Mountainous area distributed photovoltaic power generation power prediction method and system

Abstract

The application relates to the technical field of power generation power prediction, in particular to a mountainous area distributed photovoltaic power generation power prediction method and system. The method comprises the following steps: collecting photovoltaic power generation power of a target mountainous area photovoltaic power station at each time and various types of environmental data; wherein the environmental data comprises weather states, the environmental contribution change degree of each time period is determined, and the noise interference degree of each time period is obtained; according to the correlation between the photovoltaic power generation power and various types of environmental data, the possibility that various types of data collected at each time is noise data is evaluated, the correlation improvement coefficient of each time data point is calculated, the noise interference degree of the time period to which each time belongs is combined, and the sliding window size when the collected data at each time is subjected to smoothing processing is determined; and the photovoltaic power generation power is predicted by using all types of data subjected to smoothing processing. Therefore, the accuracy of mountainous area photovoltaic power generation power prediction is improved.

Description

Technical Field

[0001] This application relates to the field of power generation prediction technology, specifically to a method and system for predicting distributed photovoltaic power generation in mountainous areas. Background Technology

[0002] With the increasing demand for clean energy, renewable energy technologies are being applied more and more widely. Among them, distributed photovoltaic (PV) power generation, as one of the most important uses of renewable energy technologies, is widely used in pastoral areas, mountainous regions, and rural areas. However, PV power generation has significant randomness and volatility. Large-scale distributed PV grid connection can affect the operational stability of the power system. Therefore, it is necessary to predict the power generation of distributed PV and rationally arrange its operation mode and countermeasures to improve the reliability of the power system.

[0003] Existing technologies typically use neural network models to predict the power generation of distributed photovoltaic systems. However, when training the neural network model using various data collected from the distributed photovoltaic system, the data needs to be preprocessed to reduce the impact of missing and noisy data on the accuracy of the final model prediction.

[0004] Currently, the moving average method is commonly used to fill in missing values ​​and smooth the acquired data sequentially to achieve data cleaning. However, when the moving average method smooths the collected data to suppress noise, it often uses a fixed window to smooth all the data, ignoring the abnormal changes in photovoltaic power generation caused by changes in the environment of distributed photovoltaics. This is especially true in mountainous areas with variable weather. As a result, when using this fixed-window moving average method to smooth the collected data, data that reflects the true power generation characteristics of distributed photovoltaics in mountainous areas is misjudged as noise data and over-smoothed, thus affecting the accuracy of the final power generation prediction. Summary of the Invention

[0005] To address the aforementioned technical problems, the purpose of this application is to provide a method and system for predicting distributed photovoltaic power generation in mountainous areas. The specific technical solution adopted is as follows: In a first aspect, embodiments of this application provide a method for predicting the power generation of distributed photovoltaic power in mountainous areas, the method comprising the following steps: The system collects photovoltaic power generation data of the target mountain photovoltaic power station at various times, as well as various environmental data, including weather conditions. Determine the impact of various environmental data (excluding weather conditions) on the photovoltaic power generation of a photovoltaic power station within the same time period, analyze the differences in the impact between each time period and its adjacent time periods, and determine the degree of change in environmental contribution for each time period. By determining the similarity of weather conditions between each time period and its adjacent time periods, the weather stability coefficient of each time period is determined. Combined with the environmental contribution variation, the noise interference degree of each time period is obtained. Based on the correlation between photovoltaic power generation and various environmental data, the possibility that the data collected at each time point is noise data is assessed, the correlation enhancement coefficient of the data points at each time point is calculated, and the sliding window size is determined when smoothing the data collected at each time point by combining the noise interference of the time period to which each time point belongs. The photovoltaic power generation output is predicted using all types of data after smoothing.

[0006] In one embodiment, determining the impact of various environmental data (excluding weather conditions) on the photovoltaic power generation of the photovoltaic power station within the same time period includes: Using photovoltaic power generation at any given time period as the dependent variable and various environmental data (excluding weather conditions) as independent variables, a multivariate nonlinear regression model is established to obtain a regression coefficient vector composed of all regression coefficients. This vector is used to represent the degree of influence of various environmental data (excluding weather conditions) on the photovoltaic power generation of the photovoltaic power station within the given time period.

[0007] In one embodiment, the degree of change in environmental contribution is determined based on the positive fusion result of the differences between the regression coefficient vectors of each time period and all its adjacent time periods.

[0008] In one embodiment, determining the weather stability coefficient includes: The weather conditions at all times within each time period are converted into word vector sequences. The similarity between the word vector sequences of each time period and its adjacent time periods is calculated. The weather stability coefficient is positively correlated with the similarity.

[0009] In one embodiment, the noise interference degree is the product of the environmental contribution variation degree and the weather stability coefficient for each time period, wherein the environmental contribution variation degree and the weather stability coefficient for each time period are both in the range of [-1, 1].

[0010] In one embodiment, calculating the correlation enhancement coefficient of data points at each time point includes: The correlation between photovoltaic power generation and various environmental data within the time period of each moment was analyzed before and after removing all types of data collected at each time point, and the correlation enhancement coefficient of the data points at each time point was obtained.

[0011] In one embodiment, determining the sliding window size when smoothing the acquired data at each time point includes: The mean of the normalized result of the correlation enhancement coefficient of the data points at each time point and the normalized result of the noise interference degree of the time period to which each time point belongs is calculated and used as the window adjustment coefficient at each time point. Combined with the maximum and minimum values ​​of the sliding window size, the final sliding window size at each time point is calculated.

[0012] In one embodiment, the difference between the maximum value and the minimum value is calculated, and the product of the difference and the window adjustment coefficient is rounded down. The final sliding window size at each time step is the sum of the minimum value and the rounded down result.

[0013] In one embodiment, the prediction of photovoltaic power generation includes: The smoothed data of various types is used as training data to train the neural network model, and the trained neural network model is used to obtain the prediction results of photovoltaic power generation.

[0014] Secondly, embodiments of this application also provide a distributed photovoltaic power generation prediction system for mountainous areas, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of any of the methods described above.

[0015] This application has at least the following beneficial effects: This application collects power generation and environmental data from photovoltaic power plants in target mountainous areas to quantify the impact of different environmental factors on power generation. This allows subsequent power generation predictions to fully consider real-time environmental changes, thereby reducing prediction errors under variable mountain climate conditions. In-depth analysis of the influence between various environmental data and photovoltaic power generation enhances the understanding of how environmental factors affect photovoltaic power generation and improves the reliability of noise identification in the collected data. Furthermore, by analyzing noise interference and correlation enhancement coefficients at different time periods, the true power generation characteristics can be identified, ensuring that important signals are not mistakenly identified as noise. This allows for dynamic adjustment of the sliding window size to adapt to abnormal fluctuations in photovoltaic power generation caused by environmental changes at different time periods, effectively avoiding the over-smoothing problem caused by fixed windows. This flexible data smoothing significantly improves the accuracy of photovoltaic power generation prediction, making the predictions more consistent with reality, reducing misjudgments and improving data quality. Ultimately, this helps improve the power generation efficiency of photovoltaic power plants, reduce operational risks, enhance adaptability to external environmental changes, and ensure the continuous and efficient operation of the power plants under variable climate conditions. Attached Figure Description

[0016] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art 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.

[0017] Figure 1 A flowchart illustrating the steps of a method for predicting the power output of distributed photovoltaic power generation in mountainous areas, as provided in one embodiment of this application. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by this application to achieve the intended purpose of the invention, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a method and system for predicting distributed photovoltaic power generation in mountainous areas proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0020] The following description, in conjunction with the accompanying drawings, details the specific scheme of the method and system for predicting distributed photovoltaic power generation in mountainous areas provided in this application.

[0021] Please see Figure 1 The diagram illustrates a flowchart of a method for predicting the power generation of distributed photovoltaic power in mountainous areas, according to an embodiment of this application. The method includes the following steps: S1 collects the photovoltaic power generation of the target mountain photovoltaic power station at various times, as well as various environmental data, including weather conditions.

[0022] Historical photovoltaic (PV) datasets for each PV power station in the target mountain area are obtained. These datasets include PV power generation data collected within the same preset historical time period at the same preset data sampling interval, as well as various environmental data. Environmental data includes temperature, humidity, wind speed, and solar radiation intensity data, along with weather status data for each PV power station. Weather status data consists of word vectors converted from common mountain weather types such as sunny, cloudy, overcast, rainy, snowy, foggy, hazy, and strong wind. This data can be obtained in real-time from weather forecast websites for the area where each PV power station is located. The conversion of word vectors utilizes existing known techniques. The preset time period is at least one year and only includes sunshine hours (typically from sunrise to sunset) to ensure that the final distributed PV power generation prediction model can better capture the impact of seasonal factors on the accuracy of the distributed PV power generation prediction results in the target mountain area and reduce the interference of useless PV data collected at night. In this embodiment, the preset historical time period is set to the period from 9:00 AM to 3:00 PM every day within one year, and the preset data sampling interval is set to 10 minutes. Implementers can adjust this setting according to their specific circumstances. Photovoltaic power generation data, temperature data, humidity data, wind speed data, and solar radiation intensity data can be collected using power meters, temperature sensors, humidity sensors, wind speed sensors, and solar radiation meters, respectively. Photovoltaic power generation data and its various environmental data are all stored in time-series format.

[0023] For each type of historical photovoltaic dataset acquired from each photovoltaic power plant, pre-cleaning processing is performed to obtain pre-cleaned historical photovoltaic datasets for each power plant. The pre-cleaning process includes using linear interpolation to impute missing values ​​for each type of data, and using box plots to detect and remove outliers for each type of data after the missing value imputation process. After outlier removal, linear interpolation is used to impute missing data, thereby reducing the impact of missing data and obvious anomalies caused by noise in the acquired historical photovoltaic datasets on subsequent photovoltaic power generation prediction. Linear interpolation and box plots are well-known techniques, and their specific processes are not detailed here.

[0024] The Min-Max normalization method was used to normalize various types of data in the historical photovoltaic dataset of the photovoltaic power plant after pre-cleaning, in order to eliminate the influence of data units on subsequent data processing. The Min-Max normalization method is a well-known technique, and its specific process will not be elaborated further.

[0025] S2, determine the degree of influence of various environmental data (excluding weather conditions) on the photovoltaic power generation of the photovoltaic power station within the same time period, analyze the differences in the degree of influence between each time period and its adjacent time periods, and determine the degree of change of environmental contribution in each time period.

[0026] Generally, the solar radiation intensity, temperature, humidity, and wind speed of the environment where a photovoltaic (PV) power station is located are the main factors affecting its power generation. This is because PV power generation is a technology that converts solar energy into electrical energy, and temperature affects the operating conditions of PV panels. The power generation efficiency of PV panels generally decreases as the temperature rises; for every 1°C increase in temperature, the power generation efficiency decreases by 0.3%. The solar radiation intensity and temperature are the direct factors affecting the power generation of photovoltaic power plants, while humidity reflects the relative moisture content in the air and indirectly affects the power generation of photovoltaic power plants by affecting the light transmittance of the air. Changes in wind speed will change the speed of the flow of tiny particles and gases in the air, and indirectly affect the power generation of photovoltaic power plants by affecting the light transmittance and heat dissipation of photovoltaic panels.

[0027] In the short term, such as within a few consecutive hours, the aging degree and dust accumulation of photovoltaic panels in the same photovoltaic power station usually do not change significantly. Therefore, when the weather in the environment where the same photovoltaic power station is located does not change significantly, the contribution of solar radiation intensity, temperature, humidity and wind speed in the environment of the same photovoltaic power station to the power generation of the photovoltaic power station usually does not change significantly in the short term. However, when the weather in the environment where the same photovoltaic power station is located changes significantly in the short term, it usually causes abrupt changes in the data of solar radiation intensity, temperature, humidity and wind speed. However, the noise data in the historical photovoltaic dataset of the same photovoltaic power station usually does not have this change characteristic due to its randomness.

[0028] Based on the above analysis, taking the time series B corresponding to the photovoltaic power generation data in the historical photovoltaic dataset A (after pre-cleaning and normalization processing) of any photovoltaic power station P in the target mountainous area as an example, the sliding window algorithm is used to divide the time series B into multiple time series segments to represent the time series formed by the photovoltaic power generation data collected in different historical time periods. In this embodiment, the window size and sliding step size in the sliding window algorithm are both set to 30, which can be set by the implementer. The average timestamp of all data points in the time series segment is used as the timestamp of that time series segment.

[0029] Taking any time series segment b in time series B as an example, based on the timestamps of each time series segment, the adjacent time series segments of time series segment b in time series B are obtained. Specifically, the absolute value of the difference between the timestamps of each time series segment in time series B (excluding time series segment b) and time series segment b is calculated, and these absolute values ​​are sorted in ascending order. The time series segments corresponding to the first preset number of absolute differences are taken as the adjacent time series segments of time series segment b. In this embodiment, the preset number is set to 4. Implementers can set it according to the actual situation, and this embodiment does not impose any restrictions on it.

[0030] Using the same method as time series B, obtain all time series segments corresponding to each type of environmental data in the historical photovoltaic dataset A.

[0031] From all time series segments corresponding to each type of environmental data in the historical photovoltaic dataset A (excluding weather type data), a time series segment with the same timestamp as time series segment b is obtained and denoted as the environmental data time series segment of time series segment b. This segment is used to characterize the time series formed by each type of environmental data of photovoltaic power station P collected within the historical time period of time series segment b. Based on time series segment b and all data in all environmental data time series segments, a multivariate nonlinear regression model is established. In this embodiment, the multivariate nonlinear regression model uses a second-order polynomial function for regression fitting. The multivariate nonlinear regression model uses the data in time series segment b as the dependent variable and the data in each environmental data time series segment of time series b as an independent variable. The vector formed by all regression coefficients of the multivariate nonlinear regression model in a fixed order is denoted as the regression coefficient vector of time series segment b. This vector is used to characterize the degree of influence of each environmental factor (excluding weather conditions) on the photovoltaic power generation of photovoltaic power station P within the historical time period of time series b. The establishment of the multivariate nonlinear regression model is a well-known technique, and the specific process will not be described in detail.

[0032] Based on the differences in regression coefficient vectors between time series segment b and all its adjacent time series segments, the degree of change in the environmental contribution of time series segment b is calculated. This is used to assess whether the impact of each environmental factor (excluding weather conditions) on the photovoltaic power generation of photovoltaic power plant P has changed significantly during the historical period of time series segment b. The greater the difference in the regression coefficient vectors, the more significant the change in the degree of impact in the short term, and the greater the degree of change in the environmental contribution. Specifically, the positive fusion result of the differences in the regression coefficient vectors between time series segment b and all its adjacent time series segments is calculated to determine the degree of change in the environmental contribution of time series segment b. Positive fusion means combining multiple variables, which can be calculated using methods such as addition, multiplication, or averaging.

[0033] In this embodiment, the method for calculating the change in environmental contribution of time series segment b is as follows: In the formula, The environmental contribution of time series segment b. This represents the metric distance between time series segment b and its i-th adjacent time series segment, where n represents the number of adjacent time series segments of time series segment b, and f() is a user-defined normalization function used to normalize the regression coefficient vectors. The value is mapped to the range of -1 to 1, and the custom normalization function can be f(x) = The distance metric is either arctan(lnx) or f(x) = tanh(lnx), and this embodiment uses f(x) = tanh(lnx). This embodiment uses the distance metric to measure the difference in regression coefficient vectors between time series segment b and all its adjacent time series segments. Specifically, the distance metric uses Euclidean distance. Implementers can choose other feasible distance metric calculation methods, such as DTW distance. This represents the positive fusion result of the differences in the regression coefficient vectors between time series segment b and all its adjacent time series segments. It should be noted that during the normalization process using the function f(x) = tanh(lnx), if the independent variable x is 0, to ensure the calculation is meaningful, the environmental contribution change of the corresponding time series segment is directly set to -1.

[0034] S3. By determining the similarity of weather conditions between each time period and its adjacent time periods, the weather stability coefficient of each time period is determined. Combined with the environmental contribution variation, the noise interference degree of each time period is obtained.

[0035] Obtain the time series C corresponding to the weather type data in the historical photovoltaic dataset A of the photovoltaic power station P in the target mountain area after pre-cleaning and normalization. From all time series segments of time series C, obtain the time series segments with the same timestamp as time series segment b and each of its adjacent time series segments. These segments are used to characterize the weather changes of photovoltaic power station P in the historical time period of time series segment b and each of its adjacent time series segments.

[0036] Based on the similarity of weather types between time series segment b and all its adjacent time series segments, the weather stability coefficient of time series segment b is calculated. This coefficient is used to assess whether the weather conditions within the historical time period in which time series segment b is located have undergone significant changes in the short term. The higher the similarity, the smaller the changes in weather conditions in the short term, and thus the larger the weather stability coefficient. Specifically, the weather stability coefficient of time series segment b is calculated as follows: Obtain the word vector sequence of the time series consisting of the weather states at all times in the time period where time series segment b is located, and the word vector sequence of the time series consisting of the weather states at all times in the time period where each adjacent time series segment b is located. The expression for the weather stationarity coefficient of time series segment b is: In the formula, G2 represents the weather stability coefficient of time series segment b; represents the cosine similarity of word vector sequences between time series segment b and its i-th adjacent time series segment; n represents the number of adjacent time series segments of time series segment b.

[0037] Taking time series segment b as an example, the product of the environmental contribution variability and the weather stability coefficient of time series segment b is used as the noise interference level of time series segment b. This is used to assess whether there are abnormal data changes (i.e., noise interference) in all photovoltaic power generation data and environmental data collected by photovoltaic power plant P within the historical time period of time series segment b. The larger the environmental contribution variability and the smaller the weather stability coefficient, the more likely there is noise data in the photovoltaic power generation data and environmental data, resulting in a higher noise interference level.

[0038] S4. Based on the correlation between photovoltaic power generation and various environmental data, assess the possibility that the data collected at each time point is noise data, calculate the correlation enhancement coefficient of the data points at each time point, and determine the sliding window size when smoothing the collected data at each time point by combining the noise interference of the time period to which each time point belongs.

[0039] Since noise data typically refers to outliers caused by factors such as electronic noise within sensors, unstable signal transmission, and environmental electromagnetic interference, the noise data present in the collected photovoltaic power generation data and environmental data often disrupts or weakens the physical correlation that should exist between the photovoltaic power generation data and environmental data. For example, the presence of noise data reduces the overall positive correlation between the collected photovoltaic power generation data and solar radiation intensity data. Therefore, when the correlation between the collected photovoltaic power generation data and environmental data is significantly improved after removing a certain data, it usually indicates that the removed data is noise data that does not conform to the physical laws between photovoltaic power generation data and environmental data, rather than normal data.

[0040] Based on the above analysis, in order to more accurately distinguish between noisy data and normal data in the collected historical photovoltaic dataset, and to assign a relatively large window to noisy data and a relatively small window to normal data when using the moving average method to smooth the data in the historical photovoltaic dataset, thereby effectively suppressing noisy data in the historical photovoltaic dataset while retaining normal data in the historical photovoltaic dataset that can reflect the real power generation characteristics of distributed photovoltaic in mountainous areas, the following processing is performed.

[0041] Specifically, taking any photovoltaic power generation data point in time series segment b as the corresponding time point... For example, by comparing the correlation between time series segment b and all its environmental data time series segments, and at the removal time... Photovoltaic power generation data, all environmental data time series segments at time After obtaining environmental data, the correlation between time series segment b and all its environmental data time series segments is calculated. The correlation enhancement coefficient of the corresponding data points is used to characterize the removal time. The correlation between photovoltaic power generation data and environmental data is analyzed, and the degree of improvement in correlation is considered. A higher correlation improvement coefficient indicates that the photovoltaic power station P at time [time value missing] is performing well. The collected photovoltaic power generation data and environmental data may contain noise data that does not conform to the physical laws between photovoltaic power generation data and environmental data.

[0042] In this embodiment, time The expression for the correlation enhancement coefficient of the corresponding data points is: In the formula, U represents time. The correlation enhancement coefficient for corresponding data points; This indicates the degree of correlation between time series segment b and its j-th environmental data time series segment. Indicates the time of removal After collecting all types of data, the correlation between time series segment b and its j-th environmental data time series segment is determined. This embodiment uses the absolute value of the Spearman correlation coefficient to quantify the correlation. Implementers can use other methods for quantification, such as the absolute value of cosine similarity. m represents the number of environmental data time series segments in time series segment b; in this embodiment, m=4. The preset value is a very small positive number greater than 0, used to avoid a denominator of 0. In this embodiment... The implementer can set it according to the actual situation, and this embodiment does not impose any restrictions on it.

[0043] In another embodiment, time The expression for the correlation enhancement coefficient of the corresponding data points is: In the formula, U represents time. The correlation enhancement coefficient for corresponding data points; This represents the absolute value of the Spearman correlation coefficient between time series segment b and its j-th environmental data time series segment. Indicates the time of removal After collecting all types of data, the absolute value of the Spearman correlation coefficient between time series segment b and its j-th environmental data time series segment, where e is the natural constant and m is the number of environmental data time series segments in time series segment b.

[0044] The noise interference of all time series segments in time series B is normalized using the Min-Max normalization method, and the correlation enhancement coefficient of all data points in time series B is also normalized, so that the noise interference and correlation enhancement coefficient after normalization are both within the value range of [0,1]. The Min-Max normalization method is a well-known technique, and the specific process will not be described in detail.

[0045] The mean of the normalized results of the correlation enhancement coefficient and the normalized results of the noise interference in the time period to which each data point belongs is calculated and used as the window adjustment coefficient for each time point. This coefficient is then used to adjust the window size corresponding to each data point when performing data smoothing processing on each type of data in the collected historical photovoltaic dataset using the moving average method. The larger the normalized results of the correlation enhancement coefficient and the noise interference, the more likely the photovoltaic power generation data and environmental data collected by the photovoltaic power plant P at the corresponding time point contain abnormal data that do not conform to the physical laws between photovoltaic power generation data and environmental data, which are not caused by weather changes. In this case, the window size should be larger to better smooth the noise data, thus requiring a larger window adjustment coefficient.

[0046] Based on the window adjustment coefficient at each time point, the window size is calculated when smoothing various types of data collected at each time point. The specific calculation method is as follows: L = L² + round(Q) (L1-L2)); where L represents the window size when smoothing the data of each type collected at each time, and Q represents the window adjustment coefficient at each time; L1 and L2 represent the preset maximum and minimum values ​​of the window size in the moving average method, respectively, and the maximum and minimum values ​​are both odd numbers, which can be obtained by experimental methods. In this embodiment, the maximum and minimum values ​​are set to 11 and 3, respectively; round() is a custom rounding function used to ensure that the output of the round() function is the nearest even number not greater than the input value.

[0047] Taking the historical photovoltaic dataset of photovoltaic power plant P after pre-cleaning as an example, the moving average method is used to smooth the data of each type in the historical photovoltaic dataset, including photovoltaic power generation data and each type of environmental data, to obtain the historical photovoltaic dataset E of photovoltaic power plant P after data smoothing. The window size of the moving average method is the window size L calculated above when smoothing each type of data point at each time point. The moving average method is a well-known technique, and the specific process will not be elaborated further.

[0048] S5 uses all types of data after smoothing to predict photovoltaic power generation.

[0049] Furthermore, training and testing sets are constructed to train the neural network model and achieve photovoltaic power generation prediction. In this embodiment, all types of data collected in the historical photovoltaic dataset E during one week of the 2nd, 5th, 8th, and 11th months of the year (the first week is selected in this embodiment) are used as the testing set, and the remaining data in the historical photovoltaic dataset E other than the testing set are used as the training set. This allows the training of the neural network model to cover photovoltaic data of the target mountain photovoltaic power station under all different seasons and weather conditions, thereby enabling the neural network model to learn more comprehensive power generation patterns.

[0050] Based on the acquired training and test sets, the neural network model is trained. In this embodiment, the neural network model is a Long Short-Term Memory (LSTM) neural network model. Implementers can choose other feasible existing neural network models, such as RNN neural network models. During training, this embodiment divides the time-series data of each type in the training and test sets according to a time step, forming individual time-series samples. In this embodiment, the time step is set to 4 hours, but implementers can set it according to their actual situation. During training, for any given time step, the time-series samples corresponding to various environmental data types at that time step are used as training data, and the time-series samples corresponding to photovoltaic power generation data in the next time step are used as labels. This completes the training of the neural network model. The root mean square error is selected as the evaluation index for the prediction results of the LSTM neural network model. The training of the LSTM neural network model is a well-known technique, and the specific process will not be elaborated further.

[0051] The various environmental data collected in real time at a complete time step of the photovoltaic power station are smoothed using the moving average method and then used as input to the trained LSTM neural network model. The output is the photovoltaic power generation prediction result of the photovoltaic power station for the next time step, thus realizing the prediction of photovoltaic power generation.

[0052] Based on the same inventive concept as the above method, this application embodiment also provides a mountain distributed photovoltaic power generation prediction system, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described mountain distributed photovoltaic power generation prediction methods.

[0053] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments of this specification have been described above. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0054] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0055] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A method for predicting distributed photovoltaic power generation in mountainous areas, characterized in that, The method includes the following steps: The system collects photovoltaic power generation data of the target mountain photovoltaic power station at various times, as well as various environmental data, including weather conditions. Determine the impact of various environmental data (excluding weather conditions) on the photovoltaic power generation of a photovoltaic power station within the same time period, analyze the differences in the impact between each time period and its adjacent time periods, and determine the degree of change in environmental contribution for each time period. By determining the similarity of weather conditions between each time period and its adjacent time periods, the weather stability coefficient of each time period is determined. Combined with the environmental contribution variation, the noise interference degree of each time period is obtained. Based on the correlation between photovoltaic power generation and various environmental data, the possibility that the data collected at each time point is noise data is assessed, the correlation enhancement coefficient of the data points at each time point is calculated, and the sliding window size is determined when smoothing the data collected at each time point by combining the noise interference of the time period to which each time point belongs. The photovoltaic power generation output is predicted using all types of data after smoothing.

2. The method for predicting distributed photovoltaic power generation in mountainous areas as described in claim 1, characterized in that, The determination of the impact of various environmental data (excluding weather conditions) on the photovoltaic power generation of a photovoltaic power station within the same time period includes: Using photovoltaic power generation at any given time period as the dependent variable and various environmental data (excluding weather conditions) as independent variables, a multivariate nonlinear regression model is established to obtain a regression coefficient vector composed of all regression coefficients. This vector is used to represent the degree of influence of various environmental data (excluding weather conditions) on the photovoltaic power generation of the photovoltaic power station within the given time period.

3. The method for predicting distributed photovoltaic power generation in mountainous areas as described in claim 2, characterized in that, The degree of change in environmental contribution is determined based on the positive fusion result of the differences in the regression coefficient vectors between each time period and all its adjacent time periods.

4. The method for predicting distributed photovoltaic power generation in mountainous areas as described in claim 1, characterized in that, The determination of the weather stability coefficient includes: The weather conditions at all times within each time period are converted into word vector sequences. The similarity between the word vector sequences of each time period and its adjacent time periods is calculated. The weather stability coefficient is positively correlated with the similarity.

5. The method for predicting distributed photovoltaic power generation in mountainous areas as described in claim 1, characterized in that, The noise interference level is the product of the environmental contribution variation degree and the weather stability coefficient for each time period, wherein the environmental contribution variation degree and the weather stability coefficient for each time period are both in the range of [-1, 1].

6. The method for predicting distributed photovoltaic power generation in mountainous areas as described in claim 1, characterized in that, The calculation of the correlation enhancement coefficient for data points at each time point includes: The correlation between photovoltaic power generation and various environmental data within the time period of each moment was analyzed before and after removing all types of data collected at each time point, and the correlation enhancement coefficient of the data points at each time point was obtained.

7. The method for predicting distributed photovoltaic power generation in mountainous areas as described in claim 1, characterized in that, Determining the sliding window size when smoothing the acquired data at each time point includes: The mean of the normalized result of the correlation enhancement coefficient of the data points at each time point and the normalized result of the noise interference degree of the time period to which each time point belongs is calculated and used as the window adjustment coefficient at each time point. Combined with the maximum and minimum values ​​of the sliding window size, the final sliding window size at each time point is calculated.

8. The method for predicting distributed photovoltaic power generation in mountainous areas as described in claim 7, characterized in that, Calculate the difference between the maximum value and the minimum value, obtain the rounded result of the product of the difference and the window adjustment coefficient, and the final sliding window size at each time step is the sum of the minimum value and the rounded result.

9. The method for predicting distributed photovoltaic power generation in mountainous areas as described in claim 1, characterized in that, The prediction of photovoltaic power generation includes: The smoothed data of various types is used as training data to train the neural network model, and the trained neural network model is used to obtain the prediction results of photovoltaic power generation.

10. A distributed photovoltaic power generation prediction system for mountainous areas, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-9.

Images