A photovoltaic power prediction method, device and equipment considering pollution loss rate

Abstract

The application discloses a kind of photovoltaic power prediction method, device and equipment considering pollution loss rate, it is related to photovoltaic power generation technical field, the method includes: using pollution loss rate prediction model predicts the pollution loss rate prediction value of future setting period;The pollution loss rate prediction model is obtained by training to first machine learning model;Using ideal power prediction model predicts the ideal photovoltaic power prediction value of future setting period;The ideal power prediction model is obtained by training to second machine learning model;According to the pollution loss rate prediction value and ideal photovoltaic power prediction value of future setting period, determine the photovoltaic power prediction result of future setting period, the application can realize the quantification of component surface pollution effect, so as to consider the influence of pollution loss on power, improve photovoltaic power prediction precision.

Description

Technical Field

[0001] This application relates to the field of photovoltaic power generation technology, and in particular to a photovoltaic power prediction method, apparatus and equipment that takes into account pollution loss rate. Background Technology

[0002] Against the backdrop of a global energy transition, new energy sources, represented by photovoltaic (PV) power generation, have experienced rapid development. However, PV power generation is inherently intermittent and volatile, with its output power significantly affected by weather conditions, posing a significant challenge to the safe and stable operation of the power grid and the economic dispatch of the electricity market. Therefore, accurate prediction of the output power of PV power plants is a key technology for ensuring grid acceptance capacity, improving power plant operation and maintenance efficiency, and participating in electricity market transactions.

[0003] Currently, mainstream photovoltaic power prediction methods are mainly divided into two categories: physical models and data-driven models. Physical models are based on the physical characteristics of photovoltaic modules and meteorological parameters (such as irradiance and temperature), but these models are complex and require high parameter accuracy. Data-driven models, such as Artificial Neural Networks (ANN), Support Vector Machines (SVM), and Long Short-Term Memory (LSTM), make predictions by learning the complex mapping relationship between historical power data and meteorological data, and have become a hot topic in current research and application.

[0004] However, existing technologies generally suffer from a key flaw in their modeling: they tend to simplify photovoltaic modules as idealized devices with constant performance, and their predictive logic primarily revolves around meteorological factors. But in real-world operating environments, the surface of photovoltaic modules accumulates pollutants such as dust, bird droppings, and industrial dust, leading to shading—a phenomenon known as the "soiling effect." This effect significantly reduces the amount of solar irradiance received by the module, causing its actual output power to be far lower than the theoretical value under ideal clean conditions. This power loss is typically quantified using the soiling loss ratio (SLR).

[0005] More importantly, pollution loss is a dynamic process. During prolonged dry, windy weather, pollutants accumulate, causing the SLR (Soil Reduction Rate) to gradually increase; conversely, effective rainfall can act as a natural cleansing agent, rapidly reducing the SLR. Traditional forecasting models fail to effectively capture and quantify this dynamic process. They either completely ignore the impact of pollution or use only a fixed, empirical reduction factor, failing to reflect real-time changes in pollution. This directly leads to a persistent and increasing overestimation bias in forecasts during prolonged periods of sunny, rainless weather, while underestimation may occur after rainfall, severely impacting forecast accuracy. Summary of the Invention

[0006] The purpose of this application is to provide a photovoltaic power prediction method, apparatus, and equipment that considers the pollution loss rate, which can quantify the pollution effect on the module surface, thereby taking into account the impact of pollution loss on power and improving the accuracy of photovoltaic power prediction.

[0007] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a photovoltaic power prediction method that considers pollution loss rate, including: A pollution loss rate prediction model is used to predict the pollution loss rate for a future specified period; the pollution loss rate prediction model is obtained by training a first machine learning model. An ideal power prediction model is used to predict the ideal photovoltaic power for a future specified period; the ideal power prediction model is obtained by training a second machine learning model. Based on the predicted pollution loss rate and the predicted ideal photovoltaic power for the future set time period, the photovoltaic power prediction result for the future set time period is determined.

[0008] Optionally, a pollution loss rate prediction model is used to predict the pollution loss rate for a future specified period, specifically including: Obtain the pollution loss rate sampling points within the first historical time period; Linear interpolation is performed on the sampling points of pollution loss rate within the first historical time period to obtain a time series of pollution loss rate that is continuous within the first historical time period; The pollution loss rate time series and environmental characteristic data for a future set period are input into the pollution loss rate prediction model to obtain the predicted pollution loss rate for the future set period.

[0009] Optionally, each pollution loss rate sampling point within the first historical time period is represented as: ; ; ; in, For pollution loss rate, This is the standard power under the pollution conditions sampled. This is the standard power under clean conditions during this sampling. and These represent the maximum photovoltaic power output under clean and unclean conditions, respectively. and The figures show the irradiance under clean and uncleaned conditions, respectively. and The component temperatures were measured under clean and uncleaned conditions, respectively, in this sampling. It is the power temperature coefficient of the component.

[0010] Optionally, the environmental characteristic data includes at least one of rainfall-related characteristics, atmospheric particulate matter concentration data, and meteorological data; the rainfall-related characteristics include rainfall intensity and / or the time since the last rainfall; the atmospheric particulate matter concentration data includes PM2.5 and / or PM10 concentration data; and the meteorological data includes humidity and wind direction.

[0011] Optionally, an ideal power prediction model is used to predict the ideal photovoltaic power for a future specified period, specifically including: A historical ideal photovoltaic power time series is generated based on the historical actual photovoltaic power time series and the historical pollution loss rate time series. The historical ideal photovoltaic power time series and meteorological data for a future set period are input into the ideal power prediction model to obtain the ideal photovoltaic power prediction value for the future set period.

[0012] Optionally, a historical ideal photovoltaic power time series is generated based on the historical actual photovoltaic power time series and the historical pollution loss rate time series, specifically including: calculating each ideal photovoltaic power in the historical ideal photovoltaic power time series according to the following; ; in, Indicates time Ideal photovoltaic power, Indicates time Actual photovoltaic power at that time Indicates time Pollution loss rate.

[0013] Optionally, the photovoltaic power prediction result for the future set period is determined based on the predicted pollution loss rate and the predicted ideal photovoltaic power for the future set period. Specifically, this includes determining the predicted photovoltaic power value for each future set period according to the following formula. ; in, Indicates the predicted time The corresponding photovoltaic power forecast value, Indicates the predicted time The corresponding ideal photovoltaic power prediction value, Indicates the predicted time The corresponding predicted pollution loss rate.

[0014] Optionally, the first machine learning model adopts a gradient boosting decision tree model, and the second machine learning model adopts a long short-term memory network.

[0015] Secondly, this application provides a photovoltaic power prediction device that takes into account pollution loss rate, comprising: A pollution loss rate prediction model is used to predict the pollution loss rate for a future specified period; the pollution loss rate prediction model is obtained by training a first machine learning model. An ideal power prediction model is used to predict the ideal photovoltaic power for a future specified period; the ideal power prediction model is obtained by training a second machine learning model. The photovoltaic power prediction result generation module is used to determine the photovoltaic power prediction result for a future set period based on the predicted pollution loss rate and the predicted ideal photovoltaic power for the future set period.

[0016] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the photovoltaic power prediction method considering pollution loss rate as described above.

[0017] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a photovoltaic power prediction method, apparatus, and equipment that considers pollution loss rate. It uses a pollution loss rate prediction model to predict the pollution loss rate for a future set period, realizing a quantitative estimate of the pollution loss rate. It considers the impact of pollution loss on power, solves the problem that existing power prediction models fail to dynamically quantify the surface pollution (dust accumulation / scouring) effect of the components. Based on the pollution loss rate prediction value and the ideal photovoltaic power prediction value for the future set period, the photovoltaic power prediction result for the future set period is determined. It realizes the coupling of the ideal photovoltaic power prediction value and the pollution loss rate prediction value, obtains a power prediction result that is close to the actual operating conditions, and improves the accuracy of photovoltaic power prediction. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic flowchart illustrating a photovoltaic power prediction method that takes into account pollution loss rate, as provided in an embodiment of this application.

[0020] Figure 2 This is a detailed implementation flowchart of a photovoltaic power prediction method that takes into account pollution loss rate, provided as an embodiment of this application.

[0021] Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0023] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] In one exemplary embodiment, this application provides a photovoltaic power prediction method that considers pollution loss rate, such as... Figure 1 As shown, the photovoltaic power prediction method considering pollution loss rate includes steps 101-103.

[0025] Step 101: Use a pollution loss rate prediction model to predict the pollution loss rate for a future set period; the pollution loss rate prediction model is obtained by training the first machine learning model.

[0026] Step 102: Use the ideal power prediction model to predict the ideal photovoltaic power for a future set period; the ideal power prediction model is obtained by training the second machine learning model.

[0027] Step 103: Determine the photovoltaic power prediction result for the future set time period based on the predicted pollution loss rate and the predicted ideal photovoltaic power for the future set time period.

[0028] This application realizes the quantitative estimation of pollution loss rate, takes into account the impact of pollution loss on power, and determines the photovoltaic power prediction result by coupling the ideal photovoltaic power prediction value with the pollution loss rate prediction value. It effectively integrates the dynamic and nonlinear time-varying factor of pollution loss of photovoltaic modules into photovoltaic power prediction, and establishes a prediction method that can simultaneously respond to meteorological changes and changes in module surface condition, thereby improving the accuracy of photovoltaic power prediction and the interpretability of prediction results.

[0029] In an exemplary embodiment, step 101 specifically includes: using a pollution loss rate prediction model to predict the pollution loss rate of the target photovoltaic module for a future set period.

[0030] Step 102 specifically includes: using an ideal power prediction model to predict the ideal photovoltaic power of the target photovoltaic module for a set period of time in the future.

[0031] In an exemplary embodiment, step 101 specifically includes steps 201-202.

[0032] Step 201: Obtain the pollution loss rate sampling points within the first historical time period.

[0033] Step 202: Perform linear interpolation on the sampling points of pollution loss rate within the first historical time period to obtain a time series of pollution loss rate that is continuous within the first historical time period.

[0034] Step 203: Input the pollution loss rate time series and environmental characteristic data within a future set time period into the pollution loss rate prediction model to obtain the predicted pollution loss rate value for the future set time period.

[0035] Specifically, the pollution loss rate time series at time step t and the environmental characteristic data at time step t+1 are input into the pollution loss rate prediction model to obtain the predicted pollution loss rate at time step t+1.

[0036] In an exemplary embodiment, training the first machine learning model specifically includes: acquiring historical pollution loss rate time series data. Based on the historical pollution loss rate time series data and a first set of preset feature data, a pollution loss rate prediction model is trained to predict future SLR values. The first set of preset feature data consists of sample data of environmental feature data.

[0037] The acquisition of historical pollution loss rate time series data specifically includes: within the same measurement period, first measuring the performance of the photovoltaic module under pollution conditions to obtain the power under pollution conditions; then cleaning the photovoltaic module and measuring its performance under clean conditions within the shortest possible time (preset time) to obtain the power under clean conditions; standardizing the power under pollution conditions and the power under clean conditions; and calculating the SLR value at the measurement time based on the standardized pollution power and clean power. Interpolation processing is then performed on the discrete SLR values ​​obtained through measurement to generate temporally continuous historical SLR time series data.

[0038] The training of the second machine learning model specifically includes: generating an ideal power time series as the training target using historical actual power data and the historical pollution loss rate time series data. Based on the ideal power time series and the second set of preset feature data, an ideal power prediction model is trained to predict future ideal power values. The second set of preset feature data includes sample meteorological data.

[0039] In one exemplary embodiment, the core of quantifying pollution loss is to “sample” the evolution of pollution through periodic, precise physical measurements to obtain a series of valid but time-discontinuous SLR data points, and then maintain their continuity in the time dimension through linear interpolation.

[0040] In this application, the effective pollution loss rate sampling point refers to a pollution loss rate data point that meets preset sampling validity conditions. The preset sampling validity conditions include at least one of the following: pollution state measurement and clean state measurement are performed on the same photovoltaic module; the time interval between pollution state measurement and clean state measurement is not greater than a preset time threshold; the solar irradiance is within a preset effective range during sampling, and the rate of change of irradiance between the two measurements is not greater than a first change threshold; the change in module temperature between the two measurements is not greater than a second change threshold. The preset time threshold, the first variation threshold, and the second variation threshold can all be preset based on the sampling equipment accuracy, sampling cycle, site environmental stability, and prediction time resolution. The preset time threshold limits the maximum permissible time interval between pollution state measurements and clean state measurements to reduce the impact of environmental condition changes on the pollution loss rate calculation results. The first variation threshold limits the maximum permissible variation in solar irradiance between the two measurements. The second variation threshold limits the maximum permissible variation in component temperature between the two measurements. As a specific example, the preset time threshold is 1 minute to 5 minutes, the first variation threshold is 2% to 5%, and the second variation threshold is 1°C to 3°C.

[0041] For pollution loss rate data points that do not meet the above-mentioned preset sampling validity conditions, they are marked as abnormal sampling points, and the abnormal sampling points are removed, resampled, or corrected by using adjacent valid pollution loss rate sampling points.

[0042] Each pollution loss rate sampling point within the first historical time period is represented as follows: ; in, For pollution loss rate, This is the standard power under the pollution conditions sampled. This is the standard power under clean conditions for this sampling. Using standard power is necessary to eliminate the influence of minute changes in light and temperature; the measured power needs to be standardized. Standard test conditions (1000 W / m²) are typically used. 2 The power at 25℃ was compared. The standardized formula is as follows: ; ; in, and These represent the maximum photovoltaic power output under clean and unclean conditions, respectively. and The figures show the irradiance under clean and uncleaned conditions, respectively. and The component temperatures were measured under clean and uncleaned conditions, respectively, in this sampling. It is the power temperature coefficient of the component.

[0043] Because the obtained pollution loss rate sampling points are relatively sparse in the time dimension, while photovoltaic power prediction typically requires continuous pollution loss rate data on a daily, hourly, or preset time interval basis, it is necessary to perform linear interpolation on the discrete pollution loss rate sampling points to generate a continuous pollution loss rate time series. Specifically, let two adjacent pollution loss rate sampling points be... and For those located and Interpolation time between Its pollution loss rate The calculation method is as follows: ; in, and These represent the sampling times of two adjacent pollution loss rate sampling points, respectively. and These represent the pollution loss rates corresponding to the two adjacent sampling times, respectively. Indicates the time to be interpolated. This represents the pollution loss rate to be obtained through interpolation.

[0044] After obtaining the time series of pollution loss rate within the first historical time period using the above interpolation method, the series is used as input features and fed into the pollution loss rate prediction model to output the predicted pollution loss rate for the future set time period.

[0045] The environmental characteristic data includes at least one of rainfall-related characteristics, atmospheric particulate matter concentration data, and meteorological data; the rainfall-related characteristics include rainfall intensity and / or the time since the last rainfall; the atmospheric particulate matter concentration data includes PM2.5 and / or PM10 concentration data; and the meteorological data includes humidity and wind direction.

[0046] The environmental characteristic data is a time series corresponding to the pollution loss rate time series that is continuous within the first historical time period.

[0047] The first machine learning model employs a gradient boosting decision tree model (XGBoost). XGBoost is a first machine learning model that trains multiple decision trees sequentially and sums their predictions to obtain the final result. Its core idea is to use subsequent trees to specifically learn and correct the prediction errors of the previous tree. Its optimization objective function... for: ; in, To predict the loss, calculate the model's predicted value ( ) and the true value ( The difference is denoted as n, where n is the number of samples and i is the i-th sample. The complexity penalty term for the k-th tree, This represents the k-th decision tree; This is a regularization term used to penalize complex models and prevent overfitting.

[0048] In one exemplary embodiment, this application provides a prediction method that combines ideal power prediction with actual power prediction. The core idea of ​​this prediction method is to first complete the ideal power prediction.

[0049] Step 102 specifically includes steps 301-302.

[0050] Step 301: Generate a historical ideal photovoltaic power time series based on the historical actual photovoltaic power time series and the historical pollution loss rate time series. Specifically, this includes: generating a historical ideal photovoltaic power time series based on the historical actual photovoltaic power time series and the historical pollution loss rate time series within the second historical time period.

[0051] The first historical period and the second historical period can be the same time period.

[0052] Step 302: Input the historical ideal photovoltaic power time series and meteorological data within the future set period into the ideal power prediction model to obtain the ideal photovoltaic power prediction value for the future set period.

[0053] Specifically, the historical ideal photovoltaic power time series at time step t and the meteorological data at time step t+1 are input into the ideal power prediction model to obtain the ideal photovoltaic power prediction value at time step t+1.

[0054] The ideal photovoltaic power prediction described in this application is achieved through... This was done through correction. To eliminate the contamination effect in historical photovoltaic power data, after acquiring... After sequencing, it is compared with the historical actual photovoltaic power time series. Alignment in the time dimension.

[0055] Specifically, step 301 includes: calculating each ideal photovoltaic power in the historical ideal photovoltaic power time series according to the following: ; in, Indicates time Ideal photovoltaic power, Indicates time Actual photovoltaic power at that time Indicates time Pollution loss rate.

[0056] A new time series is obtained based on the above formula. This refers to the historical ideal photovoltaic power time series, which represents the power curve under "pollution-free" conditions. This is compared to the historical actual photovoltaic power time series. The peak value of this curve will be higher and more stable on sunny days, and it will not show a slow decline due to long-term dust accumulation.

[0057] Step 301 also includes calculating Perform boundary processing, which includes at least one of the following: First item: Will The value should be adjusted to not exceed the rated installed capacity of the photovoltaic power station.

[0058] Second item: When When the value is close to 1, specifically: When the value is less than the set value, it is the denominator. Set a non-zero lower bound value.

[0059] In one exemplary embodiment, the second machine learning model employs a Long Short-Term Memory (LSTM) network. Each memory unit in the hidden layer of the LSTM is controlled by an input gate, a forget gate, and an output gate. Indicates time The input of the LSTM unit, in this application, has the following characteristics. This mainly includes the historical ideal photovoltaic power sequence calculated by the aforementioned steps. and with time The corresponding meteorological forecast characteristics (such as irradiance, component temperature, etc.). , and These represent the outputs of the input gate, forget gate, and output gate, respectively. Additionally, the output of the candidate memory unit is denoted as... The output of the memory unit is denoted as In time At that time, each state The calculation formula is as follows: ; in, , , and These are the weight matrices for the input gate, forget gate, output gate, and memory unit, respectively. , , and These are all weight matrices of the hidden layers; , and These are the bias vectors for the input gate, forget gate, and output gate, respectively. For activation functions; For element-wise multiplication; It is the hyperbolic tangent function.

[0060] The final photovoltaic power prediction and output is composed of a pollution loss rate prediction model and an ideal power prediction model. Its core logic lies in adjusting the ideal power prediction through dynamic pollution.

[0061] At each prediction time step, the final output power consists of two parts, first predicted by the pollution loss rate prediction model. The time panel Then, the ideal power prediction model is used to predict... Ideal output power at all times .

[0062] Combining these two parts according to the following formula yields the result. Final power at time .

[0063] Step 103 specifically includes: determining the predicted photovoltaic power value for each future time period according to the following formula; ; in, Indicates the predicted time The corresponding photovoltaic power forecast value, Indicates the predicted time The corresponding ideal photovoltaic power prediction value, Indicates the predicted time The corresponding predicted pollution loss rate.

[0064] The photovoltaic power prediction method based on pollution loss rate correction proposed in this application can be summarized as a "decoupling-separate modeling-recoupling" framework. This framework mainly consists of three steps. The core of the first step is to quantify the loss rate through experiments and build a loss rate prediction model, namely the pollution loss rate prediction model. The core of the second step is to establish a pure "weather-power" model, namely the ideal power prediction model. The core of the third step is to combine the prediction results of the two specialized models in a way that conforms to physical laws, generating a final, more accurate, and more robust photovoltaic power prediction.

[0065] In one exemplary embodiment, such as Figure 2 As shown, a photovoltaic power prediction method considering pollution loss rate includes the following steps S1-S3.

[0066] Step S1: Obtain pollution loss rate data.

[0067] Step S10: Preparation. To obtain SLR data points, at least one representative photovoltaic module can be selected as a pollution monitoring module in the photovoltaic power station, or multiple photovoltaic modules that can reflect the average dust accumulation state of the station can be selected as pollution monitoring modules. The pollution monitoring modules are installed at preset typical locations that can reflect the average dust accumulation state of the station. By performing paired measurements on the output performance of the pollution monitoring modules before and after cleaning, basic data for calculating the pollution loss rate is obtained.

[0068] Simultaneously prepare three tools: a current (I)-voltage (V) curve tester, an irradiance meter, and a temperature sensor. The current-voltage curve tester is used to obtain the current-voltage characteristic curve of a photovoltaic module under current irradiance and module temperature conditions. It can quickly measure the current-voltage (IV) characteristic curve of a photovoltaic module under current light and temperature, and provide key parameters such as maximum power point (Pmax), open-circuit voltage (Voc), and short-circuit current (Isc). The irradiance detection device and temperature detection device can be integrated with the current-voltage curve tester or set up independently, as long as they can simultaneously record the total planar irradiance (GHI) and module backsheet temperature at the moment of measurement.

[0069] Step S11: Data Acquisition. Contamination and cleanliness measurements are performed within a timeframe that meets the sampling validity requirements to ensure the reliability of the measurement data. Throughout the measurement cycle, consistency in measurement time periods should be maintained as much as possible.

[0070] First, without treating the sample components in any way, the performance under contaminated conditions was measured using an IV tester. This refers to photovoltaic power. Then record the maximum power at this point. and synchronous irradiance and component temperature .

[0071] Then, immediately perform a thorough cleaning of the same sample component to ensure the surface is free of any dust and dirt. Measure the cleanliness status at time intervals not exceeding a preset time threshold, and then measure again using an IV tester. Record the maximum power at this point. Considering the extremely short time, the irradiance can be assumed to be... and temperature Same as the previous step.

[0072] To eliminate the effects of minute changes in light and temperature, the measurement power needs to be standardized. Standard test conditions (1000 W / m²) are typically used. 2 The power output at 25°C was compared. The maximum photovoltaic power output under uncleaned conditions was obtained using a standardized formula. Maximum photovoltaic power under clean conditions .

[0073] Pollution loss rate at that moment: .

[0074] Step S12: Interpolation. Since the obtained SLR is relatively sparse in the time dimension, while power prediction requires data from each day or even each hour, it is necessary to perform simple linear interpolation on these discrete points to generate a continuous SLR time series.

[0075] Step S2: Establish a pollution loss rate prediction model. This process is broken down into two main stages: feature engineering and model training and evaluation.

[0076] Step S20: Feature Construction. Based on the pollution accumulation process and the rainfall washout process, construct environmental feature data to characterize the changing patterns of pollution loss rates. Features that clearly describe the pollution accumulation and washout processes need to be constructed. The goal of this step is to create variables that explicitly indicate "accumulation" or "washout".

[0077] The first step is to clean the relevant features, which mainly involves imputing missing data. In this example, linear interpolation can be used; please refer to step S12 for details. Feature engineering mainly involves the following tasks: First, a binary flag (0 or 1) is used to represent a rainfall event; 1 indicates effective rainfall on that day, and 0 indicates otherwise. Second, the maximum rainfall intensity (Rainfall_24h) ​​and the last rainfall time (Days_Since_Last_Rain) over the past 24 hours are calculated. Simultaneously, daily PM2.5 or PM10 concentration, humidity, and wind direction are used as cumulative correlation features. SLR values ​​from the past few days, such as SLR(t-1), SLR(t-2), ... SLR(t-7), also need to be calculated. This incorporates the "inertia" of the system state; the current pollution loss rate has a temporal correlation with the pollution loss rate at one or more historical moments. Therefore, the lag term of the historical pollution loss rate can be used as a model input feature.

[0078] Next, model building is required. Here, XGBoost is used as the prediction base model, mainly because tree models are very good at handling tabular data and can automatically learn the nonlinear relationships and interactions between features. For example, this model can learn the nonlinear mapping relationship between environmental features and changes in pollution loss rate: "When Days_Since_Last_Rain>10 and PM10>100, the SLR increment will increase"; or "SLR will only decrease significantly when Rainfall_24h>5mm".

[0079] To predict SLR(t+1), the actual value of SLR(t) and the weather and environmental forecast data at time t+1 are required as input. This constitutes an autoregressive prediction process: after predicting SLR(t+1), it is used as input to predict SLR(t+2), and so on.

[0080] Step S3: Construction of the ideal power prediction model and power synthesis. This step mainly consists of building the ideal power prediction model and combining it with the final power prediction.

[0081] Step S30: Building the Ideal Power Prediction Model. The ideal power prediction model is used to predict the ideal photovoltaic power of photovoltaic modules under clean conditions; its output is determined solely by weather changes. To build this model, training targets must first be generated. This is the starting point of the entire process, and its purpose is to eliminate contamination effects in historical power data. Each data point in the sequence Apply the following formula: ; In the formula This represents the actual photovoltaic power in the historical actual photovoltaic power time series. To and A time-aligned, estimated historical pollution loss rate time series. This series can be obtained through periodic measurements and interpolation, or through data-driven methods.

[0082] In this process, there are some important boundary handling aspects: 1. When it is evening Then, It is also 0.

[0083] 2. When When the value is very close to or equal to 0, a lower limit needs to be set to prevent division by zero: .

[0084] 3. The value should not exceed the rated installed capacity of the power plant. Therefore, trimming is necessary: .

[0085] In this application, the ideal power prediction model is an LSTM, used to learn the temporal mapping relationship between historical ideal photovoltaic power, historical meteorological characteristics, historical time characteristics, and meteorological prediction characteristics at the prediction time, and future ideal photovoltaic power. Specifically, for each time step j within the historical window, an input feature vector is constructed. : ; For prediction time step t, the input feature sequence within the historical window, along with the meteorological prediction features and temporal features corresponding to prediction time step t, are input into the Long Short-Term Memory network model to obtain the ideal photovoltaic power prediction value: ; in, Indicates the time step within the historical window The corresponding input feature vector; Indicates time step Corresponding historical ideal photovoltaic power; Indicates time step The corresponding historical meteorological feature vector; Indicates time step The corresponding time feature vector; Indicates the prediction time step The corresponding ideal photovoltaic power prediction value; The parameter is Long Short-Term Memory Network Model; Indicates the prediction time step The corresponding meteorological forecast feature vector; Indicates the prediction time step The corresponding time feature vector; Set the preset history window length.

[0086] Step S31: Final power prediction output. This is the stage where the two decoupled models are recombined to obtain the final result.

[0087] Suppose we need to predict power levels for the next 24 hours. First, we need to load the trained pollution dynamics model: the predicted pollution loss rate. And a well-trained ideal power prediction model.

[0088] The core logic of the final power prediction output performs two tasks in parallel at each prediction time step. The first is to predict, using a pollution dynamic model, how "dirty" the panel will be at time t. The second assumption is that the prediction would be made if the panel were clean. How much power generation potential does it have at any given moment? Then, multiply these two pieces of information together: ; Ultimately, we arrive at the prediction result that is closest to reality.

[0089] This application aims to address the problem of inaccurate power prediction caused by surface contamination of photovoltaic modules. The core idea of ​​this method is to decouple the prediction task into two parallel sub-models: dynamic prediction of contamination loss rate and prediction of ideal clean power. First, discrete SLR data points are obtained by measuring the performance of the sentinel modules before and after cleaning, and these are interpolated to generate a historical SLR time series. Simultaneously, this historical SLR series is used to correct the historical actual power, generating an ideal power time series free from contamination effects. Second, two prediction models are constructed: an SLR prediction model that uses environmental characteristics such as rainfall and PM concentration to learn the accumulation and cleaning patterns of contamination; and an ideal power prediction model that learns the mapping relationship between meteorological conditions and power generation under clean conditions. During the prediction phase, this method calls both models in parallel to obtain the predicted SLR value and the predicted ideal power value for future times. Finally, the predicted ideal power is multiplied by (1 - predicted SLR) to synthesize the final power prediction result, which is closer to actual operating conditions. This application significantly improves the accuracy and reliability of photovoltaic power prediction by separating the dynamics of contamination and meteorological influences.

[0090] Based on the same inventive concept, this application also provides a photovoltaic power prediction device that considers pollution loss rate as described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more photovoltaic power prediction device embodiments considering pollution loss rate provided below can be found in the limitations of the photovoltaic power prediction method considering pollution loss rate described above, and will not be repeated here.

[0091] In one exemplary embodiment, a photovoltaic power prediction device that takes into account pollution loss rate is provided, comprising: A pollution loss rate prediction model is used to predict the pollution loss rate for a future specified period; the pollution loss rate prediction model is obtained by training a first machine learning model.

[0092] An ideal power prediction model is used to predict the ideal photovoltaic power for a future specified period; the ideal power prediction model is obtained by training a second machine learning model.

[0093] The photovoltaic power prediction result generation module is used to determine the photovoltaic power prediction result for a future set period based on the predicted pollution loss rate and the predicted ideal photovoltaic power for the future set period.

[0094] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 3As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores data related to a photovoltaic power prediction method considering pollution loss rates. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a photovoltaic power prediction method considering pollution loss rates.

[0095] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0096] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0097] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0098] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0099] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0100] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0101] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A photovoltaic power prediction method considering pollution loss rate, characterized in that, include: A pollution loss rate prediction model is used to predict the pollution loss rate for a future specified period; the pollution loss rate prediction model is obtained by training a first machine learning model. An ideal power prediction model is used to predict the ideal photovoltaic power for a future specified period; the ideal power prediction model is obtained by training a second machine learning model. Based on the predicted pollution loss rate and the predicted ideal photovoltaic power for the future set time period, the photovoltaic power prediction result for the future set time period is determined.

2. The photovoltaic power prediction method considering pollution loss rate according to claim 1, characterized in that, The pollution loss rate prediction model is used to predict the pollution loss rate for a given future period, specifically including: Obtain the pollution loss rate sampling points within the first historical time period; Linear interpolation is performed on the sampling points of pollution loss rate within the first historical time period to obtain a time series of pollution loss rate that is continuous within the first historical time period; The pollution loss rate time series and environmental characteristic data for a future set period are input into the pollution loss rate prediction model to obtain the predicted pollution loss rate for the future set period.

3. The photovoltaic power prediction method considering pollution loss rate according to claim 2, characterized in that, Each pollution loss rate sampling point within the first historical time period is represented as follows: ； ； ； in, For pollution loss rate, This is the standard power under the pollution conditions sampled. This is the standard power under clean conditions during this sampling. and These represent the maximum photovoltaic power output under clean and unclean conditions, respectively. and The figures show the irradiance under clean and uncleaned conditions, respectively. and The component temperatures were measured under clean and uncleaned conditions, respectively, in this sampling. It is the power temperature coefficient of the component.

4. The photovoltaic power prediction method considering pollution loss rate according to claim 2, characterized in that, The environmental characteristic data includes at least one of rainfall-related characteristics, atmospheric particulate matter concentration data, and meteorological data; the rainfall-related characteristics include rainfall intensity and / or the time since the last rainfall; the atmospheric particulate matter concentration data includes PM2.5 and / or PM10 concentration data; and the meteorological data includes humidity and wind direction.

5. The photovoltaic power prediction method considering pollution loss rate according to claim 1, characterized in that, The ideal power prediction model is used to predict the ideal photovoltaic power for a given future period, specifically including: A historical ideal photovoltaic power time series is generated based on the historical actual photovoltaic power time series and the historical pollution loss rate time series. The historical ideal photovoltaic power time series and meteorological data for a future set period are input into the ideal power prediction model to obtain the ideal photovoltaic power prediction value for the future set period.

6. The photovoltaic power prediction method considering pollution loss rate according to claim 5, characterized in that, Generate a historical ideal photovoltaic power time series based on the historical actual photovoltaic power time series and the historical pollution loss rate time series. Specifically, this includes calculating each ideal photovoltaic power in the historical ideal photovoltaic power time series according to the following: ； in, Indicates time Ideal photovoltaic power, Indicates time Actual photovoltaic power at that time Indicates time Pollution loss rate.

7. The photovoltaic power prediction method considering pollution loss rate according to claim 1, characterized in that, Based on the predicted pollution loss rate and the predicted ideal photovoltaic power for a future set period, the photovoltaic power prediction results for the future set period are determined, specifically including: determining the predicted photovoltaic power value for each future set period according to the following formula; ； in, Indicates the predicted time The corresponding photovoltaic power forecast value, Indicates the predicted time The corresponding ideal photovoltaic power prediction value, Indicates the predicted time The corresponding predicted pollution loss rate.

8. The photovoltaic power prediction method considering pollution loss rate according to claim 1, characterized in that, The first machine learning model uses a gradient boosting decision tree model, and the second machine learning model uses a long short-term memory network.

9. A photovoltaic power prediction device considering pollution loss rate, characterized in that, The photovoltaic power prediction device considering pollution loss rate applies the photovoltaic power prediction method considering pollution loss rate as described in any one of claims 1-8, and the photovoltaic power prediction device considering pollution loss rate includes: A pollution loss rate prediction model is used to predict the pollution loss rate for a future specified period; the pollution loss rate prediction model is obtained by training a first machine learning model. An ideal power prediction model is used to predict the ideal photovoltaic power for a future specified period; the ideal power prediction model is obtained by training a second machine learning model. The photovoltaic power prediction result generation module is used to determine the photovoltaic power prediction result for a future set period based on the predicted pollution loss rate and the predicted ideal photovoltaic power for the future set period.

10. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the photovoltaic power prediction method considering pollution loss rate as described in any one of claims 1-8.

Images