Wind power plant monthly generating capacity prediction method and system based on meteorological similar daily clustering

Abstract

The invention relates to the technical field of wind power prediction. The invention provides a wind power plant monthly generating capacity prediction method and system based on meteorological similar daily clustering. The method comprises the following steps: collecting, standardizing and preprocessing historical monthly meteorological characteristic data, generating capacity data and macroscopic meteorological background data; constructing a monthly meteorological similarity evaluation index system, and screening key meteorological factors; calculating the weight by adopting an entropy weight method, and constructing a weighted Euclidean distance similarity measurement model; selecting an optimal similar monthly sample based on K-means clustering; training a prediction model based on an attention mechanism and a long-short term memory neural network by taking similar sample meteorological characteristics as input and generating capacity as output; and performing secondary correction on the initial prediction value in combination with a macroscopic meteorological background anomaly correction coefficient, and outputting a final monthly generating capacity prediction result. The problems of low prediction precision and poor stability caused by error accumulation and insufficient similar day strategies when an existing wind power short-term power prediction method is expanded to monthly generating capacity prediction are solved.

Description

Technical Field

[0001] This invention relates to the field of wind power prediction technology, and more specifically, to a method and system for predicting monthly power generation of wind farms based on meteorological similarity day clustering. Background Technology

[0002] As an important clean energy source, wind power prediction is crucial for power system dispatch, operation, and market transactions. Several methods have been developed for wind power prediction in the existing technology. For example, patent document CN117060407A discloses a wind power cluster power prediction method based on similar day division. This method clusters the target wind farm cluster, selects representative units, and uses a BiGRU model to predict their short-term power. Finally, it combines the power transformation relationship of historical similar days to calculate the short-term power prediction value of the entire cluster. Another patent document, CN119443338A, proposes a short-term power prediction method and system for wind farms based on meteorological element similar days. Targeting multi-energy complementary scenarios of hydropower, wind power, and solar power, it clusters different types of power stations, establishes single-energy candidate prediction models, and selects the optimal combination model to ultimately achieve short-term power prediction for combined wind, solar, and hydropower. The above-mentioned existing technologies all improve prediction accuracy by introducing similar days or clustering concepts, but they mainly focus on short-term (hourly or day-ahead) power prediction.

[0003] However, existing methods face new challenges when the forecasting target is extended from short-term power to a longer scale of monthly power generation. Monthly power generation forecasting not only requires models to capture the long-term effects of complex factors such as weather and topography, but also places higher demands on forecast stability and data utilization efficiency. On the one hand, directly applying short-term forecasting models to a monthly scale will severely affect accuracy due to the gradual accumulation of errors; on the other hand, existing strategies for selecting short-term similar days are insufficient to fully characterize macroscopic patterns such as seasonal and interannual variations affecting monthly power generation. Therefore, how to construct a method that can effectively utilize the principle of meteorological similarity to accurately and stably predict the monthly power generation of wind farms has become an urgent problem to be solved in the current technical field. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for predicting monthly power generation of wind farms based on meteorological similarity day clustering, which aims to solve the problems of low prediction accuracy and poor stability caused by error accumulation and insufficient similarity day strategy when the existing short-term power prediction method for wind power is extended to monthly power generation prediction.

[0005] This invention is achieved through the following technical solution:

[0006] A method for predicting monthly power generation of wind farms based on meteorological similarity day clustering includes the following steps:

[0007] Historical monthly meteorological characteristic data, historical monthly power generation data, and regional macro-meteorological background data of the target wind farm were collected. The three types of data were standardized and preprocessed to remove outliers and form the corresponding monthly sample dataset.

[0008] A monthly meteorological similarity evaluation index system was constructed, and key local meteorological factors affecting monthly power generation were screened based on meteorological characteristic data in the monthly sample dataset.

[0009] The entropy weight method is used to calculate the weights of each key meteorological factor, and a weighted Euclidean distance similarity measurement model is constructed.

[0010] Based on the clustering algorithm, monthly similar day samples are selected. According to the similarity measurement model, the K-means clustering algorithm is used to cluster the monthly sample dataset to obtain several sample clusters with similar meteorological characteristics. From each sample cluster, the top N samples with the highest similarity to the meteorological characteristics of the month to be predicted are selected as the optimal similar monthly samples.

[0011] Meteorological feature data from the best similar monthly samples are used as input, and the corresponding historical monthly power generation data are used as output. A monthly power generation prediction model based on attention mechanism and long short-term memory artificial neural network is trained. The contribution weight of key meteorological factors to power generation is strengthened through attention mechanism.

[0012] Meteorological forecast data for the month to be predicted is obtained, and after standardized preprocessing, it is input into the trained monthly power generation prediction model to obtain the initial monthly power generation prediction value. Combined with the macro-meteorological background anomaly correction coefficient for the month to be predicted, the initial prediction value is corrected a second time, and the final monthly power generation prediction result of the target wind farm is output.

[0013] Optionally, the specific process of constructing a monthly meteorological similarity evaluation index system, and screening key meteorological factors affecting monthly power generation based on meteorological characteristic data in the monthly sample dataset, is as follows:

[0014] Candidate meteorological factors related to the long-term cumulative effect of wind farm energy capture process are extracted from meteorological characteristic data. Candidate meteorological factors include monthly average wind speed, effective duration of wind speed, monthly cumulative precipitation, and monthly average air pressure.

[0015] The correlation between each candidate meteorological factor and historical monthly power generation data was analyzed using the Pearson correlation coefficient method, and meteorological factors with correlation coefficients higher than the preset threshold were selected.

[0016] Multicollinearity diagnosis was performed on the screened meteorological factors, and meteorological factors with variance inflation factors greater than the preset value were removed, and the key meteorological factors were finally determined.

[0017] Optionally, the specific process of calculating the weights of each key meteorological factor using the entropy weight method and constructing a weighted Euclidean distance similarity measurement model is as follows:

[0018] Based on key meteorological factors, the observation values ​​of each key meteorological factor are obtained on all samples in the monthly sample dataset;

[0019] The observed values ​​of each key meteorological factor were standardized to eliminate the influence of dimensions.

[0020] For each standardized key meteorological factor, its proportion in all samples is calculated;

[0021] Calculate the information entropy of each key meteorological factor based on its corresponding weight.

[0022] Based on information entropy, the entropy weight of each key meteorological factor is calculated. The entropy weight is used to characterize the objective weight of the corresponding key meteorological factor in measuring monthly meteorological similarity.

[0023] By using the entropy weights of each key meteorological factor obtained through calculation, a similarity measurement model based on weighted Euclidean distance is constructed. The similarity measurement model is used to calculate the weighted distance between any two monthly samples on meteorological characteristics. The smaller the distance, the higher the similarity between the monthly meteorological characteristics.

[0024] Optionally, when constructing the weighted Euclidean distance similarity measurement model, the influence of key factors of the regional-scale macro-meteorological background is simultaneously incorporated. The specific process is as follows:

[0025] Candidate macroeconomic factors with long-term correlation to the monthly power generation of the target wind farm are extracted from the same period's regional-scale macro-meteorological background data. These candidate macroeconomic factors include the monthly average atmospheric circulation index, monsoon intensity level, and regional-scale pressure field anomaly amplitude.

[0026] The Pearson correlation coefficient method was used to analyze the correlation between each candidate macroeconomic factor and historical monthly power generation data, and candidate macroeconomic factors with correlation coefficients higher than the first preset threshold were selected.

[0027] Multicollinearity diagnosis was performed on the screened candidate macroeconomic factors, and candidate macroeconomic factors with variance inflation factors greater than the second preset value were removed to determine the key factors of macro-meteorological background.

[0028] The objective weights of key macro-meteorological background factors were calculated using the entropy weight method.

[0029] The weights of key macro-meteorological background factors and key meteorological factors are normalized and fused to construct a composite weighted Euclidean distance similarity measurement model. The formula for calculating the composite weighted Euclidean distance is as follows:

[0030]

[0031] in, For the first Monthly sample and the first The composite weighted distance of monthly samples; The number of key meteorological factors; For the first The weights of key meteorological factors; For the first The first sample Standardized values ​​of key meteorological factors; For the first The first sample Standardized values ​​of key meteorological factors; The number of key factors in the macro-meteorological background; For the first The weights of key macro-meteorological background factors; For the first The first sample Standardized values ​​of key macro-meteorological background factors; For the first The first sample Standardized values ​​of key macro-meteorological background factors.

[0032] Optionally, the specific process of using the K-means clustering algorithm to cluster the monthly sample dataset into several sample clusters with similar meteorological characteristics is as follows:

[0033] The clustering process is initialized using a weighted Euclidean distance similarity metric as the distance standard for clustering.

[0034] Based on the complexity of meteorological features and the total number of samples, the number of clusters K in the K-means clustering algorithm is determined, and K cluster centroids are randomly initialized.

[0035] Based on the weighted Euclidean distance, calculate the weighted distance from each monthly sample in the monthly sample dataset to each cluster centroid;

[0036] Based on the principle of minimum weighted distance, each monthly sample is assigned to the cluster containing the nearest cluster centroid;

[0037] Based on the allocation results, the mean value of key meteorological factor features of all samples in each cluster is recalculated, and the cluster centroid of each cluster is updated.

[0038] Iteratively execute the sample allocation and centroid update steps until the position change of all cluster centroids is less than a preset threshold.

[0039] After completing the clustering, the cluster category and corresponding cluster centroid of each sample are recorded to obtain several sample clusters with similar meteorological characteristics.

[0040] Optionally, the specific construction process of the monthly power generation prediction model is as follows:

[0041] Based on the optimal similar monthly samples, a model training dataset is prepared; wherein, the standardized meteorological feature data corresponding to each optimal similar monthly sample is used as the input feature sequence, and the corresponding historical monthly power generation data is used as the target output label.

[0042] An initial monthly power generation prediction model was constructed, integrating an attention mechanism and a long short-term memory (LSTM) artificial neural network. The initial monthly power generation prediction model includes a sequentially connected input layer, an LSM network layer, an attention mechanism layer, a fully connected layer, and an output layer. The LSM network layer extracts deep feature information with temporal dependencies from the input feature sequence and outputs a hidden state sequence containing feature information from different time steps. The attention mechanism layer, connected to the LSM network layer, receives the hidden state sequence and dynamically strengthens the weights of key meteorological factors that significantly contribute to monthly power generation in the hidden state sequence by calculating attention weights.

[0043] The model training dataset is input into the initial monthly power generation prediction model for training. The predicted monthly power generation is calculated using forward propagation. The error between the predicted value and the target output label is calculated using the loss function. The model parameters are optimized using backpropagation until the model converges, thus completing the construction of the monthly power generation prediction model.

[0044] Optionally, the specific process of combining the macro-meteorological background anomaly correction coefficient of the month to be predicted to perform a second correction on the initial predicted value and output the final monthly power generation prediction result of the target wind farm is as follows:

[0045] Based on the same period's regional-scale macro-meteorological background data, calculate the degree of anomalousness of the macro-meteorological background indicators corresponding to the month to be predicted compared with the historical climate state of the same period.

[0046] Based on the degree of anomaly of macro-meteorological background indicators, and combined with the pre-established correlation database of meteorological background anomalies and power generation deviations, the correction coefficient of macro-meteorological background anomalies for the month to be predicted is matched and determined.

[0047] The initial monthly power generation forecast was corrected using the macro-meteorological background anomaly correction coefficient to obtain the second-corrected monthly power generation forecast.

[0048] The output is the monthly power generation forecast value after secondary correction, which is used as the final forecast result of the monthly power generation of the target wind farm.

[0049] Based on the same inventive concept, this invention also provides a wind farm monthly power generation prediction system based on meteorological similarity day clustering, used to implement the aforementioned wind farm monthly power generation prediction method based on meteorological similarity day clustering, including:

[0050] The data acquisition and preprocessing module is used to collect historical monthly meteorological characteristic data, historical monthly power generation data, and macro-meteorological background data of the target wind farm at the same time, and to perform standardization and outlier cleaning on the three types of data to form a standardized monthly sample dataset.

[0051] The key factor screening and weight configuration module is connected to the data acquisition and preprocessing module. It is used to extract and screen local key meteorological factors and macro meteorological background key factors affecting power generation from the standardized monthly sample dataset, and assign objective weights to the two types of key factors based on the entropy weight method.

[0052] The similarity measurement model construction module is connected to the key factor screening and weight configuration module. It is used to construct a composite weighted Euclidean distance similarity measurement model that integrates the local key meteorological factors and the macro meteorological background key factors and their corresponding objective weights, so as to quantify the meteorological similarity between any two-month samples.

[0053] The optimal similar monthly sample screening module is connected to the similarity measurement model construction module. It is used to perform cluster analysis on the monthly sample dataset using the composite weighted Euclidean distance similarity measurement model as the distance measurement standard and the K-means clustering algorithm to obtain sample clusters with similar meteorological characteristics. Then, it selects the top N historical monthly samples that are most similar to the meteorological characteristics of the month to be predicted from each sample cluster as the optimal similar monthly samples.

[0054] The monthly power generation prediction model module is connected to the data acquisition and preprocessing module and the optimal similar monthly sample screening module, respectively, and is used to construct and train a prediction model based on attention mechanism and long short-term memory artificial neural network. The model is trained by taking the meteorological feature data of the optimal similar monthly sample as input and the corresponding power generation data as output, and strengthening the contribution of key factors through attention mechanism. When the model is applied, it receives the preprocessed monthly meteorological forecast data to be predicted and outputs the initial monthly power generation prediction value.

[0055] The macro-background correction module, connected to the monthly power generation prediction model module, is used to determine the macro-meteorological background anomaly correction coefficient based on the degree of regional-scale macro-meteorological background anomaly of the month to be predicted, and to use the correction coefficient to perform a secondary correction on the initial monthly power generation prediction value, and output the final monthly power generation prediction result of the target wind farm.

[0056] Based on the same inventive concept, the present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the above-described method for predicting monthly power generation of wind farms based on meteorological similarity day clustering.

[0057] Based on the same inventive concept, the present invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for predicting monthly power generation of wind farms based on meteorological similarity day clustering.

[0058] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0059] By constructing an indicator system that integrates key local meteorological factors with regional macro-meteorological background, and introducing a similarity measurement model based on weighted Euclidean distance, the meteorological similarity affecting monthly power generation can be characterized more scientifically. Combined with the optimal similar monthly samples selected by clustering, long-term patterns such as seasonality and interannual variation are effectively captured, reducing the error accumulation problem caused by directly accumulating short-term forecasts from the source, and significantly improving the forecast accuracy and reliability at the monthly scale.

[0060] The entropy weight method is used to objectively determine the weights of each meteorological factor, avoiding subjective assignment bias. Furthermore, a prediction model combining attention mechanism and long short-term memory neural network is used to dynamically strengthen the contribution of key factors to power generation during training. This not only improves the model's prediction performance but also provides better interpretability of the meteorological-power generation correlation, offering decision-making basis for operators.

[0061] By standardizing and integrating multi-source meteorological and power generation data, a high-quality monthly sample dataset was formed. By clustering and screening similar days, the similarity patterns in historical data were fully explored, reducing the dependence on massive high-frequency data. While ensuring the prediction effect, the computational efficiency was improved. It can be flexibly applied to wind farms with different geographical and climatic characteristics, and has strong versatility and scalability.

[0062] Based on the model prediction, a second correction is made by combining the regional-scale macro-meteorological anomaly correction coefficient of the month to be predicted. This can effectively cope with the impact of large-scale meteorological changes such as climate anomalies and interannual fluctuations, and enhance the adaptability and anti-interference ability of the prediction system under actual complex meteorological conditions. Attached Figure Description

[0063] Figure 1 This is a flowchart illustrating the wind farm monthly power generation prediction method based on meteorological similarity day clustering, as an embodiment of the present invention.

[0064] Figure 2This is a schematic diagram of the structure of the wind farm monthly power generation prediction system based on meteorological similarity day clustering, according to an embodiment of the present invention. Detailed Implementation

[0065] The following is a detailed description of the embodiments, in conjunction with the accompanying drawings.

[0066] Reference Figure 1 A method for predicting monthly power generation of wind farms based on meteorological similarity day clustering includes the following steps:

[0067] Step 1: Collect historical monthly meteorological characteristic data, historical monthly power generation data, and regional macro-meteorological background data of the target wind farm. Perform standardization preprocessing on the three types of data, remove outliers, and form the corresponding monthly sample dataset.

[0068] In some embodiments, historical time-series data covering at least five full years can be collected from the target wind farm's monitoring and data acquisition system and weather stations. Specifically, this includes:

[0069] Historical monthly meteorological characteristic data: Extract and aggregate to form a local meteorological characteristic sequence with "month" as the statistical unit. Typical characteristics include, but are not limited to: monthly average wind speed, effective duration of monthly wind speed (e.g., number of hours ≥3m / s), monthly cumulative precipitation, monthly average air pressure, monthly average temperature, and monthly average air density.

[0070] Historical monthly power generation data: Extract and summarize the total power generation sequence with "month" as the statistical unit, in megawatt-hours (MWh).

[0071] Obtain large-scale historical meteorological reanalysis data or observational data from national or international climate data centers, contemporaneous with the target wind farm's location. Collect and process these data to form a macro-meteorological background sequence fully synchronized with the aforementioned monthly data timestamps. Typical data include, but are not limited to: monthly average atmospheric circulation indices (such as the North Atlantic Oscillation Index and the East Asian Monsoon Index), regional average sea level pressure field, 500 hPa geopotential height anomalies, and large-scale monsoon intensity levels.

[0072] The three types of data from different sources were strictly aligned and matched according to the "year-month" timestamp to ensure that each month's sample contained the corresponding three types of data. Outlier detection and cleaning were performed for each type of data sequence. Combining wind turbine operation logs (such as maintenance and power curtailment records), months with abnormally low power generation due to non-meteorological factors (such as prolonged shutdowns for maintenance or grid dispatch curtailment) were identified and removed. For meteorological feature data, statistical methods (such as the 3σ principle or box plot method) were used to identify and remove extreme outliers. For the few missing data points after removal, reasonable filling was performed using the average of adjacent months or time series interpolation. To eliminate differences in dimensions and orders of magnitude between different meteorological factors, the feature dimensions of the three types of cleaned data were standardized. For each feature in each type of data (such as monthly average wind speed, circulation index, etc.), the Z-Score standardization method was used.

[0073] The three types of data processed through the above steps are integrated to form a structured monthly sample dataset that can be used for model training and similarity analysis.

[0074] Step 2: Construct a monthly meteorological similarity evaluation index system, and screen key local meteorological factors that affect monthly power generation based on meteorological characteristic data in the monthly sample dataset.

[0075] In some embodiments, the specific process of constructing a monthly meteorological similarity evaluation index system and screening key meteorological factors affecting monthly power generation based on meteorological characteristic data in the monthly sample dataset is as follows:

[0076] From meteorological characteristic data, candidate meteorological factors related to the long-term cumulative effect of wind farm energy capture are extracted. These candidate meteorological factors include monthly average wind speed, effective wind speed duration, monthly cumulative precipitation, and monthly average air pressure. The set of candidate meteorological factors can be defined as follows: ,in, This represents the average monthly wind speed. Indicates the effective duration of monthly wind speed (e.g., the number of hours ≥3m / s). This indicates the cumulative monthly precipitation. This indicates the average monthly air pressure.

[0077] The Pearson correlation coefficient method was used to analyze the correlation between each candidate meteorological factor and historical monthly power generation data, and meteorological factors with correlation coefficients higher than a preset threshold were selected. For each candidate meteorological factor, its Pearson correlation coefficient with historical monthly power generation was calculated, as shown in the following formula:

[0078]

[0079] in, Let be the Pearson correlation coefficient, and let represent the th... Candidate meteorological factors Compared with historical monthly power generation The degree of linear correlation between them, with values ​​ranging from [-1, 1], where positive values ​​indicate positive correlation, negative values ​​indicate negative correlation, and the larger the absolute value, the stronger the correlation. Covariance is used to measure... and The correlation between the overall errors of the two variables; The standard deviation of the meteorological factors; The standard deviation of electricity generation; Indicates the first In the sample of the month, the first Observed values ​​of meteorological factors; Indicates the first The sample mean of each meteorological factor; Indicates the first Monthly power generation observations; This represents the sample mean of monthly power generation. This represents the total number of samples for each month.

[0080] The filtering criteria are shown in the following formula:

[0081]

[0082] in, A preset threshold (e.g., 0.3 or 0.4) is set for relevance screening. The resulting set of factors is shown in the following formula:

[0083]

[0084] in, This represents the set of filtered meteorological factors;

[0085] Multicollinearity was diagnosed on the selected meteorological factors, and those with variance inflation factors exceeding a preset value were removed, ultimately determining the key meteorological factors. For each factor... Construct the following regression model:

[0086]

[0087] in, and The regression coefficient represents the regression coefficient in the factor. When there is multicollinearity, compare it with all other factors. ( Perform linear regression. Indicates the intercept. Represents the coefficients of each independent variable; represents the random error term. Traversal Except All factors other than those mentioned above. The coefficient of determination is obtained by fitting the regression model using the least squares method. The variance inflation factor (VIF) is calculated as follows:

[0088]

[0089] in, This represents the variance inflation factor, used to measure the variance inflation factor. The variance inflation factor is eliminated because it is collinear with other factors, causing the variance of its regression coefficient to increase by a factor of magnitude. The conditions for eliminating the variance inflation factor are shown in the following formula:

[0090]

[0091] in, A preset threshold (5 or 10) is set for the variance inflation factor. When the condition is met, the factor is considered to be positive. If severe multicollinearity exists, the factors will be removed from the screened set of meteorological factors. Remove from the middle;

[0092] The final set of key meteorological factors is shown in the following formula:

[0093]

[0094] in, This represents the set of key meteorological factors, which are used to construct a similarity measurement model later.

[0095] Step 3: Calculate the weights of each key meteorological factor using the entropy weight method, and construct a similarity measurement model based on weighted Euclidean distance.

[0096] In some embodiments, the specific process of calculating the weights of each key meteorological factor using the entropy weight method and constructing a weighted Euclidean distance similarity measurement model is as follows:

[0097] Based on key meteorological factors, the observed values ​​of each key meteorological factor are obtained in all samples of the monthly sample dataset; the observed values ​​of each key meteorological factor are standardized to eliminate the influence of dimensions; for each standardized key meteorological factor, its proportion in all samples is calculated.

[0098] Based on the weight of each key meteorological factor, the information entropy of each key meteorological factor is calculated as follows:

[0099]

[0100] in, Indicates the first The information entropy of a key meteorological factor, derived from information theory, is used here to measure the dispersion or uncertainty of the distribution of observed values ​​for that meteorological factor. The larger the value, the smaller the difference of the factor among different samples, the less information it provides, and the lower its importance in distinguishing sample similarity; Indicates the first Key meteorological factors in the first The proportion in a monthly sample; stipulating when hour, Information entropy .

[0101] Based on information entropy, the entropy weight of each key meteorological factor is calculated. The entropy weight is used to characterize the objective weight of the corresponding key meteorological factor in measuring monthly meteorological similarity. The entropy weight of each key meteorological factor is shown in the following formula:

[0102]

[0103] in, Indicates the first Entropy weights of key meteorological factors; This indicates the total number of key meteorological factors; Indicates the first Entropy weights of key meteorological factors, subscripts Used to iterate through all One factor; entropy weight The larger the value, the higher the objective importance of the factor in measuring monthly weather similarity.

[0104] By using the entropy weights of each key meteorological factor obtained through calculation, a similarity measurement model based on weighted Euclidean distance is constructed. The similarity measurement model is used to calculate the weighted distance between any two monthly samples on meteorological characteristics. The smaller the distance, the higher the similarity between the monthly meteorological characteristics.

[0105] In some embodiments, when constructing the weighted Euclidean distance similarity measurement model, the influence of key factors of the regional-scale macro-meteorological background is simultaneously incorporated. The specific process is as follows:

[0106] Candidate macroeconomic factors with long-term correlation to the monthly power generation of the target wind farm are extracted from the same period's regional-scale macro-meteorological background data. These candidate macroeconomic factors include the monthly average atmospheric circulation index, monsoon intensity level, and regional-scale pressure field anomaly amplitude.

[0107] The Pearson correlation coefficient method was used to analyze the correlation between each candidate macroeconomic factor and historical monthly power generation data, and candidate macroeconomic factors with correlation coefficients higher than the first preset threshold were selected.

[0108] Multicollinearity diagnosis was performed on the screened candidate macroeconomic factors, and candidate macroeconomic factors with variance inflation factors greater than the second preset value were removed to determine the key factors of macro-meteorological background.

[0109] The objective weights of key macro-meteorological background factors were calculated using the entropy weight method.

[0110] The weights of key macro-meteorological background factors and key meteorological factors are normalized and fused to construct a composite weighted Euclidean distance similarity measurement model. The formula for calculating the composite weighted Euclidean distance is as follows:

[0111]

[0112] in, For the first Monthly sample and the first The composite weighted distance of monthly samples; The number of key meteorological factors; For the first The weights of key meteorological factors; For the first The first sample Standardized values ​​of key meteorological factors; For the first The first sample Standardized values ​​of key meteorological factors; The number of key factors in the macro-meteorological background; For the first The weights of key macro-meteorological background factors; For the first The first sample Standardized values ​​of key macro-meteorological background factors; For the first The first sample Standardized values ​​of key macro-meteorological background factors.

[0113] Step 4: Screen monthly similar day samples based on clustering algorithm. Based on the similarity measurement model, the K-means clustering algorithm is used to cluster the monthly sample dataset to obtain several sample clusters with similar meteorological characteristics. From each sample cluster, the top N samples with the highest similarity to the meteorological characteristics of the month to be predicted are selected as the optimal similar monthly samples.

[0114] In some embodiments, the specific process of using the K-means clustering algorithm to cluster the monthly sample dataset into several sample clusters with similar meteorological characteristics is as follows:

[0115] The clustering process is initialized using a weighted Euclidean distance similarity metric as the distance standard for clustering.

[0116] Based on the complexity of meteorological features and the total number of samples, the number of clusters K in the K-means clustering algorithm is determined, and K cluster centroids are randomly initialized.

[0117] Based on the weighted Euclidean distance, calculate the weighted distance from each monthly sample in the monthly sample dataset to each cluster centroid;

[0118] Based on the principle of minimum weighted distance, each monthly sample is assigned to the cluster containing the nearest cluster centroid;

[0119] Based on the allocation results, the mean value of key meteorological factor features of all samples in each cluster is recalculated, and the cluster centroid of each cluster is updated.

[0120] Iteratively execute the sample allocation and centroid update steps until the position change of all cluster centroids is less than a preset threshold.

[0121] After completing the clustering, the cluster category and corresponding cluster centroid of each sample are recorded to obtain several sample clusters with similar meteorological characteristics.

[0122] Step 5: Take the meteorological feature data from the best similar monthly sample as input and the corresponding historical monthly power generation data as output, and train a monthly power generation prediction model based on attention mechanism and long short-term memory artificial neural network. The attention mechanism is used to strengthen the contribution weight of key meteorological factors to power generation.

[0123] In some embodiments, the specific construction process of the monthly power generation prediction model is as follows:

[0124] Based on the optimal similar monthly samples, a model training dataset is prepared; wherein, the standardized meteorological feature data corresponding to each optimal similar monthly sample is used as the input feature sequence, and the corresponding historical monthly power generation data is used as the target output label.

[0125] An initial monthly power generation prediction model was constructed, integrating an attention mechanism and a long short-term memory (LSTM) artificial neural network. The initial monthly power generation prediction model includes a sequentially connected input layer, an LSM network layer, an attention mechanism layer, a fully connected layer, and an output layer. The LSM network layer extracts deep feature information with temporal dependencies from the input feature sequence and outputs a hidden state sequence containing feature information from different time steps. The attention mechanism layer, connected to the LSM network layer, receives the hidden state sequence and dynamically strengthens the weights of key meteorological factors that significantly contribute to monthly power generation in the hidden state sequence by calculating attention weights.

[0126] The model training dataset is input into the initial monthly power generation prediction model for training. The predicted monthly power generation is calculated using forward propagation. The error between the predicted value and the target output label is calculated using the loss function. The model parameters are optimized using backpropagation until the model converges, thus completing the construction of the monthly power generation prediction model.

[0127] Step 6: Obtain the meteorological forecast data for the month to be predicted, and after standardized preprocessing, input it into the trained monthly power generation prediction model to obtain the initial monthly power generation prediction value. Combine the macro-meteorological background anomaly correction coefficient for the month to be predicted to perform a second correction on the initial prediction value, and output the final monthly power generation prediction result of the target wind farm.

[0128] In some embodiments, the specific process of combining the macro-meteorological background anomaly correction coefficient of the month to be predicted, performing a secondary correction on the initial predicted value, and outputting the final monthly power generation prediction result of the target wind farm is as follows:

[0129] Based on the same period's regional-scale macro-meteorological background data, calculate the degree of anomalousness of the macro-meteorological background indicators corresponding to the month to be predicted compared with the historical climate state of the same period.

[0130] Based on the degree of anomaly of macro-meteorological background indicators, and combined with the pre-established correlation database of meteorological background anomalies and power generation deviations, the correction coefficient of macro-meteorological background anomalies for the month to be predicted is matched and determined.

[0131] The initial monthly power generation forecast was corrected using the macro-meteorological background anomaly correction coefficient to obtain the second-corrected monthly power generation forecast.

[0132] The output is the monthly power generation forecast value after secondary correction, which is used as the final forecast result of the monthly power generation of the target wind farm.

[0133] Based on the same inventive concept, and corresponding to any of the above embodiments, refer to... Figure 2 This invention provides a wind farm monthly power generation prediction system based on meteorological similarity day clustering, used to implement the aforementioned wind farm monthly power generation prediction method based on meteorological similarity day clustering, including:

[0134] The data acquisition and preprocessing module is used to collect historical monthly meteorological characteristic data, historical monthly power generation data, and macro-meteorological background data of the target wind farm at the same time, and to perform standardization and outlier cleaning on the three types of data to form a standardized monthly sample dataset.

[0135] The key factor screening and weight configuration module is connected to the data acquisition and preprocessing module. It is used to extract and screen local key meteorological factors and macro meteorological background key factors affecting power generation from the standardized monthly sample dataset, and assign objective weights to the two types of key factors based on the entropy weight method.

[0136] The similarity measurement model construction module is connected to the key factor screening and weight configuration module. It is used to construct a composite weighted Euclidean distance similarity measurement model that integrates the local key meteorological factors and the macro meteorological background key factors and their corresponding objective weights, so as to quantify the meteorological similarity between any two-month samples.

[0137] The optimal similar monthly sample screening module is connected to the similarity measurement model construction module. It is used to perform cluster analysis on the monthly sample dataset using the composite weighted Euclidean distance similarity measurement model as the distance measurement standard and the K-means clustering algorithm to obtain sample clusters with similar meteorological characteristics. Then, it selects the top N historical monthly samples that are most similar to the meteorological characteristics of the month to be predicted from each sample cluster as the optimal similar monthly samples.

[0138] The monthly power generation prediction model module is connected to the data acquisition and preprocessing module and the optimal similar monthly sample screening module, respectively, and is used to construct and train a prediction model based on attention mechanism and long short-term memory artificial neural network. The model is trained by taking the meteorological feature data of the optimal similar monthly sample as input and the corresponding power generation data as output, and strengthening the contribution of key factors through attention mechanism. When the model is applied, it receives the preprocessed monthly meteorological forecast data to be predicted and outputs the initial monthly power generation prediction value.

[0139] The macro-background correction module, connected to the monthly power generation prediction model module, is used to determine the macro-meteorological background anomaly correction coefficient based on the degree of regional-scale macro-meteorological background anomaly of the month to be predicted, and to use the correction coefficient to perform a secondary correction on the initial monthly power generation prediction value, and output the final monthly power generation prediction result of the target wind farm.

[0140] Based on the same inventive concept, corresponding to any of the above embodiments, the present invention provides an electronic device, including a memory and a processor. The memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the wind farm monthly power generation prediction method based on meteorological similarity day clustering of the embodiments.

[0141] Alternatively, the aforementioned electronic device may be a server.

[0142] In addition, this embodiment also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the wind farm monthly power generation prediction method based on meteorological similarity day clustering of the embodiment.

[0143] It is understood that the processor in the embodiments of the present invention may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0144] The method steps in the embodiments of the present invention can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.

[0145] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a storage medium or transmitted through a storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive (SSD)).

Claims

1. A wind farm monthly energy production prediction method based on meteorologically similar day clustering, characterized in that, The method comprises the following steps: Collecting historical monthly meteorological characteristic data, historical monthly power generation data and synchronous regional macro-meteorological background data of a target wind farm, standardizing and preprocessing the three types of data, and forming corresponding monthly sample data sets after removing outliers; Building a monthly meteorological similarity evaluation index system, screening local key meteorological factors affecting monthly power generation based on the meteorological characteristic data in the monthly sample data sets; Calculating the weight of each key meteorological factor by entropy weight method, and constructing a similarity measurement model of weighted Euclidean distance; Based on the clustering algorithm, similar monthly sample data is screened, and the K-means clustering algorithm is used to cluster the monthly sample data sets based on the similarity measurement model, to obtain a plurality of sample clusters with similar meteorological characteristics, and the first N samples with the highest similarity to the to-be-predicted monthly meteorological characteristics are selected from each sample cluster as optimal similar monthly samples; Taking the meteorological characteristic data in the optimal similar monthly samples as input and the corresponding historical monthly power generation data as output, a monthly power generation prediction model based on attention mechanism and long short-term memory artificial neural network is trained, and the contribution weight of the key meteorological factors to the power generation is strengthened through the attention mechanism; Obtaining the meteorological forecast data of the to-be-predicted month, inputting the data into the trained monthly power generation prediction model after standardization preprocessing, obtaining the initial monthly power generation prediction value, combining the macro-meteorological background anomaly correction coefficient of the to-be-predicted month, and performing secondary correction on the initial prediction value to output the final monthly power generation prediction result of the target wind farm.

2. The weather analog day clustering based wind farm monthly energy production prediction method of claim 1, wherein The specific process of building a monthly meteorological similarity evaluation index system based on the meteorological characteristic data in the monthly sample data sets and screening key meteorological factors affecting monthly power generation is as follows: From the meteorological characteristic data, candidate meteorological factors related to the long-term cumulative effect of the wind farm energy capture process are extracted, including monthly average wind speed, wind speed effective time, monthly cumulative precipitation and monthly average air pressure; The correlation between each candidate meteorological factor and the historical monthly power generation data is analyzed by using the Pearson correlation coefficient method, and the meteorological factors with a correlation coefficient higher than a preset threshold are screened out; The selected meteorological factors are subjected to multicollinearity diagnosis, and the meteorological factors with a variance inflation factor greater than a preset value are removed, and finally the key meteorological factors are determined.

3. The weather analog-based clustering of wind farm monthly energy production prediction method of claim 2, wherein, The specific process of calculating the weight of each key meteorological factor by entropy weight method and constructing a similarity measurement model of weighted Euclidean distance is as follows: Based on the key meteorological factors, the observation values of each key meteorological factor on all samples in the monthly sample data set are obtained; The observation values of each key meteorological factor are standardized for eliminating the influence of dimension; For each standardized key meteorological factor, the proportion of the key meteorological factor in all samples is calculated; According to the proportion of each key meteorological factor, the information entropy of each key meteorological factor is calculated; According to the information entropy, the entropy weight of each key meteorological factor is calculated, and the entropy weight is used to represent the objective weight of the corresponding key meteorological factor in measuring the monthly meteorological similarity; The entropy weight of each key meteorological factor is calculated, and a similarity measurement model of weighted Euclidean distance is constructed. The similarity measurement model is used to calculate the weighted distance of any two monthly samples in meteorological characteristics. The smaller the distance is, the higher the monthly meteorological similarity is.

4. The weather analog-based clustering of wind farm monthly energy production prediction method of claim 3, wherein, When the similarity measurement model of weighted Euclidean distance is constructed, the influence of the key factor of the regional scale macro-meteorological background is simultaneously taken into account, and the specific process is as follows: From the regional scale macro-meteorological background data of the same period, candidate macro factors that have long-term correlation with the monthly power generation of the target wind farm are extracted, and the candidate macro factors include monthly atmospheric circulation index, monsoon intensity grade, and regional scale pressure field anomaly amplitude. The correlation between each candidate macro factor and the historical monthly power generation data is analyzed by using the Pearson correlation coefficient method, and the candidate macro factors with a correlation coefficient higher than a first preset threshold are selected. Multiple collinearity diagnosis is performed on the selected candidate macro factors, and the candidate macro factors with a variance inflation factor greater than a second preset value are removed to determine the key factor of the macro-meteorological background. The objective weight of each key factor of the macro-meteorological background is calculated by using the entropy weight method. The weights of the key factors of the macro-meteorological background and the weights of the key meteorological factors are normalized and fused to construct a composite weighted Euclidean distance similarity measurement model, and the calculation formula of the composite weighted Euclidean distance is as follows: ； wherein, is the composite weighted distance of the th monthly sample and the th monthly sample; is the number of key weather factors; is the weight of the th key weather factor; is the normalized value of the th key weather factor of the th sample; is the normalized value of the th key weather factor of the th sample; is the number of macro weather background key factors; is the weight of the th macro weather background key factor; is the normalized value of the th macro weather background key factor of the th sample; is the normalized value of the th macro weather background key factor of the th sample.

5. The weather analog-based clustering of wind farm monthly energy production prediction method of claim 1, wherein, The specific process of clustering and grouping the monthly sample data set by using the K-means clustering algorithm to obtain a plurality of sample clusters with similar meteorological characteristics is as follows: The similarity measurement model of weighted Euclidean distance is used as the distance measurement standard for clustering to initialize the clustering process. Based on the complexity of meteorological characteristics and the total amount of samples, the number K of clusters of the K-means clustering algorithm is determined, and K cluster centroids are randomly initialized. The weighted distance of each monthly sample in the monthly sample data set to each cluster centroid is calculated according to the weighted Euclidean distance. Each monthly sample is assigned to the cluster in which the nearest cluster centroid is located according to the minimum weighted distance principle. Based on the assignment result, the mean value of the characteristic values of the key meteorological factors of all samples in each cluster is recalculated, and the cluster centroid of each cluster is updated. The sample assignment and centroid update steps are iteratively performed until the position change of all cluster centroids is less than a preset threshold. After clustering is completed, the cluster category and the corresponding cluster centroid to which each sample belongs are recorded to obtain a plurality of sample clusters with similar meteorological characteristics.

6. The weather analog-based clustering of wind farm monthly energy production prediction method of claim 1, wherein, The specific construction process of the monthly power generation prediction model is as follows: Based on the optimal similar monthly sample, a model training data set is prepared. In the model training data set, the standardized meteorological characteristic data corresponding to each optimal similar monthly sample is used as the input feature sequence, and the historical monthly power generation data corresponding to each optimal similar monthly sample is used as the target output label. The initial monthly power generation prediction model fusing the attention mechanism and the long short-term memory artificial neural network is constructed, and the initial monthly power generation prediction model comprises sequentially connected input layer, long short-term memory network layer, attention mechanism layer, full connection layer and output layer; wherein, the long short-term memory network layer is used to extract deep feature information with time sequence dependence from the input feature sequence, and output a hidden state sequence containing feature information of different time steps; the attention mechanism layer is connected with the long short-term memory network layer, used to receive the hidden state sequence, and through calculating the attention weight, dynamically strengthen the weight of time sequence features in the hidden state sequence which are significantly contributed by the key meteorological factors to the monthly power generation; The model training data set is input into the initial monthly power generation prediction model for training, to calculate the predicted monthly power generation by forward propagation, calculate the error between the predicted value and the target output label by loss function, and optimize the model parameters by back propagation until the model converges, to complete the construction of the monthly power generation prediction model.

7. The weather analog-based clustering of wind farm monthly energy production prediction method of claim 1, wherein, The specific process of combining the macro-weather background anomaly correction coefficient of the to-be-predicted month to perform secondary correction on the initial predicted value, and outputting the final target wind farm monthly power generation prediction result is as follows: Based on the regional scale macro-weather background data of the same period, the abnormal degree of the macro-weather background index of the to-be-predicted month compared with the historical same period climatic state is calculated; According to the abnormal degree of the macro-weather background index, the macro-weather background anomaly correction coefficient of the to-be-predicted month is matched and determined by combining the pre-established meteorological background anomaly-power generation deviation correlation relationship library; The macro-weather background anomaly correction coefficient is used to correct the initial monthly power generation prediction value, to obtain the secondary corrected monthly power generation prediction value; The secondary corrected monthly power generation prediction value is output as the final prediction result of the target wind farm monthly power generation.

8. A wind farm monthly energy production prediction system based on meteorologically similar day clustering, for implementing the wind farm monthly energy production prediction method based on meteorologically similar day clustering according to any one of claims 1-7, characterized in that, It comprises: A data acquisition and preprocessing module for acquiring historical monthly meteorological feature data, historical monthly power generation data and same period regional scale macro-weather background data of a target wind farm, and performing standardization and outlier cleaning processing on the three types of data to form a standardized monthly sample data set; A key factor screening and weight configuration module connected with the data acquisition and preprocessing module, used to extract and screen local key meteorological factors and macro-weather background key factors affecting power generation from the standardized monthly sample data set, and respectively assign objective weights to the two types of key factors based on entropy weight method; A similarity measurement model construction module connected with the key factor screening and weight configuration module, used to construct a composite weighted Euclidean distance similarity measurement model fusing the two types of factors according to the local key meteorological factors and macro-weather background key factors and their corresponding objective weights, to quantify the meteorological similarity degree between any two monthly samples; An optimal similar monthly sample screening module is connected with the similarity measure model construction module, configured to use the composite weighted Euclidean distance similarity measure model as a distance measurement standard, perform cluster analysis on the monthly sample data set by using a K-means clustering algorithm, obtain sample clusters similar in meteorological features, and select the first N historical monthly samples most similar to the monthly meteorological features to be predicted from each sample cluster as optimal similar monthly samples; A monthly power generation prediction model module is connected with the data collection and preprocessing module and the optimal similar monthly sample screening module, configured to construct and train a prediction model based on an attention mechanism and a long short-term memory artificial neural network; wherein the training of the model uses the meteorological feature data of the optimal similar monthly samples as input and the corresponding power generation data as output, and strengthens the contribution of key factors through the attention mechanism; in application, the model receives the preprocessed monthly meteorological forecast data to be predicted and outputs an initial monthly power generation prediction value; A macro background correction module is connected with the monthly power generation prediction model module, configured to determine a macro meteorological background anomaly correction coefficient according to the regional scale macro meteorological background anomaly degree of the monthly data to be predicted, and perform secondary correction on the initial monthly power generation prediction value by using the correction coefficient to output a final target wind farm monthly power generation prediction result.

9. An electronic device, comprising: An electronic device includes a memory for storing a computer program and a processor for running the computer program to enable the electronic device to perform the wind farm monthly power generation prediction method based on meteorological similar day clustering according to any one of claims 1-7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, A computer program is executed by a processor to implement the wind farm monthly power generation prediction method based on meteorological similar day clustering according to any one of claims 1-7.

Images