Photovoltaic power station generating capacity prediction analysis method and system based on multi-source data

Abstract

The invention discloses a photovoltaic power station generating capacity prediction analysis method and system based on multi-source data, and belongs to the technical field of data processing, and the method comprises the following steps: S1, obtaining an environment disturbance label which comprises environment disturbance and environment stability; s2, if the environment disturbance label is environment disturbance, photovoltaic inverter parameter adjustment is carried out, otherwise, adjustment is not executed, and a real-time meteorological environment parameter set is updated and obtained; s3, outputting a power generation amount prediction value of the to-be-predicted power generation period; s4, obtaining a meteorological correction generating capacity prediction value; and S5, based on the meteorological correction generating capacity prediction value, analyzing to obtain a generating capacity prediction label, if the generating capacity prediction label is generating capacity prediction deviation, executing generating capacity prediction model correction, otherwise, not executing generating capacity prediction model correction. The problem that in the prior art, due to the fact that the air density and air pressure changes of the high-altitude area are not fully considered, the photovoltaic power generation capacity prediction precision is insufficient is solved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for predicting and analyzing the power generation of photovoltaic power plants based on multi-source data. Background Technology

[0002] Existing photovoltaic power plant power generation prediction systems achieve prediction by integrating equipment, meteorological, and sky image data. On the one hand, they connect with meteorological stations in high-altitude areas to obtain real-time weather data, and on the other hand, they collect equipment data from photovoltaic equipment. Then, they input the data into a mathematical model to calculate the power generation and output the predicted power generation, thus realizing power generation prediction driven by multi-source data.

[0003] For example, Chinese invention patent CN106909985B discloses a prediction method for a photovoltaic power generation prediction system, which includes: a timed start prediction module, a manual start prediction module, a prediction statistical calculation module, an equipment management module, and a weather detection module. The timed start prediction module is connected to the prediction statistical calculation module and has a preset start time. Upon reaching the start time, the prediction statistical calculation module is triggered to statistically analyze equipment and weather data. The weather detection module connects to an international meteorological station server via a detection interface to read real-time weather conditions and transmits the data to the prediction statistical calculation module. The equipment management module connects to the photovoltaic power generation equipment server to obtain the equipment's serial number, latitude and longitude, placement angle, and photovoltaic material type data, and transmits this data to the prediction statistical calculation module. The prediction statistical calculation module combines the statistical weather data and equipment data and uses a mathematical model to calculate the predicted power generation.

[0004] For example, the Chinese invention patent with announcement number CN107133685B discloses a method and system for predicting the power generation of a photovoltaic power generation system, which includes: using the proportion of cloud area to sky area in a sky image corresponding to a first preset time to estimate the proportion of cloud area to sky area in a second preset time; then using the proportion of cloud area to sky area in a sky image corresponding to the second preset time to calculate the total solar radiation at the second preset time; and finally using the total solar radiation and temperature at the second preset time to predict the power generation of the photovoltaic power generation system at the second preset time.

[0005] The above-mentioned technology has at least the following technical problems: In existing technologies, power generation prediction models typically only use altitude as a fixed correction factor in calculations, without fully considering the significant fluctuations in air density and pressure at high altitudes. At high altitudes, air density decreases and air pressure changes are substantial. These changes are amplified, leading to a continuous accumulation of discrepancies between the power generation calculated by the prediction model and the actual power generation. Therefore, there is a problem of insufficient accuracy in photovoltaic power generation predictions due to the inadequate consideration of changes in air density and pressure at high altitudes. Summary of the Invention

[0006] To address the problem of insufficient accuracy in photovoltaic power generation prediction caused by inadequate consideration of changes in air density and pressure at high altitudes in existing technologies, this invention provides a method and system for predicting and analyzing photovoltaic power generation based on multi-source data. The technical solution is as follows: On the one hand, a method for predicting and analyzing photovoltaic power generation based on multi-source data is provided. This method includes: S1, acquiring photovoltaic power generation environmental parameters at each monitoring point of a photovoltaic power station located in a high-altitude area within a preset time period, analyzing the environmental disturbance value to obtain an environmental disturbance label, which includes both environmental disturbance and environmental stability. The environmental disturbance value is used to characterize the degree of influence of the power generation environment on the accuracy of photovoltaic power generation prediction; S2, if the environmental disturbance label indicates environmental disturbance, adjusting the photovoltaic inverter parameters; otherwise, no adjustment is performed, and the real-time meteorological environmental parameter set is updated; S3, acquiring the power generation cycle to be predicted, and combining it with the environmental disturbance value analysis to obtain the predicted power generation cycle. The multi-source data training set, together with the real-time meteorological environmental parameter set, is input into the power generation prediction model to output the power generation prediction value for the power generation cycle to be predicted; S4, obtain the air pressure value and air density at each moment within the preset time period, and analyze them to obtain the air pressure change label and the light refraction feature label, thereby obtaining the meteorological corrected power generation prediction value; S5, based on the meteorological corrected power generation prediction value, analyze to obtain the power generation prediction label. If the power generation prediction label is a deviation from the power generation prediction, then the power generation prediction model correction is performed; otherwise, the power generation prediction model correction is not performed. The power generation prediction model correction includes offline power generation prediction model correction and online adaptive power generation prediction model correction.

[0007] On the other hand, a photovoltaic power plant power generation prediction and analysis system based on multi-source data is provided. This system includes: an environmental disturbance assessment module, a photovoltaic inverter adjustment module, a power generation prediction module, a power generation correction module, and a model correction module. The environmental disturbance assessment module acquires the photovoltaic power generation environmental parameters of each monitoring point at a photovoltaic power plant located in a high-altitude area within a preset time period, analyzes the environmental disturbance value, and obtains an environmental disturbance label. The photovoltaic inverter adjustment module adjusts the photovoltaic inverter parameters if the environmental disturbance label indicates environmental disturbance; otherwise, it does not perform the adjustment and updates the real-time meteorological environmental parameter set. The power generation prediction module acquires the power generation cycle to be predicted and, combined with the environmental disturbance value analysis, obtains the power generation prediction value. A multi-source data training set for predicting power generation cycles, combined with a real-time meteorological environmental parameter set, is input into the power generation prediction model to output the predicted power generation value for the desired power generation cycle. A power generation correction module is used to acquire air pressure and air density at various times within a preset time period, and analyze them to obtain air pressure change labels and light refraction feature labels, thereby obtaining the meteorologically corrected power generation prediction value. A model correction module is used to analyze the power generation prediction label based on the meteorologically corrected power generation prediction value. If the power generation prediction label indicates a deviation from the prediction, power generation prediction model correction is performed; otherwise, it is not performed. Power generation prediction model correction includes offline correction and online adaptive correction.

[0008] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following: 1. The photovoltaic power generation prediction and analysis method based on multi-source data provided by this invention obtains the photovoltaic power generation environmental parameters of each monitoring point of the photovoltaic power station in high-altitude areas, and forms environmental disturbance labels based on the analysis of environmental disturbance values, thereby dynamically judging whether the environment is stable and then adjusting the inverter parameters in real time, which improves the accuracy of photovoltaic power generation prediction and effectively solves the problem of insufficient photovoltaic power generation prediction accuracy caused by insufficient consideration of air density and air pressure changes in high-altitude areas in the existing technology.

[0009] 2. This invention constructs a multi-source data training set for the power generation cycle to be predicted by jointly building environmental disturbance values ​​and real-time meteorological environmental parameters. This enables the analysis and adaptive screening of complex correlations between photovoltaic power generation environmental parameters, thereby improving the adaptability and robustness of the power generation prediction model in high-altitude areas and effectively solving the problem of prediction accuracy deviation caused by insufficient adaptability of existing power generation prediction models.

[0010] 3. This invention obtains the air pressure and air density at each moment within a preset time period, and analyzes them to obtain air pressure change labels and light refraction feature labels, thereby introducing meteorological correction power generation prediction values. This enables real-time correction of power generation deviations caused by air density and air pressure changes and light refraction fluctuations in high-altitude areas, effectively solving the problem of inaccurate power generation prediction caused by meteorological disturbances in the prior art.

[0011] 4. This invention analyzes the power generation prediction labels and selects to perform offline or online adaptive correction of the power generation prediction model based on the degree of prediction deviation. This enables dynamic optimization and iterative updating of the power generation prediction model parameters, thereby ensuring continuous improvement in the accuracy and reliability of photovoltaic power generation prediction during long-term operation. It effectively solves the problem of inaccuracy of photovoltaic power generation prediction models under continuous environmental disturbances in the prior art. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A macroscopic flowchart of the photovoltaic power generation prediction and analysis method based on multi-source data provided in the embodiments of this application; Figure 2 A flowchart illustrating the steps of the photovoltaic power generation prediction and analysis method based on multi-source data provided in this application embodiment; Figure 3 A flowchart illustrating the multi-source data training set acquisition process for the photovoltaic power generation prediction and analysis method based on multi-source data provided in this application embodiment; Figure 4 A schematic diagram of the modules of the photovoltaic power generation prediction and analysis system based on multi-source data provided in the embodiments of this application. Detailed Implementation

[0014] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0015] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0016] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent.

[0017] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] like Figure 1 The diagram shows a flowchart of a photovoltaic power plant power generation prediction and analysis method based on multi-source data provided in this application embodiment. The method includes the following steps: S1. Obtaining photovoltaic power generation environmental parameters at each monitoring point of a photovoltaic power plant located in a high-altitude area within a preset time period, analyzing and obtaining environmental disturbance values, thereby obtaining environmental disturbance labels. The environmental disturbance labels include environmental disturbance and environmental stability. The environmental disturbance value is used to characterize the degree of influence of the power generation environment on the accuracy of photovoltaic power generation prediction; S2. If the environmental disturbance label is environmental disturbance, then adjusting the photovoltaic inverter parameters; otherwise, no adjustment is performed, and the real-time meteorological environmental parameter set is updated; S3. Obtaining the power generation cycle to be predicted, and combining it with the environmental disturbance value analysis, obtaining… The multi-source data training set for the power generation cycle to be predicted, together with the real-time meteorological environmental parameter set, is input into the power generation prediction model to output the power generation prediction value for the power generation cycle to be predicted; S4, obtain the air pressure value and air density at each moment within the preset time period, and analyze them to obtain the air pressure change label and the light refraction feature label, thereby obtaining the meteorological corrected power generation prediction value; S5, based on the meteorological corrected power generation prediction value, analyze to obtain the power generation prediction label. If the power generation prediction label is a deviation from the power generation prediction, then the power generation prediction model correction is performed; otherwise, the power generation prediction model correction is not performed. The power generation prediction model correction includes offline power generation prediction model correction and online adaptive power generation prediction model correction.

[0020] In this embodiment, as Figure 2 As shown, Figure 2The flowchart of the photovoltaic power generation prediction and analysis method based on multi-source data provided in this application embodiment describes the steps of the photovoltaic power station operation in high-altitude areas. First, the photovoltaic power generation environmental parameters of each monitoring point are acquired and the environmental disturbance value is analyzed. Then, the environmental disturbance label is determined based on the threshold. If there is an environmental disturbance, the photovoltaic inverter parameters are adjusted, including calculating the photovoltaic power generation impact difference, matching reference samples in the database and calculating the photovoltaic inverter processing coefficient, and adjusting the MPPT sampling interval and disturbance step size. Next, the real-time meteorological environmental parameter set is updated, and the multi-source data training set corresponding to the power generation cycle to be predicted is acquired. The training set and real-time meteorological parameters are input into the power generation prediction model, and the predicted power generation is output. Subsequently, the air pressure and air density data at each moment are analyzed to obtain the air pressure change label and the light refraction feature label, and the predicted value is corrected in combination with the constraint factor. Finally, the deviation between the predicted value and the actual power generation is calculated, the power generation prediction label is determined based on the preset threshold, and offline or online adaptive correction is performed to achieve gradual convergence of the model.

[0021] In the process of predicting the power generation of photovoltaic power plants, the prediction accuracy of the model depends not only on the quality of the input environmental parameters, but also on whether the model parameters can be adaptively updated according to the dynamic changes in the operating environment. Since the operating environment of photovoltaic power plants is usually affected by the combined effects of multiple environmental factors such as air pressure, temperature, and air density, especially in mountainous or complex terrain areas, the changes in air pressure and air density gradients caused by altitude differences can lead to periodic fluctuations in solar conversion efficiency, making the power generation prediction model prone to parameter drift and prediction bias after long-term operation. Therefore, this scheme, after obtaining the predicted power generation value based on multi-source data, further introduces two training methods for the power generation prediction model: offline calibration and online adaptive calibration. The core objective is to ensure that the prediction model maintains stable prediction accuracy under different environmental disturbances through continuous iterative updates of model parameters and dynamic weight adjustments.

[0022] Offline calibration primarily utilizes historical power generation data, environmental parameter data, and long-term operational error records for global optimization. Through offline calibration, the nonlinear relationship between input parameters and power generation can be refitted over a large timescale, thereby eliminating model mismatch issues caused by long-term environmental changes (such as seasonal climate differences and equipment aging). The advantages of offline calibration include ample computational resources, the ability to search for optimal solutions in the model's global parameter space, and the ability to effectively improve the model's long-term stability and global generalization capability.

[0023] Online adaptive calibration continuously receives environmental data from various monitoring points during the photovoltaic power plant's power generation process. It dynamically adjusts model parameters based on the deviation between the current predicted and actual power generation, achieving rapid response and local optimization. By introducing a covariance update mechanism, forgetting factor control, and calibration gain vector calculation, the online calibration model can adjust weights according to real-time environmental fluctuations, gradually converging the prediction error. Its advantages lie in its ability to quickly adapt to sudden weather disturbances and changes in equipment operating status, ensuring dynamic consistency between predicted and actual power generation.

[0024] A robust closed-loop relationship is established between model training and power generation prediction in the power generation forecasting model. During the prediction phase, predicted power generation values ​​are generated based on the current model parameters, while the training phase utilizes prediction error feedback to adjust the model parameters. This closed loop enables the model to continuously learn, maintaining prediction stability and convergence characteristics even under the influence of multi-source data disturbances, thereby achieving accurate prediction and dynamic optimization of photovoltaic power generation.

[0025] The database is used to store various historical data (such as historical reference datasets, which refer to photovoltaic power generation data in various regions under the same altitude conditions, as well as photovoltaic power generation data in the region where the photovoltaic power generation equipment is located or adjacent regions), various thresholds, various reference sets, real-time detection data, and equipment operation logs, etc. Among them, the historical data can be obtained through big data, the thresholds are specifically the state critical values ​​in the historical data as thresholds, and the reference sets refer to the set of optimal values ​​of parameters in the historical data.

[0026] Furthermore, the environmental disturbance value is obtained through the following method: Photovoltaic power generation environmental parameters are acquired at each monitoring point of the photovoltaic power station within a preset time period. These parameters include horizontal illuminance, ambient temperature, average surface temperature of the photovoltaic panel, and short-term irradiance variation rate. A preset photovoltaic power generation environmental standard set is obtained from the database, including standard values ​​for horizontal illuminance, ambient temperature, average surface temperature of the photovoltaic panel, and short-term irradiance variation rate. The photovoltaic power generation environmental parameters detected at each monitoring point are averaged to obtain the mean values. The mean values ​​are then divided by the corresponding photovoltaic power generation environmental standard set after absolute difference processing to obtain the results of each degree of difference processing. These results are multiplied by the corresponding directionality coefficient to obtain the results of each direction processing. It is then determined whether each direction processing result is greater than zero. If it is greater than zero, the corresponding spatial weight is introduced for spatial merging to obtain the environmental disturbance value; otherwise, the direction processing result is ignored.

[0027] In this embodiment, the photovoltaic power generation environmental parameters include horizontal illuminance, ambient temperature, average surface temperature of the photovoltaic panel (obtained by averaging the temperatures at the center points of each photovoltaic panel detected by each monitoring point), and short-term irradiance variation rate, which can be obtained by querying the computer background data management system. The computer background data management system is connected to various sensors (including digital illuminance meters, digital temperature sensors, infrared thermometers, and photovoltaic irradiance sensors). Each sensor detects the actual photovoltaic power generation environment and transmits the detected data values ​​to the computer background data management system for storage after detection.

[0028] The environmental disturbance value is obtained by the following method: ; In the formula, This represents the environmental disturbance value at time t, where t represents the time number, t=1,2,...,t max ,t max This indicates the final time step, where i represents the number of the photovoltaic power generation environmental parameter, i=1,2,3,4, γ i γ represents the weighting factor for the i-th photovoltaic power generation environmental parameter, γ1 represents the horizontal illuminance weighting factor, γ2 represents the ambient temperature weighting factor, γ3 represents the average surface temperature of the photovoltaic panel weighting factor, γ4 represents the short-time irradiance change rate weighting factor, max represents taking the maximum value, and s i s represents the directionality coefficient of the i-th photovoltaic power generation environmental parameter, where s is the coefficient when i=1. i =-1, when i=2, 3 or 4 s i =1, x i,j Let x represent the photovoltaic power generation environmental parameter of the i-th monitoring point at time t, where j = 1, 2, 3, ..., J, and J represents the total number of monitoring points. i,ref This represents the standard value of the i-th photovoltaic power generation environmental parameter.

[0029] The horizontal illuminance weighting factor is obtained based on the horizontal illuminance data collected by the real-time illuminance sensor. First, illuminance records under the same time period and the same lighting conditions in the past week are retrieved from the database to establish a time series correspondence between the current illuminance and the historical illuminance. The difference in the fluctuation range between the current illuminance series and the historical series within the same sampling interval is used as the basis for comparison. The ratio between the illuminance change range per unit time and the historical average change range is calculated. The result of the time moving average processing of this ratio is taken as the value of the horizontal illuminance weighting factor. The ambient temperature weighting factor is obtained based on the regression analysis results of the inverter's operating efficiency on the sensitivity of ambient temperature changes. It is obtained by extracting the power deviation rate within different temperature ranges, calculating the slope of the temperature deviation versus power loss rate, and then linearly normalizing the result to obtain the ambient temperature weighting factor. The average surface temperature weighting factor of the photovoltaic panel is derived from temperature sequence data collected by the panel temperature sensor (thermocouple). The Pearson correlation coefficient between the temperature fluctuation amplitude and the photovoltaic output current fluctuation rate within adjacent time windows is statistically analyzed, and the average of multiple samples is used to obtain the average surface temperature weighting factor of the photovoltaic panel. The short-term irradiance change rate weighting factor is obtained based on short-term sampling data from the irradiance sensor. The rate of change of irradiance at adjacent moments is calculated through continuous sampling periods (e.g., every 5 or 10 seconds) and compared with typical irradiance rate samples recorded in the database to obtain the rate difference. The rate difference is then processed by moving average over multiple continuous sampling periods to eliminate occasional fluctuation interference, and the smoothed average ratio is used as the value of the short-term irradiance change rate weighting factor. It should be noted that after obtaining the values ​​of the above four weighting factors (horizontal illuminance weighting factor, ambient temperature weighting factor, average surface temperature of photovoltaic panels weighting factor, and short-term irradiance change rate weighting factor), a normalization process is required. This means that the four weighting factors are allocated according to the proportion of their respective values ​​to the total value, so that the sum of the values ​​of the four weighting factors is 1.

[0030] It should be noted that the above four weighting factors are automatically generated based on historical statistical samples in the database during the system initialization phase, and are adaptively corrected based on real-time monitoring data during subsequent operation, thereby ensuring that each weighting factor has reproducibility and dynamic adaptability.

[0031] By analyzing environmental parameters of photovoltaic power generation, including horizontal illuminance, ambient temperature, average surface temperature of photovoltaic panels, and short-term irradiance variation rate, environmental disturbance values ​​are obtained. This analysis considers the interrelationships between these parameters. For example, horizontal illuminance directly determines the incident solar energy intensity and is the primary factor affecting the output power of photovoltaic modules. However, when the ambient temperature rises, it leads to a decrease in air density and an increase in module surface temperature, thereby reducing the open-circuit voltage of the photovoltaic cells and consequently decreasing the conversion efficiency. Simultaneously, the increase in the average surface temperature of the photovoltaic panels alters the radiative heat dissipation balance of the modules, causing a shift in the response characteristics of the short-term irradiance variation rate, manifesting as a lag or amplification in the response to changes in illuminance. Furthermore, drastic fluctuations in the short-term irradiance variation rate (such as rapid cloud cover) not only cause transient changes in illuminance but also lead to periodic fluctuations in module surface temperature, creating localized thermal stress differences and further affecting the stability of power generation output.

[0032] By simultaneously collecting and analyzing the horizontal illuminance, ambient temperature, average surface temperature of photovoltaic panels, and short-term irradiance variation rate at multiple monitoring points in a photovoltaic power plant, a quantitative characterization of environmental disturbances in photovoltaic power generation is achieved. By constructing a standard set of photovoltaic power generation environmental data and conducting multi-dimensional difference analysis between the real-time collected parameters and the standard set, the comprehensive impact of power plant environmental fluctuations on power generation prediction can be reflected, enabling early identification and correction of environmental disturbances during the model input stage. Furthermore, by introducing directional coefficients and spatial weights, a comprehensive spatial mapping of environmental disturbance values ​​is achieved. This allows for the dynamic quantification of the directionality of environmental changes (such as irradiance attenuation or local shading) and regional differences (such as temperature gradients caused by terrain), thereby enabling dynamic correction of the input data to the photovoltaic power generation prediction model. This significantly improves the stability and accuracy of power generation prediction, ultimately achieving intelligent prediction and adaptive control of photovoltaic power plants under complex environmental conditions.

[0033] Furthermore, the environmental disturbance label is obtained by: obtaining a preset environmental disturbance threshold in the database and comparing it with the environmental disturbance value to obtain the environmental disturbance label. If the environmental disturbance value is above the environmental disturbance threshold, the environmental disturbance label is environmental disturbance; otherwise, the environmental disturbance label is environmental stability.

[0034] In this embodiment, by performing a stability analysis of the photovoltaic power plant's operating environment before power generation prediction, setting an environmental disturbance threshold, and classifying the disturbance values ​​based on real-time calculations, the power plant's operating status is categorized into two types: "environmental disturbance" and "environmental stability," thus achieving quantitative identification of the degree of external disturbance to photovoltaic power generation. This analysis allows for the determination of whether there are drastic fluctuations in current weather conditions before the prediction model is implemented, thereby deciding whether to adjust photovoltaic inverter parameters and correct model inputs. This avoids the error amplification problem caused by directly performing power generation prediction under drastic environmental changes, achieving pre-screening of the dynamic characteristics of the photovoltaic power plant's operating environment.

[0035] Furthermore, the photovoltaic inverter parameters are adjusted using the following method: The difference between the environmental disturbance value and the environmental disturbance threshold is processed to obtain the photovoltaic power generation impact difference; the photovoltaic power generation impact difference is matched with the database to obtain the photovoltaic inverter processing coefficient; the MPPT sampling interval of the photovoltaic inverter is adjusted by decreasing the photovoltaic inverter processing coefficient; the real-time power generation during the preset second operating period after the MPPT sampling interval is adjusted is obtained, and the power generation improvement value is analyzed. The power generation improvement value is compared with the preset power generation improvement threshold in the database. If the power generation improvement value is above the power generation improvement threshold, the photovoltaic inverter parameter adjustment is completed; otherwise, based on the power generation improvement value and the current altitude value, the comprehensive adjustment value of the MPPT disturbance step size is obtained, and the MPPT disturbance step size is increased.

[0036] In this embodiment, the difference between the environmental disturbance value and the environmental disturbance threshold is processed (the environmental disturbance value is subtracted from the environmental disturbance threshold) to obtain the difference in the impact of photovoltaic power generation.

[0037] The photovoltaic (PV) inverter processing coefficient is obtained by matching the PV power generation impact difference with a database. Specifically, a set of known PV power generation impact difference samples and their corresponding PV inverter processing coefficients stored in the database is formed. The database is then searched for PV power generation impact difference samples that are closest to the current PV power generation impact difference (the difference between the current PV power generation impact difference sample and the current PV power generation impact difference is less than the power generation impact sample error difference). These samples are marked as PV power generation impact difference reference sample values, thus obtaining a PV power generation impact difference reference sample value range. Based on the proportion of the current PV power generation impact difference within this range, the PV inverter processing coefficients at both ends of the range are weighted proportionally to obtain the current PV inverter processing coefficient. If the current PV power generation impact difference is less than the minimum reference sample value in the database, the PV inverter processing coefficient of the minimum reference sample is used; if it is greater than the maximum reference sample value in the database, the PV inverter processing coefficient of the maximum reference sample is used.

[0038] The MPPT sampling interval of the photovoltaic inverter is reduced based on the photovoltaic inverter processing coefficient. Specifically, the MPPT sampling interval of the photovoltaic inverter is reduced by subtracting the product of the MPPT sampling interval and the photovoltaic inverter processing coefficient from the current MPPT sampling interval of the photovoltaic inverter.

[0039] The improvement value of power generation is obtained as follows: the photovoltaic inverter is operated with the MPPT sampling interval reduced and the real-time power generation during the second operating period (e.g., 1 minute) is obtained. The time period with the same duration as the second operating period before the MPPT sampling interval of the photovoltaic inverter was reduced and adjusted is marked as the comparison period (e.g., 1 minute). The photovoltaic power generation during the comparison period is obtained. The real-time power generation during the second operating period is subtracted from the photovoltaic power generation during the comparison period to obtain the improvement value of power generation.

[0040] Based on the power generation improvement value and the current altitude value of the region, the comprehensive adjustment value of the MPPT perturbation step size is obtained. The specific method is as follows: subtract the power generation improvement threshold from the power generation improvement value to obtain the power generation improvement difference value; divide the power generation improvement difference value by the power generation improvement threshold to obtain the power generation improvement difference degree value; match the power generation improvement difference degree value with the database to obtain the MPPT perturbation step size increase adjustment value; obtain the current altitude value of the region and match it with the database to obtain the altitude correction constraint factor; amplify the MPPT perturbation step size increase adjustment value based on the altitude correction constraint factor to obtain the altitude-corrected MPPT perturbation step size increase adjustment value, and mark it as the comprehensive adjustment value of the MPPT perturbation step size. Therefore, the current MPPT perturbation step size is increased and adjusted (additive processing) based on the comprehensive adjustment value of the MPPT perturbation step size.

[0041] Based on matching the power generation improvement difference value with the database, the MPPT perturbation step size adjustment value is obtained. Specifically, the method is as follows: A set of known power generation improvement difference value samples and corresponding historical MPPT perturbation step size adjustment values ​​stored in the database are formed. Then, the database is searched for power generation improvement difference value samples that are close to the current power generation improvement difference value (the difference between the current power generation improvement difference value sample and the current power generation improvement difference value is less than the improvement sample error difference), and these are marked as reference sample values ​​for power generation improvement difference value. This yields the reference sample value range for power generation improvement difference value. Based on the current power generation improvement... The proportion of the difference value within the reference sample value range of the difference value of power generation improvement is used to obtain the current MPPT disturbance step size increase adjustment value reference value by proportionally weighting the historical MPPT disturbance step size increase adjustment values ​​at both ends of the reference sample value range of the difference value of power generation improvement, and this value is used as the MPPT disturbance step size increase adjustment value. If the current difference value of power generation improvement is less than the minimum reference sample value in the database, the historical MPPT disturbance step size increase adjustment value of the minimum reference sample is taken as the MPPT disturbance step size increase adjustment value. If it is greater than the maximum reference sample value in the database, the historical MPPT disturbance step size increase adjustment value of the maximum reference sample is taken as the MPPT disturbance step size increase adjustment value.

[0042] By adaptively adjusting the parameters of the photovoltaic inverter when environmental disturbances exceed a threshold, the dynamic response capability of the system under complex weather conditions can be effectively improved. Specifically, by calculating the difference in the impact of photovoltaic power generation and matching it with the photovoltaic inverter processing coefficient, the sampling interval of maximum power point tracking (MPPT) is reduced and the MPPT disturbance step size is increased. This allows the inverter to respond more quickly to transient changes in environmental parameters such as irradiance, temperature, and wind speed, thereby shortening the power point tracking lag time, reducing energy loss, and avoiding the accumulation of tracking errors caused by excessively low sampling frequency or insufficient disturbance amplitude. This achieves dynamic and steady-state balanced control of the photovoltaic inverter under unstable environmental conditions, improving the power generation stability and real-time power output accuracy of the photovoltaic power plant.

[0043] Furthermore, a multi-source data training set for the power generation cycle to be predicted is obtained. Specifically, the following methods are employed: First, a pre-defined historical reference dataset is acquired from the database, including historical reference data for each time period. Second, the power generation cycle to be predicted is matched with each stored prediction cycle to obtain the corresponding historical reference dataset. Third, based on environmental disturbance value analysis, if the environmental disturbance value is above the environmental disturbance threshold, a historical data demand reference value is obtained by matching the environmental disturbance value with the database and marked as the historical data demand. If the environmental disturbance value is below the environmental disturbance threshold, a pre-defined historical data demand baseline value is acquired from the database and marked as the historical data demand. Fourth, the data in the historical reference dataset corresponding to the power generation cycle to be predicted is sorted according to time order (the order in which the times occur), and a pre-defined number of data points closest to the current time are selected. This pre-defined number of data points is then used as the multi-source data training set for the power generation cycle to be predicted. The pre-defined number represents the historical data demand.

[0044] In this embodiment, as Figure 3 As shown, Figure 3 The flowchart for acquiring the multi-source data training set of the photovoltaic power generation prediction and analysis method based on multi-source data provided in this application embodiment is as follows: First, the power generation cycle to be predicted is acquired as the target time period for power generation data analysis; a preset historical reference dataset is retrieved from the database as the basic sample for multi-source data training; the power generation cycle to be predicted is compared with multiple historical prediction cycles stored in the database to determine the corresponding historical reference dataset that is closest to the target cycle; after matching, the current environmental disturbance value is detected and it is determined whether it is higher than a preset threshold. If the environmental disturbance value exceeds the threshold, it indicates that the current power generation cycle is in an unstable state, and the historical data demand is calculated; if the environmental disturbance value does not exceed the threshold, the preset historical data demand benchmark value in the database is directly called as the reference standard to obtain the historical data demand. Then, according to the obtained historical data demand, the corresponding historical reference dataset is sorted in chronological order, and the preset quantity of data closest to the current time is selected to form the multi-source data training set for the power generation cycle to be predicted.

[0045] The historical reference dataset corresponding to the expected power generation cycle is obtained by matching the expected power generation cycle with each storage prediction cycle (year, month, week, day, etc.). Specifically, the expected power generation cycle (e.g., the next day) is matched with each storage prediction cycle (year, month, week, day, etc.). If the expected power generation cycle is the next day, the historical reference dataset corresponding to the expected power generation cycle may include data for the prediction day in previous years as well as data for the days before the prediction day.

[0046] It should be noted that each storage prediction period includes years, months, weeks, days, etc. Therefore, the historical reference dataset corresponding to each storage prediction period is composed of the historical reference data corresponding to each historical moment.

[0047] Based on matching environmental disturbance values ​​with the database, a reference value for historical data demand is obtained. Specifically, a set of known environmental disturbance value samples and their corresponding historical data demands stored in the database is formed. An environmental disturbance value sample that is closest to the current environmental disturbance value (the difference between the current environmental disturbance value sample and the current environmental disturbance value is less than the environmental disturbance sample error difference) is searched in the database and marked as the environmental disturbance value reference sample value. This yields an environmental disturbance value reference sample value range. Based on the proportion of the current environmental disturbance value within this range, the historical data demands at both ends of the range are weighted proportionally to obtain the current historical data demand reference value. If the current environmental disturbance value is less than the minimum reference sample value in the database, the historical data demand of the minimum reference sample is used; if it is greater than the maximum reference sample value in the database, the historical data demand of the maximum reference sample is used.

[0048] By constructing a multi-source data training set for the power generation cycle to be predicted, dynamic data filtering and adaptive training optimization under different environmental conditions are achieved. The historical reference dataset is a complete data set covering multiple historical cycles, different meteorological conditions, and multi-dimensional operating parameters in the database, with a wide range and rich information dimensions. By matching and filtering this historical reference dataset according to the power generation cycle to be predicted, only the data with the time characteristics closest to the current power generation cycle can be extracted, which can effectively avoid the interference of irrelevant historical data, improve the model's response accuracy to periodic patterns, and thus achieve targeted prediction of photovoltaic power generation output. When the environmental disturbance value is higher than the threshold, the minimum data sampling amount can be determined by matching the historical data requirement reference value obtained from the database, so as to ensure sufficient data coverage even under significant environmental fluctuations, thereby enhancing the robustness of training and the generalization ability of the prediction model, while avoiding computational redundancy and response lag caused by using all data for training. For example, under conditions of sudden cloud cover changes, training and prediction can be quickly completed using only historical data with similar disturbance characteristics. When the environmental disturbance value is below a threshold, data is screened based on a preset baseline value for the required amount of historical data to ensure that the number of input samples for the model is appropriate and to prevent insufficient training or model non-convergence due to insufficient data. This achieves adaptive construction of multi-source data training sets, improving data processing efficiency and algorithm execution performance while ensuring prediction accuracy.

[0049] Furthermore, pressure change labels and light refraction feature labels are obtained. Specifically, the following methods are used: First, the air pressure value and air density at each moment within a preset time period are obtained, thus yielding the air pressure change rate. Second, the air pressure change rate is compared with a preset air pressure change rate threshold in the database to obtain an air pressure change label. If the air pressure change rate is above the threshold, the label indicates air pressure disturbance; otherwise, it indicates stable air pressure. Third, based on the analysis of air pressure and air density at each moment, the change in the ratio of air density to air pressure is obtained and marked as an abnormal value of light refraction fluctuation. Fourth, the abnormal value of light refraction fluctuation is compared with a preset abnormal threshold of light refraction fluctuation in the database to obtain a light refraction feature label. If the abnormal value is less than the abnormal threshold, the label indicates normal light refraction; otherwise, it indicates abnormal light refraction.

[0050] In this embodiment, it should be noted that the air pressure value can be obtained in the computer background data management system, which is connected to a digital air pressure sensor, an air pressure sensor, and a temperature and humidity sensor. These sensors can perform real-time data detection and upload the detected data to the computer background data management system for storage. The air density can be calculated by combining the air pressure and temperature measured by the air pressure sensor and the temperature and humidity sensor with the air constant and applying the ideal gas law.

[0051] It should be noted that the pressure change value is obtained by subtracting the maximum and minimum air pressure values ​​at each moment within the preset time period. Dividing this pressure change value by the duration of the preset time period yields the pressure change rate. The specific calculation method for the anomaly value of light refraction fluctuation is as follows: the air density and air pressure values ​​at each moment are compared to obtain the ratio result. The absolute value of the difference between the ratio result at the end of the time period and the ratio result at the beginning of the time period is then used to obtain the anomaly value of light refraction fluctuation.

[0052] By acquiring air pressure change tags and light refraction feature tags, a refined identification and quantitative characterization of the variation patterns of light characteristics under complex mountainous terrain conditions was achieved. Due to the significant undulations and frequent altitude changes in mountainous terrain, the local air pressure and air density distribution are extremely uneven, leading to variations in refractive index and energy attenuation during light propagation. Failure to identify and calibrate these differences will directly affect the effective irradiance received by photovoltaic modules, causing deviations in power generation predictions. By analyzing the trends in air pressure and air density at various times, air pressure change tags are obtained, which can be used to determine whether current environmental air pressure fluctuations are significant. Simultaneously, by calculating the change in the ratio of air density to air pressure and generating light refraction feature tags, abnormal light refraction phenomena caused by differences in air rarefaction can be identified, thus providing data support for correcting light conversion efficiency. Therefore, by analyzing the rate of change in air pressure and abnormal values ​​of light refraction fluctuations, dynamic discrimination of the light propagation environment can be achieved in areas with significant altitude changes, facilitating more accurate correction of power generation predictions.

[0053] Furthermore, the meteorologically corrected power generation forecast is obtained using the following method: Based on pressure change label analysis, a pressure change constraint factor is obtained. If the pressure change label is "stable pressure," the pressure change constraint factor is 1; otherwise, the absolute difference between the pressure change rate and the pressure change rate threshold is processed and matched with the database to obtain the pressure change constraint factor. Based on illumination refraction feature label analysis, an illumination refraction constraint factor is obtained. If the illumination refraction feature label is "normal illumination refraction," the illumination refraction constraint factor is 1; otherwise, the absolute difference between the illumination refraction fluctuation anomaly value and the illumination refraction fluctuation anomaly threshold is processed and matched with the database to obtain the illumination refraction constraint factor. The pressure change constraint factor and the illumination refraction constraint factor are coupled to obtain the meteorological disturbance constraint factor. Based on the meteorological disturbance constraint factor, the power generation forecast for the forecast power generation cycle is corrected to obtain the meteorologically corrected power generation forecast (the meteorologically corrected power generation forecast is obtained by subtracting the product of the power generation forecast for the forecast power generation cycle and the meteorological disturbance constraint factor from the power generation forecast for the forecast power generation cycle).

[0054] In this embodiment, the absolute difference between the rate of change of air pressure and the threshold of the rate of change of air pressure is processed and matched with the database to obtain the air pressure change constraint factor. Specifically, the method is as follows: The absolute difference between the rate of change of air pressure and the threshold of the rate of change of air pressure is processed to obtain the absolute difference of the rate of change of air pressure. A set of known absolute difference samples of the rate of change of air pressure and their corresponding historical air pressure change constraint factors stored in the database are formed. The database is then searched for absolute difference samples of the rate of change of air pressure that are closest to the current absolute difference (the difference between the current absolute difference sample of the rate of change of air pressure and the current absolute difference of the rate of change of air pressure is less than the error difference of the air pressure change sample), and these are marked as absolute differences of the rate of change of air pressure. The difference reference sample value is used to obtain the reference sample value range of the absolute difference of the air pressure change rate. Based on the proportion of the current absolute difference of the air pressure change rate within the reference sample value range, the historical air pressure change constraint factors at both ends of the reference sample value range are weighted proportionally to obtain the current air pressure change constraint factor reference value, which is used as the air pressure change constraint factor. If the current absolute difference of the air pressure change rate is less than the minimum reference sample value in the database, the historical air pressure change constraint factor of the minimum reference sample is taken as the air pressure change constraint factor; if it is greater than the maximum reference sample value in the database, the historical air pressure change constraint factor of the maximum reference sample is taken as the air pressure change constraint factor.

[0055] After processing the absolute difference between outliers and threshold values ​​of light refraction fluctuations, and matching them with a database, light refraction constraint factors are obtained. Specifically, the method involves processing the absolute difference between outliers and threshold values ​​of light refraction fluctuations to obtain the absolute difference in light refraction fluctuations. A set of known absolute difference samples of light refraction fluctuations and their corresponding historical light refraction constraint factors stored in the database is formed. Then, a sample of absolute difference in light refraction fluctuations that is closest to the current absolute difference in light refraction fluctuations (the difference between the current absolute difference sample and the current absolute difference in light refraction fluctuations is less than the sample error difference) is searched in the database and marked as a light refraction fluctuation constraint factor. The absolute difference of light refraction fluctuation is used as a reference sample value to obtain the range of absolute difference of light refraction fluctuation. Based on the proportion of the current absolute difference of light refraction fluctuation within the range of reference sample values, the historical light refraction constraint factors at both ends of the range are weighted proportionally to obtain the current reference value of the light refraction constraint factor, which is then used as the light refraction constraint factor. If the current absolute difference of light refraction fluctuation is less than the minimum reference sample value in the database, the historical light refraction constraint factor of the minimum reference sample is taken as the light refraction constraint factor; if it is greater than the maximum reference sample value in the database, the historical light refraction constraint factor of the maximum reference sample is taken as the light refraction constraint factor.

[0056] By introducing air pressure change constraint factors and light refraction constraint factors, and multiplying them, a meteorological disturbance constraint factor is obtained, thereby achieving dynamic correction of the power generation prediction value. When the air pressure change label is determined to be stable or the light refraction characteristic label is determined to be normal, the corresponding constraint factor is set to 1, indicating that the impact of environmental changes on light propagation and energy conversion is negligible, and the power generation prediction value does not need to be corrected. However, when there are air pressure fluctuations or abnormal light refraction, the corresponding constraint factor is obtained by performing absolute difference analysis on the abnormal values ​​of air pressure change rate or light refraction fluctuations and their corresponding thresholds, and matching them with the database. This allows for targeted correction of the power generation prediction value. This constraint factor can quantitatively reflect the comprehensive impact of air pressure disturbances and air density changes caused by altitude changes on the incident angle of light, refraction path, and light energy transfer efficiency, thereby achieving adaptive correction of the power generation prediction value in different altitude ranges. This effectively avoids the systematic errors caused by the failure to consider altitude-related environmental fluctuations in traditional models.

[0057] Furthermore, the power generation prediction model is corrected, and the judgment criteria are as follows: First, the preset first deviation threshold and second deviation threshold for power generation prediction are obtained from the database; the actual power generation for the predicted power generation cycle is obtained (obtained through a computer backend management system; the power generation data of photovoltaic panels is uploaded to the computer backend management system in real time and stored), and the difference between this value and the meteorologically corrected power generation prediction value is analyzed (the absolute value of the difference between the meteorologically corrected power generation prediction value and the actual power generation is taken, and then the absolute difference is divided by the actual power generation), to obtain the power generation prediction deviation value; based on the power generation prediction deviation value, the first deviation threshold and the second deviation threshold for power generation prediction are respectively compared with the actual power generation prediction. A second deviation threshold is used to compare the predicted power generation with the predicted power generation. If the predicted power generation deviation is less than the first deviation threshold, the predicted power generation is labeled as normal, and no power generation prediction model correction is performed. If the predicted power generation deviation is above the first deviation threshold but less than the second deviation threshold, the predicted power generation is labeled as first deviation, and offline correction of the power generation prediction model is performed. If the predicted power generation deviation is above the second deviation threshold, the predicted power generation is labeled as second deviation, and online adaptive correction of the power generation prediction model is performed. The first and second deviations are jointly labeled as predicted power generation deviation.

[0058] In this embodiment, it should be noted that the predicted deviation value of power generation is obtained. For example, the predicted power generation in the next hour is obtained by analyzing the deviation between the predicted power generation in the next hour and the actual power generation one hour later. By analyzing the predicted deviation value of power generation, the accuracy of the power generation prediction model can be processed and verified more quickly. If it is not accurate, it can be trained more quickly, avoiding the inaccurate predicted power generation obtained by using an inaccurate power generation prediction model.

[0059] During offline training, the power generation prediction data and corresponding photovoltaic (PV) power generation environmental parameters are first collected. The PV power generation environmental parameters at each time point are used to construct a design matrix. The residual vector is obtained by subtracting the baseline power generation from the actual measured power generation at each time point. The optimal residual weight vector after offline training is obtained by multiplying the transpose of the design matrix by itself, adding a pre-defined regularization coefficient multiplied by the identity matrix, and then inverting the result. Finally, this result is multiplied by the transpose of the design matrix and the residual vector. This optimal residual weight vector represents the optimal weight of each PV power generation environmental parameter in the offline model and is used to calculate the residual of the predicted power generation. The regularization coefficient is used to control the magnitude of the weights to avoid overfitting. The dimension of the identity matrix is ​​consistent with the number of input parameters.

[0060] The offline calibration method for power generation prediction models is as follows: Offline regression target: ; In the formula, w * denoted as the offline optimal weight vector, which is the analytical solution obtained by fitting historical samples, and is the final model parameter obtained by offline training. w represents the residual basis function weight vector, r represents the residual target vector, φ represents the design matrix, which is composed of the residual basis functions at each time step, and α represents the L2 regularization coefficient.

[0061] Analytical solution (ridge regression formula): ; In the formula, I K This represents a K-dimensional identity matrix.

[0062] Further online adaptive calibration of the power generation prediction model includes the following steps: During the real-time operation of the photovoltaic power plant, photovoltaic power generation environmental parameters from various monitoring points are continuously received. These parameters are then input into the power generation prediction model and matched with the parameters stored in the previous time step, thus forming the observation relationship for the current moment. This observation relationship describes the difference between the predicted and actual power generation at the current moment. Within each time step, the deviation between the predicted and actual power generation is calculated (subtracted), and this deviation is considered an error signal. This is used for covariance updating to obtain the updated historical covariance matrix. A forgetting factor (ranging from 0.98 to 0) is set during the covariance update process. .995 (which can be automatically adjusted through cross-validation); the updated results of the historical covariance matrix are used to measure the correlation between each input parameter and its impact on prediction accuracy; the forgetting factor is used to control the weight decay of historical data during the model update process; in each round of covariance update, the correction gain vector is calculated based on the real-time deviation, and the correction gain vector is used to determine the weight correction magnitude of each photovoltaic power generation environmental parameter during the model update process; based on the correction gain vector, the parameter vector of the power generation prediction model is iteratively corrected step by step through the recursive least squares method, so that the power generation prediction value output by the power generation prediction model gradually approaches the actual power generation. When the prediction error change rate is lower than the preset convergence threshold after multiple consecutive iterations, it is determined that the power generation prediction model has reached the convergence state, and the parameter update of the current round is stopped.

[0063] In this embodiment, the formula for the observation relationship at the current moment is: ; In the formula, y t x represents the measured power generation at the current moment. t θ represents the input feature vector at the current time. t-1 e represents the parameter vector of the power generation prediction model at the previous moment. t This represents the deviation between the predicted power generation and the actual power generation at the current moment, with T indicating the transpose sign.

[0064] The correction gain vector is calculated based on the covariance matrix and forgetting factor from the previous time step. The specific method is as follows: ; In the formula, K t Let Pt-1 represent the correction gain vector, Pt-1 represent the covariance matrix of the previous time step, and λ represent the forgetting factor.

[0065] The parameters are corrected using the gain vector, specifically as follows: ; In the formula, θ tThis represents the updated model parameter vector at the current time.

[0066] The covariance matrix is ​​iteratively updated using the following method: ; In the formula, P t This represents the covariance matrix at the current time.

[0067] like Figure 4 The diagram shows a module schematic of a photovoltaic power plant power generation prediction and analysis system based on multi-source data provided in this application embodiment. The system includes: an environmental disturbance assessment module, a photovoltaic inverter adjustment module, a power generation prediction module, a power generation correction module, and a model correction module. The environmental disturbance assessment module acquires the photovoltaic power generation environmental parameters of each monitoring point at a photovoltaic power plant located at high altitude within a preset time period, analyzes the environmental disturbance value, and obtains an environmental disturbance label. The photovoltaic inverter adjustment module adjusts the photovoltaic inverter parameters if the environmental disturbance label indicates environmental disturbance; otherwise, it does not perform the adjustment and updates the acquired real-time meteorological environmental parameter set. The power generation prediction module acquires the power generation cycle to be predicted and analyzes it in conjunction with the environmental disturbance value. The system obtains a multi-source data training set for the power generation cycle to be predicted, and inputs it together with a real-time meteorological environmental parameter set into the power generation prediction model to output the predicted power generation value for the cycle. The power generation correction module is used to obtain the air pressure value and air density at each moment within a preset time period, and analyzes them to obtain air pressure change labels and light refraction feature labels, thereby obtaining the meteorologically corrected power generation prediction value. The model correction module is used to analyze the power generation prediction label based on the meteorologically corrected power generation prediction value. If the power generation prediction label indicates a deviation from the power generation prediction, the power generation prediction model correction is performed; otherwise, the power generation prediction model correction is not performed. The power generation prediction model correction includes offline power generation prediction model correction and online adaptive power generation prediction model correction.

[0068] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0069] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0070] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0071] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0072] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0073] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for predicting and analyzing the power generation of a photovoltaic power plant based on multi-source data, characterized in that, Includes the following steps: S1. Obtain the photovoltaic power generation environment parameters of each monitoring point of the photovoltaic power station located in a high-altitude area within a preset time period, analyze and obtain the environmental disturbance value, thereby obtaining the environmental disturbance label. The environmental disturbance label includes environmental disturbance and environmental stability. The environmental disturbance value is used to characterize the degree of influence of the power generation environment on the accuracy of photovoltaic power generation prediction. S2. If the environmental disturbance label is "environmental disturbance", then the photovoltaic inverter parameters will be adjusted; otherwise, the adjustment will not be performed, and the real-time meteorological and environmental parameter set will be updated. S3. Obtain the power generation cycle to be predicted, and combine it with the analysis of environmental disturbance values ​​to obtain the multi-source data training set of the power generation cycle to be predicted. Then, input it together with the real-time meteorological and environmental parameter set into the power generation prediction model and output the power generation prediction value of the power generation cycle to be predicted. S4. Obtain the air pressure and air density at each moment within the preset time period, analyze and obtain the air pressure change label and the light refraction characteristic label respectively, and thus obtain the meteorological correction power generation prediction value. S5. Based on the meteorological correction of the power generation forecast value, analyze and obtain the power generation forecast label. If the power generation forecast label is a deviation from the power generation forecast, then perform power generation forecast model correction; otherwise, do not perform power generation forecast model correction. The power generation forecast model correction includes offline power generation forecast model correction and online adaptive power generation forecast model correction.

2. The photovoltaic power generation prediction and analysis method based on multi-source data as described in claim 1, characterized in that: The method for obtaining the environmental disturbance value is as follows: The photovoltaic power generation environmental parameters of each monitoring point of the photovoltaic power station are obtained within a preset time period. The photovoltaic power generation environmental parameters include horizontal illuminance, ambient temperature, average surface temperature of photovoltaic panels, and short-term irradiance variation rate. Obtain a preset set of photovoltaic power generation environment standards from the database. The set of photovoltaic power generation environment standards includes standard values ​​for horizontal illuminance, ambient temperature, average surface temperature of photovoltaic panels, and short-time irradiance variation rate. The photovoltaic power generation environmental parameters detected at each monitoring point are averaged to obtain the average values. The absolute difference between each average value and the corresponding photovoltaic power generation environmental standard set is then calculated and divided by the corresponding photovoltaic power generation environmental standard set to obtain the results of each degree of difference. Each result of the degree of difference is then multiplied by the corresponding directional coefficient to obtain the results of each direction. It is then determined whether the results of each direction are greater than zero. If they are greater than zero, the corresponding spatial weight is introduced for spatial merging to obtain the environmental disturbance value; otherwise, the results of that direction are ignored.

3. The photovoltaic power generation prediction and analysis method based on multi-source data as described in claim 1, characterized in that: The specific method for obtaining the environmental disturbance label is as follows: Obtain the preset environmental disturbance threshold from the database and compare it with the environmental disturbance value to obtain the environmental disturbance label. If the environmental disturbance value is above the environmental disturbance threshold, the environmental disturbance label is environmental disturbance; otherwise, the environmental disturbance label is environmental stability.

4. The photovoltaic power generation prediction and analysis method based on multi-source data as described in claim 1, characterized in that: The specific method for adjusting the parameters of the photovoltaic inverter is as follows: The difference between the environmental disturbance value and the environmental disturbance threshold is used to obtain the difference in the impact of photovoltaic power generation. The photovoltaic inverter processing coefficient is obtained by matching the difference in the impact of photovoltaic power generation with the database. Adjusting the MPPT sampling interval of the photovoltaic inverter based on reducing the photovoltaic inverter processing coefficient; The real-time power generation during the preset second operating period after the MPPT sampling interval duration is adjusted is obtained, and the power generation improvement value is analyzed. The power generation improvement value is compared with the preset power generation improvement threshold in the database. If the power generation improvement value is above the power generation improvement threshold, the photovoltaic inverter parameter adjustment is completed. Otherwise, based on the power generation improvement value and the current altitude value, the comprehensive adjustment value of the MPPT disturbance step size is obtained, and the MPPT disturbance step size is increased.

5. The photovoltaic power generation prediction and analysis method based on multi-source data as described in claim 1, characterized in that: The specific method for obtaining the multi-source data training set for the power generation cycle to be predicted is as follows: Obtain a preset historical reference dataset from the database, the historical reference dataset including historical reference data at each time point; By matching the generation cycle to be predicted with each storage prediction cycle, the historical reference dataset corresponding to the generation cycle to be predicted is obtained. Based on the analysis of environmental disturbance values, if the environmental disturbance value is above the environmental disturbance threshold, the historical data demand reference value is obtained by matching the environmental disturbance value with the database and marked as the historical data demand. If the environmental disturbance value is less than the environmental disturbance threshold, the preset historical data demand benchmark value in the database is obtained and marked as the historical data demand. The data in the historical reference dataset corresponding to the power generation cycle to be predicted are sorted in chronological order, and a preset number of data that are closest to the current time are selected. This preset number of data is then used as the multi-source data training set for the power generation cycle to be predicted. The preset quantity is the historical data requirement.

6. The photovoltaic power generation prediction and analysis method based on multi-source data as described in claim 1, characterized in that: The specific method for obtaining the air pressure change label and the light refraction feature label is as follows: Obtain the air pressure and air density at each moment within a preset time period within a preset range, thereby obtaining the air pressure change rate; Based on the comparison between the rate of change of air pressure and the preset threshold of the rate of change of air pressure in the database, an air pressure change label is obtained. If the rate of change of air pressure is above the threshold, the air pressure change label is air pressure disturbance; otherwise, the air pressure change label is air pressure stability. Based on the analysis of air pressure and air density at various times, the change in the ratio of air density to air pressure is obtained and marked as anomaly value of light refraction fluctuation. The anomaly value of light refraction fluctuation is compared with the preset anomaly threshold of light refraction fluctuation in the database to obtain the light refraction feature label. If the anomaly value of light refraction fluctuation is less than the anomaly threshold of light refraction fluctuation, the light refraction feature label is normal light refraction; otherwise, the light refraction feature label is abnormal light refraction.

7. The photovoltaic power generation prediction and analysis method based on multi-source data as described in claim 1, characterized in that: The specific method for obtaining the meteorologically corrected power generation forecast is as follows: Based on the analysis of air pressure change labels, the air pressure change constraint factor is obtained. If the air pressure change label is air pressure stability, the air pressure change constraint factor is 1. Otherwise, the air pressure change constraint factor is obtained by performing absolute difference processing on the air pressure change rate and the air pressure change rate threshold and matching it with the database. Based on the analysis of light refraction feature labels, the light refraction constraint factor is obtained. If the light refraction feature label is normal light refraction, the light refraction constraint factor is 1. Otherwise, the absolute difference between the light refraction fluctuation anomaly value and the light refraction fluctuation anomaly threshold is processed and matched with the database to obtain the light refraction constraint factor. The meteorological disturbance constraint factor is obtained by coupling the air pressure change constraint factor and the light refraction constraint factor. Based on the meteorological disturbance constraint factor, the predicted power generation value for the predicted power generation cycle is corrected to obtain the meteorological corrected power generation prediction value.

8. The photovoltaic power generation prediction and analysis method based on multi-source data as described in claim 1, characterized in that: The determination condition for performing the power generation prediction model correction is as follows: Obtain the preset first deviation threshold and second deviation threshold of power generation prediction from the database; The actual power generation of the power generation cycle to be predicted is obtained, and the difference between the actual power generation and the meteorologically corrected power generation prediction is analyzed to obtain the power generation prediction deviation value. The power generation prediction deviation is compared with the first deviation threshold and the second deviation threshold of power generation prediction to obtain the power generation prediction label. If the power generation prediction deviation is less than the first deviation threshold of power generation prediction, the power generation prediction label is normal and no power generation prediction model correction is performed. If the deviation of the power generation forecast is above the first deviation threshold of the power generation forecast but less than the second deviation threshold of the power generation forecast, then the power generation forecast is labeled as the first deviation of the power generation forecast, and offline correction of the power generation forecast model is performed. If the deviation of the power generation forecast is above the second deviation threshold of the power generation forecast, the power generation forecast is labeled as the second deviation of the power generation forecast, and online adaptive correction of the power generation forecast model is performed. The first deviation from the power generation forecast and the second deviation from the power generation forecast are jointly labeled as the power generation forecast deviation.

9. The photovoltaic power generation prediction and analysis method based on multi-source data as described in claim 8, characterized in that: The online adaptive correction of the power generation prediction model includes the following steps: During the real-time operation of the photovoltaic power station, photovoltaic power generation environmental parameters from various monitoring points are continuously received. The photovoltaic power generation environmental parameters at the current moment are input into the power generation prediction model and matched with the power generation prediction model parameters stored in the power generation prediction model at the previous moment, thereby forming the observation relationship at the current moment. The observation relationship is used to describe the difference between the predicted power generation and the actual power generation at the current moment. In each time step, the deviation between the predicted power generation and the actual power generation at the current moment is calculated, and this deviation is regarded as an error signal. Covariance is then updated to obtain the updated result of the historical covariance matrix. A forgetting factor is set in the covariance update process. The updated results of the historical covariance matrix are used to measure the correlation between the input parameters and their impact on prediction accuracy. The forgetting factor is used to control the weight decay of historical data during the model update process; In each round of covariance update, a correction gain vector is calculated based on the real-time deviation. The correction gain vector is used to determine the weight correction magnitude of each photovoltaic power generation environmental parameter in the model update process. Based on the correction gain vector, the parameter vector of the power generation prediction model is iteratively corrected step by step using the recursive least squares method, so that the power generation prediction value output by the power generation prediction model gradually approaches the actual power generation. When the rate of change of prediction error is lower than the preset convergence threshold after multiple consecutive iterations, the power generation prediction model is determined to have reached the convergence state, and the parameter update of the current round is stopped.

10. A system applying the photovoltaic power generation prediction and analysis method based on multi-source data as described in any one of claims 1-9, characterized in that, include: The module includes an environmental disturbance assessment module, a photovoltaic inverter adjustment module, a power generation prediction module, a power generation correction module, and a model correction module. The environmental disturbance assessment module is used to obtain the photovoltaic power generation environmental parameters of each monitoring point of the photovoltaic power station located in a high-altitude area within a preset time period, analyze and obtain the environmental disturbance value, and thus obtain the environmental disturbance label. The photovoltaic inverter adjustment module is used to adjust the photovoltaic inverter parameters if the environmental disturbance label is environmental disturbance, otherwise it will not perform the adjustment and will update the real-time meteorological and environmental parameter set. The power generation prediction module is used to obtain the power generation cycle to be predicted, and combine it with the analysis of environmental disturbance values ​​to obtain a multi-source data training set for the power generation cycle to be predicted. The training set is then input into the power generation prediction model along with the real-time meteorological and environmental parameter set, and the power generation prediction value for the power generation cycle to be predicted is output. The power generation correction module is used to obtain the air pressure value and air density at each moment within a preset time period, and analyze them to obtain the air pressure change label and the light refraction feature label, thereby obtaining the meteorological correction power generation prediction value. The model correction module is used to analyze and obtain power generation prediction labels based on meteorological correction of power generation prediction values. If the power generation prediction label indicates a deviation from the power generation prediction, power generation prediction model correction is performed; otherwise, power generation prediction model correction is not performed. The power generation prediction model correction includes offline power generation prediction model correction and online adaptive power generation prediction model correction.

Images