Power Prediction Method and System for String Photovoltaic Inverters

Abstract

This invention discloses a power prediction method and system for string photovoltaic inverters, relating to the field of photovoltaic inverter technology. This invention addresses the problems of large differences in the dimensions of original data and asynchronous timing by processing dynamic anomaly data, aligning multiple sources with time series, and constructing multi-dimensional features. The generated fused feature vector provides high-quality input for modeling, laying the foundation for prediction accuracy. A dynamic response model is constructed by combining physical principles and data-driven approaches, and targeted calibration is achieved through a multi-condition feature library to accurately capture the impact of transient changes and non-ideal factors. The dynamic calibration mechanism of the efficiency mapping model further improves the accuracy of power assessment, solving the problem of poor adaptability of traditional models under different operating conditions. Ultra-short-term, short-term, and medium-term prediction sub-models are constructed and dynamically fused. Combined with error analysis and uncertainty quantification, multi-confidence prediction results are output, meeting the needs of different time scales such as grid dispatching and planning, and improving photovoltaic absorption and grid stability.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic inverter technology, and in particular to a method and system for predicting the power of string photovoltaic inverters. Background Technology

[0002] String photovoltaic (PV) inverters are widely used due to their high flexibility, but their output power fluctuates significantly due to environmental factors such as irradiance and temperature, as well as non-ideal factors such as component aging and shading, posing challenges to grid dispatch and absorption. Chinese patent application CN106156466A discloses a method for plotting a predicted output power curve for a PV inverter, including the following steps: S1. Selecting an initial reference curve; S2. Correcting the reference curve based on the measured actual output power of the PV inverter on the current day and the actual output power of the initial reference curve on the corresponding day; S3. Obtaining the current time, the actual output power of the PV inverter, and the power of the reference curve, inputting prediction coefficients, and calculating the output power at the next time moment; S4. Connecting the calculated predicted output power to form a predicted output power curve; S5. Determining whether the current time is between a first preset time and a second preset time; if so, executing step S3 after the first preset time period. This method, by correcting the reference curve and introducing prediction coefficients to calculate the output power at the next time moment, makes the PV inverter output power curve more consistent with the actual output power curve.

[0003] While the aforementioned patents improve the fit of the photovoltaic inverter output power curve, the following problems still exist:

[0004] 1. The correction is based solely on the overall output power of the inverter, ignoring the differences in power characteristics of each photovoltaic string due to different geographical locations, module aging levels, and shading conditions, making it difficult to accurately reflect the distributed characteristics of string inverters;

[0005] 2. The baseline curve correction relies on historical data and cannot effectively capture transient power fluctuations caused by non-ideal factors such as sudden changes in irradiance and high-temperature aging. It lacks a dynamic adjustment mechanism, resulting in significant errors under extreme operating conditions.

[0006] 3. The time scale is singular, only providing short-term power forecasts for adjacent moments, which cannot meet the diverse needs of real-time grid dispatch and intraday planning. Furthermore, it does not involve the quantification of the uncertainty of the forecast results, making it difficult to support the assessment of decision reliability. Summary of the Invention

[0007] The purpose of this invention is to provide a power prediction method and system for string photovoltaic inverters. It combines physical principles with data-driven construction of a dynamic response model, achieves targeted calibration through a multi-condition feature library, accurately captures the impact of transient changes and non-ideal factors, constructs and dynamically merges ultra-short-term, short-term, and medium-term prediction sub-models, and outputs multi-confidence prediction results by combining error analysis and uncertainty quantification. This meets the needs of different time scales such as grid dispatching and planning, improves photovoltaic absorption and grid stability, and solves the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] Power prediction methods for string photovoltaic inverters include:

[0010] Photovoltaic inverter parameter acquisition: Real-time acquisition of photovoltaic inverter operating parameters based on distributed sensor network, while acquiring external environmental data, and preprocessing the acquired operating parameters and external environmental data;

[0011] The dynamic response model of the photovoltaic inverter is constructed based on the preprocessed operating parameters and external environmental data, and the transient and steady-state characteristics of the output power of the photovoltaic inverter as key factors change are extracted.

[0012] Photovoltaic inverter power assessment establishes a mapping relationship between the input power and output power of the photovoltaic inverter. Based on the conversion efficiency and internal losses of the photovoltaic inverter at different operating nodes, the actual output power of the photovoltaic inverter is calculated.

[0013] The prediction results are integrated to obtain historical operating parameters of string photovoltaic inverters. Combined with real-time operating parameters and external environmental data, a power prediction model is constructed. The ultra-short-term, short-term and medium-term prediction results are fused over time, and a multi-time-scale collaborative prediction result is generated through a weighted average strategy.

[0014] Furthermore, the photovoltaic inverter parameter acquisition undergoes preprocessing, including:

[0015] Dynamic anomaly data processing: An adaptive threshold model is constructed based on the distribution characteristics of historical data. The deviation between the collected operating parameters and the external environment data is monitored in real time based on the adaptive threshold model. Anomalies are identified based on the data deviation, smoothing is performed, and missing data is imputed or labeled using polynomial fitting.

[0016] Multi-source data timing alignment: A reference time axis is established through a distributed clock synchronization protocol to perform time alignment and synchronization processing on electrical parameters and external environment data;

[0017] Multidimensional feature construction: Based on the preprocessed electrical parameters and external environment data, dynamic trend features, cross-domain correlation features and state assessment features are generated and quantified to generate a fusion feature vector of electrical parameters and external environment data.

[0018] Furthermore, the construction of the dynamic response model for the photovoltaic inverter includes:

[0019] Key variable screening and correlation analysis: Based on the fused feature vector, the correlation degree between each feature and the actual output power of the photovoltaic inverter is calculated. The core input variables are screened based on the preset correlation degree threshold, and the mutual correlation strength between the core input variables is determined. The correlation strength is compared with the preset redundant correlation threshold, and redundant features are eliminated.

[0020] Basic response model construction: Based on the circuit principle of photovoltaic modules, a basic response model is established between the output power of the photovoltaic inverter and the selected core input variables to obtain the basic response law of the output power of the photovoltaic inverter as the core input variables change;

[0021] Dynamic response model optimization: Based on dynamic trend characteristics and cross-domain correlation characteristics, the deviation generated by the basic response model in predicting the output power of the photovoltaic inverter is identified and quantified. The deviation is quantified and used as a correction term to correct the basic response model. The output power value of the dynamic response model of the photovoltaic inverter after correction is used as the final output.

[0022] Furthermore, the construction of the dynamic response model for photovoltaic inverters also includes:

[0023] Collect photovoltaic inverter operation data for different seasons, weather types, and time periods, and combine them with corresponding core input variables, dynamic trend features, and cross-domain correlation features to construct a multi-condition feature library;

[0024] Acquire historical operating data of string photovoltaic inverters, extract operating condition features from the historical operating data, and perform similarity matching with a multi-operating condition feature library to determine different operating conditions in the historical operating data.

[0025] Based on the actual output power of the photovoltaic inverter under different operating conditions in historical operating data, the modified dynamic response model is adaptively verified, and the deviation rate between the output power of the dynamic response model and the actual output power is calculated.

[0026] When the deviation rate exceeds the preset deviation threshold, the core input variables, dynamic trend characteristics and cross-domain correlation characteristics of the corresponding working condition are obtained, the inter-correlation strength between the core input variables is recalculated, and the parameters and correction term weights of the basic response model are adjusted.

[0027] Further, photovoltaic inverter power assessment includes:

[0028] Basic data acquisition: Based on the output power value output by the dynamic response model and combined with the real-time collected operating parameters, the input power of the photovoltaic inverter is determined, and the output voltage, current and operating temperature data of the AC side of the photovoltaic inverter are collected at the same time.

[0029] Constructing an efficiency mapping model: Based on the input power, AC side output voltage, current, and operating temperature data of the photovoltaic inverter, an efficiency mapping model of the photovoltaic inverter is constructed.

[0030] Loss quantification and actual output power calculation: Decompose the internal losses of the photovoltaic inverter, combine the conversion efficiency output by the efficiency mapping model with the quantified loss value, and calculate the actual output power of the photovoltaic inverter.

[0031] Furthermore, photovoltaic inverter power assessment also includes dynamic calibration of mapping relationships:

[0032] Based on the multi-condition feature library, a subset of efficiency mapping model parameters is established for different condition categories;

[0033] Real-time acquisition of photovoltaic inverter operating parameters and matching with corresponding operating conditions; extraction of corresponding efficiency mapping model parameter subsets to calibrate the efficiency mapping model.

[0034] When the deviation between the calculated actual output power and the actual measured value exceeds the preset deviation threshold within the sampling period, and the sampling period exceeds the preset sampling period range, the parameter recalibration mechanism is triggered.

[0035] Furthermore, the integration of prediction results also includes:

[0036] Based on historical and real-time operating parameters, ultra-short-term, short-term, and medium-term prediction sub-models are constructed respectively, and the power prediction results of each prediction sub-model are quantitatively evaluated.

[0037] Analyze the probability distribution characteristics and main occurrence periods of prediction errors, and identify the key factors that lead to errors by combining the multi-condition feature library of high-incidence conditions for positioning errors.

[0038] Based on the prediction error analysis results, each prediction sub-model is optimized to generate prediction intervals at different confidence levels. Combining the probability distribution characteristics of the prediction error, a comprehensive prediction result including the optimal prediction value, uncertainty interval, and confidence level is output.

[0039] Furthermore, after generating multi-timescale collaborative prediction results through a weighted averaging strategy, the process also includes:

[0040] Continuously and in real time acquire the multimodal application status of the multi-timescale collaborative prediction results;

[0041] Based on the aforementioned multimodal application scenario, the application value score of the multi-timescale collaborative prediction results is calculated using the following formula:

[0042]

[0043] in, The application value of the multi-timescale collaborative prediction results is scored. For the aforementioned multimodal application scenario in the first... Scores under each application value rating metric For the first Each application value scoring indicator corresponds to a preset scoring weight. For the first The preset error coefficients for each application value scoring indicator. The total number of scoring indicators. Score the application value calculated in the previous calculation. It is the maximum value among the historical application value scores calculated within the most recent preset time period. and As the preset base weights, This is a preset constant;

[0044] When the application value score is lower than the preset score threshold, an optimization basis feature set is extracted from the multimodal application situation based on the preset feature extraction template;

[0045] Based on the aforementioned feature set, a corresponding model optimization strategy is matched from a preset model optimization strategy library;

[0046] Based on the model optimization strategy, the power prediction model is optimized.

[0047] Based on the optimized power prediction model, the ultra-short-term, short-term, and medium-term prediction results are re-fused across time scales, and a new multi-time-scale collaborative prediction result is generated through a weighted averaging strategy.

[0048] Furthermore, after generating multi-timescale collaborative prediction results through a weighted averaging strategy, the process also includes:

[0049] By deploying edge computing nodes on the photovoltaic power station side, the actual output power measurement values ​​of the photovoltaic inverters are continuously received from the sensor network;

[0050] The actual output power measurement values ​​of the continuously received photovoltaic inverters are compared point by point with the multi-timescale collaborative prediction results;

[0051] Based on the results of point-by-point comparison, a time series prediction error matrix is ​​constructed;

[0052] Based on the constructed time series prediction error matrix, a multi-dimensional error source analysis is performed.

[0053] Based on the results of multi-dimensional error source analysis, prediction compensation is performed on the power prediction model.

[0054] Based on the power prediction model after prediction compensation, the ultra-short-term, short-term and medium-term prediction results are re-fused on time scales, and new multi-time-scale collaborative prediction results are generated again through a weighted average strategy.

[0055] The multi-dimensional error source analysis based on the constructed time series prediction error matrix includes:

[0056] Identify the correlation between external environmental data and prediction bias;

[0057] Extract the historical degradation curve of inverter conversion efficiency and quantify the influence coefficient of device aging on power mapping relationship.

[0058] This invention provides another technical solution, a string photovoltaic inverter power prediction system, comprising:

[0059] The distributed parameter acquisition module is configured to collect the operating parameters and environmental data of the photovoltaic inverters in real time based on the sensors deployed on each photovoltaic inverter, and perform preprocessing to generate a fused feature vector containing dynamic trend features, cross-domain correlation features, and state assessment features.

[0060] The dynamic modeling module is configured to filter core input variables and remove redundant features based on the fused feature vector, establish a basic response model between the output power of the photovoltaic inverter and the core input variables, and modify the basic response model based on dynamic trend features and cross-domain correlation features to generate a dynamic response model of the photovoltaic inverter.

[0061] The working condition feature library module is configured to build a multi-working condition feature library based on historical operating data, and store multi-working condition feature data, including core input variables, dynamic trend features, cross-domain correlation features and corresponding model parameter subsets under different working conditions.

[0062] The power assessment module is configured to determine the input power of the photovoltaic inverter based on the output power value output by the dynamic response model, construct the efficiency mapping model of the photovoltaic inverter, and dynamically calibrate the efficiency mapping model based on a multi-operating condition feature library.

[0063] The multi-timescale prediction integration module is configured to construct ultra-short-term, short-term, and medium-term prediction sub-models, quantitatively evaluate the power prediction results of each prediction sub-model, optimize each prediction sub-model based on the prediction error analysis results, and fuse the multi-timescale prediction results to generate a comprehensive prediction result that includes the optimal prediction value, uncertainty interval, and confidence level.

[0064] Compared with the prior art, the beneficial effects of the present invention are:

[0065] By processing dynamic anomaly data, aligning multiple time series data, and constructing multi-dimensional features, the problems of large differences in the dimensions of the original data and asynchronous time series are solved. The generated fused feature vector provides high-quality input for modeling, laying the foundation for prediction accuracy. A dynamic response model is constructed by combining physical principles and data-driven approaches. Targeted adjustments are made through a multi-condition feature library to accurately capture the impact of transient changes and non-ideal factors. The dynamic calibration mechanism of the efficiency mapping model further improves the accuracy of power assessment, solves the problem of poor adaptability of traditional models under different operating conditions, constructs ultra-short-term, short-term, and medium-term prediction sub-models and dynamically fuses them. Combined with error analysis and uncertainty quantification, multi-confidence prediction results are output to meet the needs of different time scales such as grid dispatching and planning, thereby improving photovoltaic consumption and grid stability.

[0066] The system continuously optimizes itself to cope with changes in different environments and load conditions, improving the accuracy of power prediction and the stability of the system.

[0067] By using edge computing nodes to achieve high-precision, real-time photovoltaic inverter power prediction and compensation, prediction errors can be accurately identified and corrected, prediction accuracy can be improved, environmental changes and equipment status can be responded to quickly, system adaptability can be enhanced, equipment lifespan can be extended, and grid dispatch can be optimized through multi-timescale prediction fusion, providing strong technical support. Attached Figure Description

[0068] Figure 1 This is a flowchart of the string photovoltaic inverter power prediction method of the present invention. Detailed Implementation

[0069] 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.

[0070] To address the technical issues of existing technologies failing to consider string-level characteristic differences, thus failing to reflect distributed characteristics, and lacking adaptability to complex operating conditions such as sudden changes in irradiance, as well as the lack of dynamic adjustment of prediction coefficients; and the single time scale failing to meet the needs of multiple scenarios and support diverse decision-making, please refer to [link to relevant documentation]. Figure 1 This embodiment provides the following technical solution:

[0071] Power prediction methods for string photovoltaic inverters include:

[0072] Photovoltaic inverter parameter acquisition: Real-time acquisition of photovoltaic inverter operating parameters based on distributed sensor network, including open-circuit voltage, short-circuit current, operating temperature and DC bus voltage of each photovoltaic inverter, and acquisition of external environmental data, including irradiance, temperature, humidity and numerical weather forecast data; and preprocessing of the acquired operating parameters and external environmental data to eliminate the influence of different units, which facilitates subsequent model processing;

[0073] The dynamic response model of the photovoltaic inverter is constructed based on the preprocessed operating parameters and external environmental data. The transient and steady-state characteristics of the output power of the photovoltaic inverter as a function of key factors such as irradiance and temperature are extracted.

[0074] Photovoltaic inverter power assessment establishes a mapping relationship between the input power and output power of the photovoltaic inverter. Based on the conversion efficiency and internal losses of the photovoltaic inverter at different operating nodes, the actual output power of the photovoltaic inverter is calculated.

[0075] The prediction results are integrated to obtain historical operating parameters of string photovoltaic inverters. Combined with real-time operating parameters and external environmental data, a power prediction model adapted to different prediction time scales is constructed. The ultra-short-term, short-term and medium-term prediction results are fused at different time scales, and a multi-time-scale collaborative prediction result is generated through a weighted average strategy.

[0076] In this embodiment, multi-source data preprocessing addresses the issues of dimensional differences and temporal asynchrony, generating a fused feature vector that provides high-quality input for modeling. Its dynamic feature construction accurately adapts to string characteristics, achieving more refined data processing than existing technologies. Based on this, a model is constructed by integrating physical principles and data-driven approaches, accurately capturing both transient and steady-state characteristics. Furthermore, multi-condition calibration addresses the poor adaptability of traditional models, significantly improving prediction accuracy in complex scenarios. Combining loss quantification and dynamic calibration mechanisms overcomes the coarseness of traditional efficiency assessments, achieving precise output of actual power output through multi-condition parameter subset adaptation. Finally, multi-timescale fusion meets diverse needs, uncertainty quantification improves decision reliability, and the weighted fusion strategy significantly enhances adaptability to power grid dispatch, forming a complete technical closed loop from data processing to decision support.

[0077] In this embodiment, the photovoltaic inverter parameter acquisition undergoes preprocessing, including:

[0078] Dynamic anomaly data processing: An adaptive threshold model is constructed based on the distribution characteristics of historical data. The deviation between the collected operating parameters and the external environment data is monitored in real time based on the adaptive threshold model. Anomalies are identified based on the data deviation, smoothing is performed, and missing data is imputed or labeled using polynomial fitting.

[0079] Multi-source data timing alignment: A reference time axis is established through a distributed clock synchronization protocol to perform time alignment and synchronization processing on electrical parameters and external environmental data. The DC side voltage jump signal of the photovoltaic inverter is used as the time anchor point to map data with different sampling frequencies to a unified time scale. The synchronization deviation is corrected by quadratic curve fitting to ensure that the time axis error does not exceed one-tenth of the sampling period.

[0080] Multidimensional feature construction: Based on the preprocessed electrical parameters and external environment data, dynamic trend features, cross-domain correlation features, and state assessment features are generated. For example, the average irradiance, irradiance change rate, and harmonic content of voltage and current within a specific time period are calculated and quantified to generate a fusion feature vector of electrical parameters and external environment data, which is directly used to construct the input layer of the photovoltaic inverter dynamic response model.

[0081] In this embodiment, the dynamic trend features are: calculating the slope of irradiance change, voltage and current fluctuation amplitude, and cumulative temperature change within three consecutive sampling windows, and strengthening the weight of recent data through a sliding weighted algorithm;

[0082] In this embodiment, the cross-domain correlation feature is: constructing a coupling coefficient matrix between environmental parameters and electrical parameters, calculating interactive features such as irradiance-open-circuit voltage correlation and temperature-short-circuit current sensitivity, and quantifying the degree of influence of environmental factors on the electrical characteristics of photovoltaic inverters;

[0083] In this embodiment, the state assessment features are as follows: the component performance degradation coefficient is calculated based on the deviation between the open-circuit voltage temperature coefficient and the operating temperature, and a health state feature vector is generated by combining the short-circuit current change rate to characterize the long-term operating state of the photovoltaic inverter.

[0084] In this embodiment, the feature values ​​are mapped to the [0,1] interval, and dynamic weight factors are assigned to different types of features. The weight values ​​are dynamically adjusted according to their mutual information values ​​with historical power data. The processed data is directly used to construct the input layer of the photovoltaic inverter dynamic response model.

[0085] In this embodiment, the construction of the dynamic response model of the photovoltaic inverter includes:

[0086] Key variable screening and correlation analysis: Based on the fused feature vector, the correlation degree between each feature and the actual output power of the photovoltaic inverter is calculated. Based on the preset correlation degree threshold, irradiance, operating temperature, open circuit voltage, fill factor and component performance degradation coefficient are selected as core input variables. The mutual correlation strength between the core input variables is determined. The correlation strength is compared with the preset redundant correlation threshold and redundant features are eliminated.

[0087] Basic response model construction: Based on the circuit principle of photovoltaic modules, a basic response model is established between the output power of the photovoltaic inverter and the selected core input variables to obtain the basic response law of the output power of the photovoltaic inverter as the core input variables change, such as the power-irradiance curve, the power-temperature curve, etc.

[0088] Dynamic response model optimization: Based on dynamic trend characteristics and cross-domain correlation characteristics, the deviations caused by neglecting rapid dynamic changes and interactions between photovoltaic inverters / between photovoltaic inverters and the environment when the basic response model predicts the output power of the photovoltaic inverter are identified and quantified. The deviations are quantified and used as correction terms to modify the basic response model, thereby improving the model's response accuracy to actual operating conditions. The output power value of the modified dynamic response model of the photovoltaic inverter is used as the final output, which reflects the comprehensive response of the photovoltaic inverter output power to the core input variables and their dynamic changes at the current and nearest time.

[0089] In this embodiment, the construction of the dynamic response model of the photovoltaic inverter also includes:

[0090] Collect photovoltaic inverter operation data for different seasons, different weather types (such as sunny, cloudy, overcast, rainy and snowy days) and different time periods (such as morning, noon and evening), and combine them with the corresponding core input variables, dynamic trend features and cross-domain correlation features to construct a multi-condition feature library;

[0091] Acquire historical operating data of string photovoltaic inverters, extract operating condition features from the historical operating data, and perform similarity matching with a multi-operating condition feature library to determine different operating conditions in the historical operating data.

[0092] Based on the actual output power of the photovoltaic inverter under different operating conditions in historical operating data, the modified dynamic response model is adaptively verified, and the deviation rate between the output power of the dynamic response model and the actual output power is calculated.

[0093] When the deviation rate exceeds the preset deviation threshold, the core input variables, dynamic trend characteristics and cross-domain correlation characteristics of the corresponding working condition are obtained, the inter-correlation strength between the core input variables is recalculated, and the parameters and correction term weights of the basic response model are adjusted.

[0094] In this embodiment, the operating condition identification results are used to classify historical operating data or model training data into corresponding operating condition categories. For each identified operating condition category, the parameters (such as temperature coefficient, irradiance response coefficient, etc.) and dynamic correction terms (such as correction factors based on dynamic trends and cross-domain correlations) in the basic response model (e.g., a power calculation model based on circuit principles) are re-estimated or optimized using a subset of data within that category. For example, for a sunny midday operating condition, the model's temperature coefficient may need to be adjusted to more accurately reflect performance degradation under high temperatures; for cloudy weather, the weight of the dynamic correction term for rapid changes in irradiance may need to be increased. Through this targeted tuning based on operating conditions, the dynamic response model can dynamically adjust its internal parameters and correction strategies according to the currently identified operating condition, thereby maintaining high prediction accuracy under various operating conditions.

[0095] In this embodiment, the photovoltaic inverter power assessment includes:

[0096] Basic data acquisition: Based on the output power value output by the dynamic response model and combined with the real-time collected operating parameters, the input power of the photovoltaic inverter is determined, and the output voltage, current and operating temperature data of the AC side of the photovoltaic inverter are collected at the same time.

[0097] Constructing an efficiency mapping model: Based on the input power, AC side output voltage, current and operating temperature data of the photovoltaic inverter, an efficiency mapping model of the photovoltaic inverter is constructed to describe the conversion efficiency curve of the photovoltaic inverter under different load rates and temperature conditions;

[0098] Loss Quantification and Actual Output Power Calculation: The internal losses of the photovoltaic inverter are decomposed into fixed losses and variable losses. Fixed losses include core losses and control circuit losses. The loss coefficient is fitted based on the photovoltaic inverter nameplate parameters and operating temperature. Variable losses include switching losses and conduction losses. The loss value is calculated by using the effective value of the input current and the switching frequency. Combining the conversion efficiency output by the efficiency mapping model with the quantified loss value, the actual output power of the photovoltaic inverter is calculated: Actual output power = Photovoltaic inverter input power × Conversion efficiency - Total loss;

[0099] Dynamic calibration of mapping relationship: Based on the multi-condition feature library, a subset of efficiency mapping model parameters is established for different condition categories (such as high load rate - high temperature, low load rate - normal temperature, etc.); the operating parameters of the photovoltaic inverter are collected in real time and matched with the corresponding operating conditions, and the corresponding subset of efficiency mapping model parameters is extracted to dynamically calibrate the efficiency mapping model; when the deviation between the calculated value of actual output power and the actual measured value exceeds the preset deviation threshold within the sampling period, and the sampling period exceeds the preset sampling period range, the parameter recalibration mechanism is triggered to reacquire the operating parameters of the photovoltaic inverter, update the polynomial fitting coefficients, and ensure the timeliness of the mapping relationship.

[0100] In this embodiment, the efficiency mapping model accurately depicts the conversion efficiency under different load rates and temperatures, and the loss quantification mechanism refines the calculation of fixed and variable losses, improving the reliability of actual output power results. The dynamic calibration of the mapping relationship, combined with a subset of parameters under multiple operating conditions, enables model adaptation under different operating conditions. The recalibration mechanism triggered by deviation exceeding limits ensures long-term timeliness, solving problems such as traditional assessment ignoring differences in operating conditions, coarse loss quantification, and time-varying model failure. This provides high-precision actual output power data for power prediction and ensures the accuracy of predictions at multiple time scales.

[0101] In this embodiment, the prediction result integration also includes:

[0102] Based on historical and real-time operating parameters, ultra-short-term (0-1 hour), short-term (1-24 hours), and medium-term (24-72 hours) prediction sub-models were constructed, and the power prediction results of each prediction sub-model were quantitatively evaluated.

[0103] Analyze the probability distribution characteristics and main occurrence periods of prediction errors, combine the multi-condition feature library to locate the high-incidence conditions of errors, and identify the key factors that cause errors, which may be the limitations of the model itself, data quality problems, or sudden changes in the external environment.

[0104] Based on the prediction error analysis results, each prediction sub-model is optimized, such as by adjusting model parameters, adding or modifying features, and improving model structure. The optimization process can combine online learning and offline retraining strategies. Prediction intervals at different confidence levels are generated through Monte Carlo simulation. Combining the probability distribution characteristics of prediction errors, a comprehensive prediction result including the optimal prediction value, uncertainty interval, and confidence level is output, allowing users to make decisions based on prediction reliability.

[0105] In this embodiment, the ultra-short-term prediction sub-model inputs dynamic trend features, focusing on capturing the transient characteristics of rapid changes in irradiance; the short-term prediction sub-model combines cross-domain correlation features with numerical weather forecast data to explore the power change patterns within the daily cycle; and the medium-term prediction sub-model introduces long-term trend parameters from the state assessment features to match historical operating conditions.

[0106] In this embodiment, after generating multi-timescale collaborative prediction results through a weighted averaging strategy, the method further includes:

[0107] Continuously and in real time acquire the multimodal application status of the multi-timescale collaborative prediction results;

[0108] Based on the aforementioned multimodal application scenario, the application value score of the multi-timescale collaborative prediction results is calculated using the following formula:

[0109]

[0110] in, The application value of the multi-timescale collaborative prediction results is scored. For the aforementioned multimodal application scenario in the first... Scores under each application value rating metric For the first Each application value scoring indicator corresponds to a preset scoring weight. For the first The preset error coefficients for each application value scoring indicator. The total number of scoring indicators. Score the application value calculated in the previous calculation. It is the maximum value among the historical application value scores calculated within the most recent preset time period. and As the preset base weights, This is a preset constant;

[0111] When the application value score is lower than the preset score threshold, an optimization basis feature set is extracted from the multimodal application situation based on the preset feature extraction template;

[0112] Based on the aforementioned feature set, a corresponding model optimization strategy is matched from a preset model optimization strategy library;

[0113] Based on the model optimization strategy, the power prediction model is optimized.

[0114] Based on the optimized power prediction model, the ultra-short-term, short-term, and medium-term prediction results are re-fused across time scales, and a new multi-time-scale collaborative prediction result is generated through a weighted averaging strategy.

[0115] In this embodiment, multimodal application refers to the performance of the photovoltaic inverter's power prediction model in multiple application scenarios under different operating conditions. These application scenarios involve different geographical locations, climate conditions, load conditions, etc., or prediction results at different time scales, such as ultra-short-term, short-term, and medium-term predictions. To more accurately predict power output, it is necessary to track and analyze these application scenarios under different conditions in real time. By collecting real-time operating data of the photovoltaic inverter and external environmental data, a dynamic monitoring system can be constructed, which can not only capture the instantaneous response of the photovoltaic inverter but also obtain its stability performance under various environments.

[0116] Once multimodal application scenarios are acquired, the system calculates an application value score for the multi-timescale collaborative prediction results based on this data. The application value score represents the performance score in a specific application scenario. A preset error coefficient for the application value score describes the error range or sensitivity of the score. The weight of the application value score represents its influence on the overall score. This weight can be adjusted according to actual needs. The previously calculated application value score represents the application value obtained within a previous time period. The maximum application value score calculated historically within the most recent preset time period is used to normalize the previously calculated application value score for comparison. Two basic weight parameters are used to balance the weighting of current and historical prediction results. These weights control the weight ratio between new and historical data. Preset constants are used for final standardization to ensure a reasonable range for the score values. Based on these calculated application value scores, the system can continuously adjust the power prediction model to achieve the best prediction results.

[0117] When the calculated application value score is lower than the preset score threshold, it means that the existing prediction model may have certain biases or deficiencies. In this case, the system will extract an optimization feature set from the real-time acquired multimodal application data based on a specific feature extraction template. These feature sets contain potential problems that the model may have, such as model failure under certain environmental conditions or performance instability within a specific time period.

[0118] The information extracted from the feature set used for optimization will be used to match corresponding optimization strategies from the model optimization strategy library. These optimization strategies may include adjusting algorithm parameters, updating data processing methods, and changing the weights of the weighted averaging strategy. Through these optimization measures, the system can improve the accuracy and reliability of the prediction model.

[0119] In the optimized prediction model, the ultra-short-term, short-term, and medium-term prediction results will be re-fused across time scales and combined with a new weighted averaging strategy to generate new multi-time-scale collaborative prediction results.

[0120] Through this process, the system continuously optimizes itself to cope with changes in different environments and load conditions, thereby improving the accuracy of power prediction and the stability of the system.

[0121] In this embodiment, after generating multi-timescale collaborative prediction results through a weighted averaging strategy, the method further includes:

[0122] By deploying edge computing nodes on the photovoltaic power station side, the actual output power measurement values ​​of the photovoltaic inverters are continuously received from the sensor network;

[0123] The actual output power measurement values ​​of the continuously received photovoltaic inverters are compared point by point with the multi-timescale collaborative prediction results;

[0124] Based on the results of point-by-point comparison, a time series prediction error matrix is ​​constructed;

[0125] Based on the constructed time series prediction error matrix, a multi-dimensional error source analysis is performed.

[0126] Based on the results of multi-dimensional error source analysis, prediction compensation is performed on the power prediction model.

[0127] Based on the power prediction model after prediction compensation, the ultra-short-term, short-term and medium-term prediction results are re-fused on time scales, and new multi-time-scale collaborative prediction results are generated again through a weighted average strategy.

[0128] The multi-dimensional error source analysis based on the constructed time series prediction error matrix includes:

[0129] Identify the correlation between external environmental data and prediction bias;

[0130] Extract the historical degradation curve of inverter conversion efficiency and quantify the influence coefficient of device aging on power mapping relationship.

[0131] In this embodiment, edge computing nodes are deployed at the photovoltaic power station to receive real-time data from a distributed sensor network. The output power of the photovoltaic inverter is a key parameter of the photovoltaic power generation system, affected by various factors such as environmental conditions, the inverter's own operating status, and equipment aging. The sensor network is responsible for collecting various operating parameters of the photovoltaic inverter in real time, such as voltage, current, frequency, temperature, and power, and transmitting the data to the edge computing nodes. The edge computing nodes not only undertake the data collection task but also play a role in local data processing and analysis.

[0132] Edge computing nodes typically possess strong processing capabilities, enabling them to process data locally in real time, reducing reliance on remote servers and achieving low-latency data feedback. Actual output power measurements collected through sensor networks are fundamental data for evaluating the actual performance of photovoltaic inverters. Edge computing nodes continuously receive this real-time data, perform preprocessing and basic analysis for subsequent comparison and compensation calculations.

[0133] Once the edge computing nodes collect the actual output power measurements of the photovoltaic inverter, the next step is to compare and analyze them with existing power prediction models. These models predict the inverter's power output using historical operating parameters, external environmental data, and the inverter's dynamic response model. However, due to various factors (such as equipment aging, environmental changes, and sudden weather events), discrepancies may exist between the actual output power and the predicted values. Therefore, it is necessary to compare the real-time collected actual output power with the model's predictions point-by-point. This comparison process generates an error matrix, which contains the differences between the predicted and actual power at each time point. The error matrix not only helps understand the accuracy of the current model but also reflects the error trend at specific time points. For example, during certain periods, weather changes, inverter load variations, and other factors may cause significant deviations between the power prediction and actual power. These differences are quantified through the error matrix, providing a data foundation for subsequent error source analysis.

[0134] After constructing the error matrix, the next crucial step is to conduct multi-dimensional error source analysis. This process involves in-depth analysis of various potential influencing factors to identify the root causes of power prediction errors. The output power of photovoltaic inverters is affected by external environmental factors such as temperature, humidity, radiation intensity, and wind speed. Therefore, it is first necessary to identify the relationship between external environmental data (such as weather changes, light intensity, and temperature changes) and prediction errors. For example, weather forecasting systems may contain errors, leading to discrepancies between actual and predicted weather conditions, thus affecting the inverter's power output. By analyzing the correlation between these external data and power prediction errors, the causes of inaccurate predictions can be identified, providing guidance for subsequent compensation adjustments. The inverter's conversion efficiency gradually decreases over time and with equipment aging, especially after long-term operation, where performance may degrade to some extent. This factor has a significant impact on the power mapping relationship. Therefore, by extracting the historical degradation curve of the inverter's conversion efficiency and quantifying its impact coefficient on power output, a more accurate basis for error analysis can be provided. Equipment aging typically leads to a decrease in conversion efficiency, resulting in actual output power lower than the predicted value. This analysis enables timely identification of signs of equipment performance degradation and provides decision support for equipment maintenance or replacement.

[0135] Based on the results of multi-dimensional error source analysis, predictive compensation is performed on the power prediction model. The goal of compensation is to adjust the original prediction model according to the measured results of actual output power and the analyzed error sources. Predictive compensation is usually achieved by correcting known errors and adjusting the model's input parameters or algorithm, so that the model can more accurately predict future power output.

[0136] Once the model has been compensated, the results of ultra-short-term, short-term, and medium-term forecasts can be fused using a weighted averaging strategy. This strategy dynamically adjusts the weights of the forecasts at different time scales based on the forecast errors, thereby generating new multi-time-scale collaborative forecasts. This process not only improves the accuracy of the forecasts but also reduces the uncertainty caused by changes in time scales.

[0137] After predictive compensation, a new power prediction model is reconstructed, and a new multi-timescale collaborative prediction result is generated through a weighted averaging strategy. Ultimately, these new prediction results can be provided to the photovoltaic power plant management system to help optimize inverter operation scheduling and fault early warning.

[0138] By using edge computing nodes to achieve high-precision, real-time photovoltaic inverter power prediction and compensation, prediction errors can be accurately identified and corrected, prediction accuracy can be improved, environmental changes and equipment status can be responded to quickly, system adaptability can be enhanced, equipment lifespan can be extended, and grid dispatch can be optimized through multi-timescale prediction fusion, providing strong technical support.

[0139] To better demonstrate the implementation of the string photovoltaic inverter power prediction method, this invention provides a string photovoltaic inverter power prediction system, comprising:

[0140] The distributed parameter acquisition module is configured to collect the operating parameters and environmental data of the photovoltaic inverters in real time based on the sensors deployed on each photovoltaic inverter, and perform preprocessing to generate a fused feature vector containing dynamic trend features, cross-domain correlation features, and state assessment features.

[0141] The dynamic modeling module is configured to filter core input variables and remove redundant features based on the fused feature vector, establish a basic response model between the output power of the photovoltaic inverter and the core input variables, and modify the basic response model based on dynamic trend features and cross-domain correlation features to generate a dynamic response model of the photovoltaic inverter.

[0142] The working condition feature library module is configured to build a multi-working condition feature library based on historical operating data, and store multi-working condition feature data, including core input variables, dynamic trend features, cross-domain correlation features and corresponding model parameter subsets under different working conditions.

[0143] The power assessment module is configured to determine the input power of the photovoltaic inverter based on the output power value output by the dynamic response model, construct the efficiency mapping model of the photovoltaic inverter, and dynamically calibrate the efficiency mapping model based on a multi-operating condition feature library.

[0144] The multi-timescale prediction integration module is configured to construct ultra-short-term, short-term, and medium-term prediction sub-models, quantitatively evaluate the power prediction results of each prediction sub-model, optimize each prediction sub-model based on the prediction error analysis results, and fuse the multi-timescale prediction results to generate a comprehensive prediction result that includes the optimal prediction value, uncertainty interval, and confidence level.

[0145] In this embodiment, the distributed parameter acquisition module ensures data quality and feature validity, laying the foundation for modeling; the dynamic modeling module combines physical principles with data-driven correction to accurately reflect the power response under different operating conditions; the operating condition feature library module provides a unified benchmark to improve system response speed and stability; the power assessment module ensures accurate calculation of actual output power through efficiency mapping and loss quantification; and the multi-timescale prediction integration module achieves multi-dimensional accurate prediction, solving problems such as data quality, model adaptability, and operating condition differences, comprehensively improving prediction accuracy and practicality.

[0146] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for predicting the power of a string photovoltaic inverter, characterized in that, include: Photovoltaic inverter parameter acquisition: Real-time acquisition of photovoltaic inverter operating parameters based on distributed sensor network, while acquiring external environmental data; The acquired operating parameters and external environment data are preprocessed. The dynamic response model of the photovoltaic inverter is constructed based on the preprocessed operating parameters and external environmental data, and the transient and steady-state characteristics of the output power of the photovoltaic inverter as key factors change are extracted. Photovoltaic inverter power assessment establishes a mapping relationship between the input power and output power of the photovoltaic inverter. Based on the conversion efficiency and internal losses of the photovoltaic inverter at different operating nodes, the actual output power of the photovoltaic inverter is calculated. The prediction results are integrated to obtain historical operating parameters of string photovoltaic inverters. Combined with real-time operating parameters and external environmental data, a power prediction model is constructed. The ultra-short-term, short-term and medium-term prediction results are fused over time, and a multi-time-scale collaborative prediction result is generated through a weighted average strategy. Following the generation of multi-timescale collaborative prediction results using a weighted averaging strategy, the process also includes: The multimodal application of the multi-timescale collaborative prediction results is continuously acquired in real time. The multimodal application refers to the performance of the photovoltaic inverter's power prediction model in multiple application scenarios under different operating conditions. Based on the aforementioned multimodal application scenario, the application value score of the multi-timescale collaborative prediction results is calculated using the following formula: in, The application value of the multi-timescale collaborative prediction results is scored. For the aforementioned multimodal application scenario in the first... Scores under each application value rating metric For the first Each application value scoring indicator corresponds to a preset scoring weight. For the first The preset error coefficients for each application value scoring indicator. The total number of scoring indicators. Score the application value calculated in the previous calculation. It is the maximum value among the historical application value scores calculated within the most recent preset time period. and As the preset base weights, This is a preset constant; When the application value score is lower than the preset score threshold, an optimization basis feature set is extracted from the multimodal application situation based on the preset feature extraction template; Based on the aforementioned feature set, a corresponding model optimization strategy is matched from a preset model optimization strategy library; Based on the model optimization strategy, the power prediction model is optimized. Based on the optimized power prediction model, the ultra-short-term, short-term, and medium-term prediction results are re-fused across time scales, and a new multi-time-scale collaborative prediction result is generated through a weighted averaging strategy.

2. The power prediction method for string photovoltaic inverters as described in claim 1, characterized in that, The acquired operating parameters and external environment data are preprocessed, including: Dynamic anomaly data processing: An adaptive threshold model is constructed based on the distribution characteristics of historical data. The deviation between the collected operating parameters and the external environment data is monitored in real time based on the adaptive threshold model. Anomalies are identified based on the data deviation, smoothing is performed, and missing data is imputed or labeled using polynomial fitting. Multi-source data timing alignment: A reference time axis is established through a distributed clock synchronization protocol to perform time alignment and synchronization processing on electrical parameters and external environment data; Multidimensional feature construction: Based on the preprocessed electrical parameters and external environment data, dynamic trend features, cross-domain correlation features and state assessment features are generated and quantified to generate a fusion feature vector of electrical parameters and external environment data.

3. The power prediction method for string photovoltaic inverters as described in claim 2, characterized in that, The construction of the dynamic response model for photovoltaic inverters includes: Key variable screening and correlation analysis: Based on the fused feature vector, the correlation degree between each feature and the actual output power of the photovoltaic inverter is calculated. The core input variables are screened based on the preset correlation degree threshold, and the mutual correlation strength between the core input variables is determined. The correlation strength is compared with the preset redundant correlation threshold, and redundant features are eliminated. Basic response model construction: Based on the circuit principle of photovoltaic modules, a basic response model is established between the output power of the photovoltaic inverter and the selected core input variables to obtain the basic response law of the output power of the photovoltaic inverter as the core input variables change; Dynamic response model optimization: Based on dynamic trend characteristics and cross-domain correlation characteristics, the deviation generated by the basic response model in predicting the output power of the photovoltaic inverter is identified and quantified. The deviation is quantified and used as a correction term to correct the basic response model. The output power value of the dynamic response model of the photovoltaic inverter after correction is used as the final output.

4. The power prediction method for string photovoltaic inverters as described in claim 3, characterized in that, The construction of the dynamic response model for photovoltaic inverters also includes: Collect photovoltaic inverter operation data for different seasons, weather types, and time periods, and combine them with corresponding core input variables, dynamic trend features, and cross-domain correlation features to construct a multi-condition feature library; Acquire historical operating data of string photovoltaic inverters, extract operating condition features from the historical operating data, and perform similarity matching with a multi-operating condition feature library to determine different operating conditions in the historical operating data. Based on the actual output power of the photovoltaic inverter under different operating conditions in historical operating data, the modified dynamic response model is adaptively verified, and the deviation rate between the output power of the dynamic response model and the actual output power is calculated. When the deviation rate exceeds the preset deviation threshold, the core input variables, dynamic trend characteristics and cross-domain correlation characteristics of the corresponding working condition are obtained, the inter-correlation strength between the core input variables is recalculated, and the parameters and correction term weights of the basic response model are adjusted.

5. The power prediction method for string photovoltaic inverters as described in claim 4, characterized in that, Photovoltaic inverter power assessment, including: Basic data acquisition: Based on the output power value output by the dynamic response model and combined with the real-time collected operating parameters, the input power of the photovoltaic inverter is determined, and the output voltage, current and operating temperature data of the AC side of the photovoltaic inverter are collected at the same time. Constructing an efficiency mapping model: Based on the input power, AC side output voltage, current, and operating temperature data of the photovoltaic inverter, an efficiency mapping model of the photovoltaic inverter is constructed. Loss quantification and actual output power calculation: Decompose the internal losses of the photovoltaic inverter, combine the conversion efficiency output by the efficiency mapping model with the quantified loss value, and calculate the actual output power of the photovoltaic inverter.

6. The power prediction method for string photovoltaic inverters as described in claim 5, characterized in that, Photovoltaic inverter power assessment also includes dynamic calibration of mapping relationships: Based on the multi-condition feature library, a subset of efficiency mapping model parameters is established for different condition categories; Real-time acquisition of photovoltaic inverter operating parameters and matching with corresponding operating conditions; extraction of corresponding efficiency mapping model parameter subsets to calibrate the efficiency mapping model. When the deviation between the calculated actual output power and the actual measured value exceeds the preset deviation threshold within the sampling period, and the sampling period exceeds the preset sampling period range, the parameter recalibration mechanism is triggered.

7. The power prediction method for string photovoltaic inverters as described in claim 6, characterized in that, The prediction results integration also includes: Based on historical and real-time operating parameters, ultra-short-term, short-term, and medium-term prediction sub-models are constructed respectively, and the power prediction results of each prediction sub-model are quantitatively evaluated. Analyze the probability distribution characteristics and main occurrence periods of prediction errors, and identify the key factors that lead to errors by combining the multi-condition feature library of high-incidence conditions for positioning errors. Based on the prediction error analysis results, each prediction sub-model is optimized to generate prediction intervals at different confidence levels. Combining the probability distribution characteristics of the prediction error, a comprehensive prediction result including the optimal prediction value, uncertainty interval, and confidence level is output.

8. The power prediction method for string photovoltaic inverters as described in claim 1, characterized in that, Following the generation of multi-timescale collaborative prediction results using a weighted averaging strategy, the process also includes: By deploying edge computing nodes on the photovoltaic power station side, the actual output power measurement values ​​of the photovoltaic inverters are continuously received from the sensor network; The actual output power measurement values ​​of the continuously received photovoltaic inverters are compared point by point with the multi-timescale collaborative prediction results; Based on the results of point-by-point comparison, a time series prediction error matrix is ​​constructed; Based on the constructed time series prediction error matrix, a multi-dimensional error source analysis is performed. Based on the results of multi-dimensional error source analysis, prediction compensation is performed on the power prediction model. Based on the power prediction model after prediction compensation, the ultra-short-term, short-term and medium-term prediction results are re-fused on time scales, and new multi-time-scale collaborative prediction results are generated again through a weighted average strategy. The multi-dimensional error source analysis based on the constructed time series prediction error matrix includes: Identify the correlation between external environmental data and prediction bias; Extract the historical degradation curve of inverter conversion efficiency and quantify the influence coefficient of device aging on power mapping relationship.

9. A string photovoltaic inverter power prediction system, applied in the string photovoltaic inverter power prediction method as described in claim 1, characterized in that, include: The distributed parameter acquisition module is configured to collect real-time operating parameters of photovoltaic inverters and external environmental data based on sensors deployed on each photovoltaic inverter, and perform preprocessing to generate a fused feature vector containing dynamic trend features, cross-domain correlation features, and state assessment features. The dynamic modeling module is configured to filter core input variables and remove redundant features based on the fused feature vector, establish a basic response model between the output power of the photovoltaic inverter and the core input variables, and modify the basic response model based on dynamic trend features and cross-domain correlation features to generate a dynamic response model of the photovoltaic inverter. The working condition feature library module is configured to build a multi-working condition feature library based on historical operating data, and store multi-working condition feature data, including core input variables, dynamic trend features, cross-domain correlation features and corresponding model parameter subsets under different working conditions. The power assessment module is configured to determine the input power of the photovoltaic inverter based on the output power value output by the dynamic response model, construct the efficiency mapping model of the photovoltaic inverter, and dynamically calibrate the efficiency mapping model based on a multi-operating condition feature library. The multi-timescale prediction integration module is configured to construct ultra-short-term, short-term, and medium-term prediction sub-models, quantitatively evaluate the power prediction results of each prediction sub-model, optimize each prediction sub-model based on the prediction error analysis results, and fuse the multi-timescale prediction results to generate a comprehensive prediction result that includes the optimal prediction value, uncertainty interval, and confidence level.

Images