm according to claim 1, characterized in that: In step 3.2, the preprocessed data is specifically divided into a training set and a test set in a ratio of 70%:30% to ensure that the training set and the test set have similar data distribution. During the training process, the early stopping method is used to prevent the model from overfitting, that is, when the performance of the model on the validation set does not improve for several consecutive rounds, the training is stopped.

6. A photovoltaic solar power generation analysis algorithm according to claim 1, characterized in that: In step 3.4, for the neural network model, L1 and L2 regularization methods are used, and regularization terms are added to the loss function to constrain the size of the model parameters. At the same time, ensemble learning technology is used to train multiple base models in parallel, and their prediction results are averaged or voted to reduce the model variance.

Images