Long-term new energy daily power generation prediction method and system based on deep learning

Abstract

The application discloses a long-term new energy daily average power generation prediction method and system based on deep learning, and the method comprises the following steps: acquiring historical new energy power generation data and historical numerical weather forecast data of a target area, constructing new energy power generation sample time series and multi-dimensional space-time characteristic index data sample series; building a long-term power prediction model, training the model with the new energy power generation sample time series and the multi-dimensional space-time characteristic index data sample series as the training sample set; building a daily average power prediction model, training the model with the new energy power generation sample time series as the training sample set; inputting the acquired numerical weather forecast data into the trained power prediction model to obtain the predicted long-term new energy daily average power generation. Through the establishment of the long-term power prediction model and the daily average power prediction model of the target area, the application realizes the accurate prediction of the new energy daily average power generation of the target area according to the predicted numerical weather forecast data.

Description

Technical Field

[0001] This invention relates to the field of new energy power generation technology, and in particular to a long-term new energy daily average power generation prediction method and system based on deep learning. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art known to those skilled in the art.

[0003] With the increasing demand for energy, and considering that the energy sources we have long relied on, such as coal, oil, and natural gas, are non-renewable and cause serious environmental pollution, the development of renewable energy sources, including solar, wind, and ocean energy, is crucial to addressing energy and environmental issues and promoting sustainable global economic development and mitigating climate change. Currently, photovoltaic and wind power generation are the main forms of renewable energy generation. Due to the randomness and volatility of renewable energy generation, for long-term electricity trading, renewable energy power plants need to forecast long-term renewable energy generation capacity for the next month and quarter to provide a basis for long-term electricity transactions.

[0004] In long-term electricity trading, such as monthly electricity trading, it is necessary to comprehensively consider the total output and trends of regional wind power, photovoltaic power, and other new energy sources on a monthly and quarterly basis to provide a basis for signing long-term electricity trading contracts. However, current forecasts of new energy power generation are mainly based on short-term and ultra-short-term forecasts, which cannot provide a basis for long-term electricity trading.

[0005] Meanwhile, as meteorological factors such as wind and solar power increase in forecast uncertainty over time, and due to the randomness of these factors, the predicted power generation varies significantly across different time periods on the same day. Predicting the total renewable energy power generation for the entire day can effectively balance power generation and provide a clearer picture of long-term renewable energy power generation trends. In other words, long-term electricity trading, such as monthly electricity trading, relies more on predicting the total monthly power generation and the average daily power generation. Because the forecast period is long (e.g., 30-60 days), long-term electricity trading focuses on the daily average forecast value, which can also reduce forecast uncertainty to some extent. Therefore, using the predicted long-term daily average wind and solar power generation of the target area as a basis can more clearly demonstrate the region's power generation situation. Currently, there is a pressing need for long-term daily average wind and solar power generation forecasts for target areas.

[0006] Furthermore, most existing power generation prediction models for wind farms and photovoltaic farms only establish a mapping model between meteorological elements and the output power of wind and photovoltaic power generation. This one-to-one modeling scheme ignores the spatial correlation between meteorological elements and the output power of new energy power generation and the spatial fluctuation distribution characteristics of weather, which reduces the accuracy of power generation prediction. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a long-term, daily average power generation prediction method and system for renewable energy based on deep learning. Considering the long-term electricity trading demand and the instability of long-term prediction results, this invention predicts the long-term daily average power generation of renewable energy in a target region, providing a clearer picture of renewable energy generation. Based on wide-area spatiotemporal numerical weather forecast data and time-series data of actual renewable energy power generation in the target region, this invention establishes a long-term power prediction model and a daily average power prediction model for the target region using deep learning algorithms. Based on the trained models, it achieves highly accurate prediction of the long-term daily average power generation of the target region.

[0008] Firstly, this disclosure provides a long-term, daily average power generation prediction method for new energy sources based on deep learning, including:

[0009] Acquire historical renewable energy power generation data and historical numerical weather forecast data for the target area, and construct a sample time series of renewable energy power generation and a sample data series of multidimensional spatiotemporal characteristic indicators.

[0010] A long-term power prediction model was built, and the training sample set was used to train the model with the time series of new energy power generation samples and the multi-dimensional spatiotemporal characteristic index data sample series.

[0011] A daily average power prediction model was built, and the model was trained using the time series of new energy power generation samples as the training sample set.

[0012] The acquired numerical weather forecast data for the target area over a future time period is input into the trained long-term power generation prediction model to obtain the predicted long-term renewable energy power generation. The long-term renewable energy power generation is then input into the daily average power prediction model to obtain the predicted long-term renewable energy daily average power generation.

[0013] A further technical solution is that the long-term power prediction model is used to output a long-term new energy power generation prediction result with the same time scale as the numerical weather prediction data, based on the input future numerical weather forecast data.

[0014] The daily average power prediction model is used to output the long-term daily average power prediction result of new energy sources based on the input long-term new energy power generation prediction value.

[0015] A further technical solution is that the long-term power prediction model is a deep neural network model based on 2DCNN-GRU, and the training process of this model is as follows:

[0016] The model is trained using time series of new energy power generation samples and multidimensional spatiotemporal characteristic index data samples with the same time scale. The mean squared error is used as the loss function, and the Adam gradient descent method is used for iterative optimization until the loss function converges or the maximum number of iterations is reached, thus completing the training of the model.

[0017] A further technical solution is that the training process of the daily average power prediction model is as follows:

[0018] Based on the time series of new energy power generation samples, a time series of average daily power generation samples of new energy is constructed.

[0019] The model is trained using a training sample set consisting of time series of new energy power generation samples and time series of daily average new energy power generation samples.

[0020] Using mean squared error as the loss function, the Adam gradient descent method is used for iterative optimization until the loss function converges numerically or the maximum number of iterations is reached, thus completing the training of the model.

[0021] A further technical solution involves constructing the multidimensional spatiotemporal feature index data sample sequence as follows: acquiring historical numerical weather forecast data for the target area, screening feature indicators affecting new energy power generation, selecting historical numerical weather forecast data based on the feature indicators, and constructing a multidimensional spatiotemporal feature index data sample sequence.

[0022] A further technical solution involves preprocessing the acquired data after obtaining historical numerical weather forecast data corresponding to the characteristic indicators affecting the power generation of new energy sources.

[0023] The preprocessing includes data splicing, outlier handling, missing value handling, and normalization of the time series data of each indicator, ultimately forming a multidimensional spatiotemporal feature indicator data sample sequence with time, longitude, dimension, and indicator as data dimensions.

[0024] Secondly, this disclosure provides a long-term daily average power generation prediction system for new energy sources based on deep learning, including:

[0025] The data acquisition module is used to acquire historical renewable energy power generation data and historical numerical weather forecast data for the target area;

[0026] The data processing module is used to construct a time series of new energy power generation samples and a multi-dimensional spatiotemporal characteristic index data sample series based on the historical new energy power generation data and historical numerical weather forecast data of the target area.

[0027] The module for building and training a long-term power prediction model is used to build a long-term power prediction model and train it using the time series of new energy power generation samples and the multi-dimensional spatiotemporal characteristic index data sample series as training sample sets.

[0028] The module for building and training the daily average power prediction model is used to build the daily average power prediction model and train it using the time series of new energy power generation samples as the training sample set.

[0029] The power prediction module is used to input the acquired numerical weather forecast data of the target area for a future time period into the trained long-term power generation prediction model to obtain the predicted long-term renewable energy power generation, and then input the long-term renewable energy power generation into the daily average power prediction model to obtain the predicted long-term renewable energy daily average power generation.

[0030] A further technical solution is that the long-term power prediction model is used to output a long-term new energy power generation prediction result with the same time scale as the numerical weather prediction data, based on the input future numerical weather forecast data.

[0031] The daily average power prediction model is used to output the long-term daily average power prediction result of new energy sources based on the input long-term new energy power generation prediction value.

[0032] Thirdly, this disclosure also provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the steps of the method described in the first aspect.

[0033] Fourthly, this disclosure also provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the steps of the method described in the first aspect.

[0034] The above one or more technical solutions have the following beneficial effects:

[0035] 1. This invention provides a long-term new energy daily average power generation prediction method and system based on deep learning. Based on wide-area spatiotemporal numerical weather forecast data and actual new energy power generation time series data of the target area, a long-term power prediction model and a daily average power prediction model of the target area are established through deep learning algorithms. Based on the trained model, the long-term daily average power generation of the target area can be predicted with high accuracy.

[0036] 2. The power generation predicted by this invention takes into account the spatial correlation between meteorological factors and the output power of new energy power generation, as well as the spatial fluctuation distribution characteristics of weather, thereby improving the accuracy of power generation prediction.

[0037] 3. This invention, by constructing and training a daily average power prediction model, enables the prediction of the long-term daily average power generation in the target area, which facilitates a clearer display of the new energy power generation situation and provides data support and basis for power plant medium- and long-term power transactions. Attached Figure Description

[0038] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0039] Figure 1 This is an overall flowchart of the method described in Embodiment 1 of the present invention;

[0040] Figure 2 This is a flowchart of the training of the long-term power prediction model according to Embodiment 1 of the present invention;

[0041] Figure 3 This is a flowchart of the training of the daily average power prediction model according to Embodiment 1 of the present invention. Detailed Implementation

[0042] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, 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 invention pertains.

[0043] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0044] Example 1

[0045] This embodiment provides a long-term, daily average power generation prediction method for new energy sources based on deep learning, such as... Figure 1 As shown, based on existing NWP forecast products, numerical weather prediction data is acquired and input into a trained power prediction model to predict the average daily power output of new energy sources in the target area for the next 30 to 60 days. The power prediction method described in this embodiment specifically includes the following steps:

[0046] Step 1: Obtain historical renewable energy power generation data for the target area, preprocess the obtained data, and construct a sample time series of renewable energy power generation.

[0047] In step 1 above, historical renewable energy power generation data for at least one full year is obtained for the target area. This historical data is then preprocessed, including data stitching, outlier handling, and missing value handling, to form a renewable energy power generation sample time series Y = [y1, y2, y3, ..., y κ ] T , where y i Let i represent the historical renewable energy power generation at time i, where i = 1, 2, 3, ..., k.

[0048] Step 2: Obtain historical numerical weather forecast data for the target area, screen characteristic indicators affecting the power generation of new energy sources, select historical numerical weather forecast data based on the characteristic indicators, and construct a multi-dimensional spatiotemporal characteristic indicator data sample sequence.

[0049] In step 2 above, historical numerical weather prediction (NWP) data for the target area is acquired. Considering that the subsequent power prediction model needs to base its predictions on weather conditions for the predicted future time period, this embodiment uses existing NWP forecast products to acquire NWP data for the target area for the future time period. Numerical Weather Prediction (NWP) is a method that, based on actual atmospheric conditions and certain initial and boundary conditions, uses large-scale computers to perform numerical calculations, solving fluid dynamics and thermodynamic equations describing weather evolution, and predicting atmospheric motion and weather phenomena for a certain future time period. NWP can provide multi-timescale prediction data for a global area, with forecast periods up to 3 months and a typical forecast time resolution of 6 hours. Using NWP to acquire weather forecast data for the future time period meets the 30-60 day timeframe requirement for power generation prediction in this embodiment. Therefore, in step 2 of this embodiment, historical NWP forecast data is also acquired based on NWP forecast products. The time interval of this historical numerical weather prediction data time series is 6 hours, for subsequent training of the power prediction model.

[0050] Furthermore, considering that NWP data contains various meteorological types, and that different meteorological types have different impacts on the power output of different new energy sources, this embodiment utilizes big data to screen out characteristic indicators affecting the power output of new energy sources. Based on these characteristic indicators, historical numerical weather forecast data is selected to construct a multi-dimensional spatiotemporal characteristic indicator sample sequence, thereby ensuring the accuracy of the trained power prediction model. New energy power generation includes wind power and photovoltaic power generation. The characteristic indicators affecting wind power generation include wind speed, wind direction, temperature, maximum temperature, minimum temperature, precipitation rate, specific humidity, total cloud cover, surface air pressure, and sea level air pressure at different altitudes. The characteristic indicators affecting photovoltaic power generation include downward longwave radiation flux, downward shortwave radiation flux, total cloud cover, clear-sky solar radiation downwards, temperature at different altitudes, maximum temperature, minimum temperature, precipitation rate, specific humidity, surface air pressure, and sea level air pressure. Combining historical new energy power generation data of the target area with the key indicators affecting the power output of new energy sources selected, a power prediction model is constructed and trained using a combination of deep learning methods to predict the output power of new energy sources in the target area for the next 30 to 60 days.

[0051] A further technical solution involves preprocessing the acquired data after obtaining historical numerical weather forecast data corresponding to the characteristic indicators affecting new energy power generation. This includes data stitching, outlier handling, and missing value handling. Simultaneously, considering the impact of different dimensions and large numerical differences between indicators on the power prediction model, adaptive normalization is performed on the time series data of each indicator. This ultimately forms a multi-dimensional spatiotemporal characteristic indicator data sample sequence X, with data dimensions of time, longitude, dimension, and indicator, where X = [x...]. (1) ,x (2) ,x (3) ,......,x (m) ], where x () Let x represent the spatiotemporal sequence sample composed of the data of the m-th feature index. (m) = [x1,x2,x3,......,x t ] T ,in, t represents the corresponding time, i represents longitude, and j represents latitude.

[0052] Combining steps 1 and 2 above, we know that the time resolution of NWP data (i.e., the time interval) is 6 hours, while the time resolution of historical renewable energy power generation data (i.e., the time interval) is 15 minutes. Here, we use historical renewable energy power generation data with a 15-minute time interval. Compared to the scheme of directly using the average of actual data with a 24-hour time interval for training, the model trained directly using the mean cannot achieve the prediction effect of the multi-model combination used in this embodiment. The average RMSE of the multi-model combination prediction results used in this embodiment can be reduced by about 10%, achieving a better prediction effect.

[0053] Because the time resolution of NWP data and historical renewable energy power generation data is inconsistent, in order to enable the prediction model to learn as many characteristics of historical renewable energy power generation as possible, this embodiment constructs a long-term power prediction model and a daily average power prediction model to learn the many-to-one time interval mapping relationship between NWP data and historical renewable energy power generation data at different time scales. This many-to-one time mapping relationship can be expressed as: in It is an integer greater than or equal to 0.

[0054] The prediction model described in this embodiment learns the above mapping relationship and constructs a long-term power prediction model and a daily average power prediction model, respectively. The long-term power prediction model is used to output a long-term renewable energy power generation prediction result at the same time scale as the input future numerical weather forecast data. The obtained long-term renewable energy power generation prediction result is input into the daily average power prediction model, which is used to output a long-term renewable energy daily average power generation prediction result based on the input long-term renewable energy power generation prediction value.

[0055] Step 3: Build a long-term power prediction model. Use the time series of new energy power generation samples and the multi-dimensional spatiotemporal characteristic index data sample series as the training sample set, and use the training sample set to train the long-term power prediction model.

[0056] The aforementioned long-term power prediction model is a deep neural network model based on 2DCNN-GRU. In this model, the 2DCNN neural network can extract both local and global spatial features of the data, possessing the characteristics of local connectivity and parameter sharing, and has advantages in processing multidimensional ordered spatiotemporal NWP data. When predicting the renewable energy power generation in a target area, using multidimensional spatiotemporal feature index data from the NWP data as input to train the model helps it learn the relationship between the spatial fluctuation distribution of weather and the regional renewable energy power generation. Simultaneously, renewable energy power generation sequence data exhibits long-term dependencies inherent in time-series data. The GRU network can connect data from different time periods, using update and reset gates as core modules to extract deeper features, thereby improving the accuracy of power prediction.

[0057] The aforementioned 2DCNN network structure includes Conv2D layers, MaxPooling2D layers, BatchNormalization layers, and Dense layers. It uses Same convolution and ReLU as the activation function. Figure 2 As shown, a deep neural network model based on 2DCNN-GRU is trained using time series of new energy power generation samples with the same time scale and multi-dimensional spatiotemporal feature index data samples, i.e., data sequences within the same time interval (6 hours). First, the input data enters the 2DCNN network. After operations such as convolution, pooling, and regularization, the convolutional layer detects local connections of features, and the pooling layer merges similar features to extract data features. Then, the feature map extracted by the 2DCNN network is stretched into multiple one-dimensional vectors and input into the GRU network. The GRU network structure includes an input layer, an output layer, and hidden layers. This embodiment uses a 3-layer GRU iterative network, and the predicted low-temporal-resolution new energy power generation output is output by a fully connected architecture.

[0058] In this embodiment, the training sample set is divided into a training set, a validation set, and a test set according to requirements. The training set is used for training the model relation mapping, the validation set is used for tuning the model's hyperparameters, and the test set is used to test the generalization ability of the prediction model. After the model is trained using the training set, the model's accuracy is verified by inputting the validation set data into the model. If it meets the requirements, the new energy power generation prediction model is output; otherwise, the model's hyperparameters are tuned based on the validation set data until the requirements are met. The long-term new energy power generation prediction model is then tested using the test set data.

[0059] In this embodiment, a deep neural network model is trained based on the training set and validation set. Both the 2DCNN network and the GRU network use Mean Square Error (MSE) as the loss function, calculated as follows:

[0060]

[0061] Where n is the number of samples, y is the model's predicted value. i This represents the actual power generation capacity of new energy sources.

[0062] The Adam gradient descent method is used for iterative optimization until the loss function converges numerically or the maximum number of iterations is reached, thus completing the training of the 2DCNN-GRU model.

[0063] Finally, the root mean square error (RMSE) was selected as the evaluation accuracy, and the test set was used to evaluate the performance of the established long-term renewable energy power generation prediction model. The calculation formula is as follows:

[0064]

[0065] Where n is the number of samples, y is the model's predicted value. i This represents the actual photovoltaic power generation.

[0066] Step 4: Build a daily average power prediction model and train it using the time series of new energy power generation samples as the training sample set.

[0067] like Figure 3 As shown, the above-mentioned daily average power prediction model takes the energy generation power data output by the long-term power prediction model as input and outputs the final daily average power generation prediction result of new energy. In this embodiment, firstly, a daily average power generation sample time series of new energy is constructed based on the new energy power generation sample time series. Then, the daily average power prediction model is trained using the new energy power generation sample time series and the daily average power generation sample time series of new energy. Specifically, a continuous date sequence is formed by concatenating four moments with a time resolution of 6 hours each day, resulting in a low time resolution new energy power generation sample time series E = [e (1) ,e (2) ,...,e (n) ], n=4, where e (n) For the nth time point, the continuous historical renewable energy power generation sequence sample, e (n) =[e1,e2,...,e t ] T t represents the date; the daily average power generation of new energy sources for the corresponding date is extracted to form a daily average power generation sample time series F = [f1, f2, ..., t ] T t represents the date.

[0068] It should be noted that the above-mentioned prediction of the daily average power is based on the power generation at four times (2:00, 8:00, 14:00, and 20:00), rather than directly using the sum of the predicted power at these four times as the daily average power. This is because the sum of the power at these four times has a certain gap with the actual daily average power value. Especially in photovoltaic power prediction, the maximum power of the day is generally between 11:00 and 1:00. At this time, the prediction results of the above four times generally cannot capture the time of maximum power. If the sum of these four points is used as the average, the predicted daily average value will be too low. Therefore, this embodiment uses the above-mentioned model learning method to learn and establish the mapping relationship between the four times and the daily average value, thereby improving the prediction accuracy of the daily average value. Compared with the summation and averaging method, it can achieve a daily average value prediction that is closer to the actual power generation.

[0069] The structure of the above-mentioned daily average power prediction model includes an input layer, an output layer, and a hidden layer. There are four hidden layers, with 96, 64, 32, and 16 neurons in each layer, respectively. Each layer uses the ReLU function as the activation function, and a combination of BatchNormalization and Dropout terms is selected to prevent the model from overfitting prematurely.

[0070] The above-mentioned daily average power prediction model is trained based on the training set and validation set, and the parameters of the network neurons are continuously updated iteratively using the Adam gradient descent method. Finally, the performance of the established daily average power prediction model for new energy power generation is evaluated using the test set.

[0071] In this embodiment, the above model uses Mean Square Error (MSE) as the loss function and Root Mean Square Error (RMSE) as the accuracy evaluation to evaluate the constructed and trained model. The calculation formula is as follows:

[0072]

[0073]

[0074] Where n is the number of samples, y is the model's predicted value. i This represents the actual power generation capacity of new energy sources.

[0075] Step 5: Input the acquired numerical weather forecast data for the target area over a future time period into the trained long-term renewable energy power prediction model to obtain the predicted long-term renewable energy power generation. Then, input the long-term renewable energy power generation into the daily average power prediction model to obtain the predicted long-term daily average renewable energy power generation. Specifically, based on existing NWP forecast products, acquire numerical weather forecast data for the target area over a future time period. Input this NWP data into the power prediction model. Through the long-term renewable energy power prediction model, obtain the photovoltaic power output at the same time interval as the NWP data over the future time period. Then, input this power into the daily average power prediction model to obtain the final long-term (30-60 days) daily average renewable energy power generation prediction result.

[0076] This embodiment establishes a long-term power prediction model and a daily average power prediction model for the target area using deep learning algorithms. Based on numerical weather forecast data for the target area in the future time period, it outputs a highly accurate daily average power generation prediction for the target area. The predicted power generation takes into account the spatial correlation between meteorological elements and the output power of new energy power generation, as well as the spatial fluctuation distribution characteristics of weather, thus improving the accuracy of power generation prediction. Moreover, by constructing and training the daily average power prediction model, it realizes the prediction of daily average power generation, which facilitates a clearer display of the new energy power generation situation and provides data support and basis for power plant medium- and long-term power trading.

[0077] Example 2

[0078] This embodiment provides a long-term, daily average power generation prediction system for new energy sources based on deep learning, including:

[0079] The data acquisition module is used to acquire historical renewable energy power generation data and historical numerical weather forecast data for the target area;

[0080] The data processing module is used to construct a time series of new energy power generation samples and a multi-dimensional spatiotemporal characteristic index data sample series based on the historical new energy power generation data and historical numerical weather forecast data of the target area.

[0081] The module for building and training a long-term power prediction model is used to build a long-term power prediction model and train it using the time series of new energy power generation samples and the multi-dimensional spatiotemporal characteristic index data sample series as training sample sets.

[0082] The module for building and training the daily average power prediction model is used to build the daily average power prediction model and train it using the time series of new energy power generation samples as the training sample set.

[0083] The power prediction module is used to input the acquired numerical weather forecast data of the target area for a future time period into the trained long-term power generation prediction model to obtain the predicted long-term renewable energy power generation, and then input the long-term renewable energy power generation into the daily average power prediction model to obtain the predicted long-term renewable energy daily average power generation.

[0084] Example 3

[0085] This embodiment provides an electronic device, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, they complete the steps in the long-term new energy daily average power generation prediction method described above.

[0086] Example 4

[0087] This embodiment also provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps in the deep learning-based long-term daily average power generation prediction method for new energy sources as described above.

[0088] The steps and methods involved in Embodiments 2 to 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0089] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0090] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0091] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A long-term, daily average power generation prediction method for new energy sources based on deep learning, characterized in that... include: Historical new energy power generation data and historical numerical weather forecast data of the target area are obtained, and a time series of new energy power generation samples and a multidimensional spatiotemporal characteristic index data sample sequence are constructed. The construction process of the multidimensional spatiotemporal characteristic index data sample sequence is as follows: historical numerical weather forecast data of the target area are obtained, characteristic indicators affecting new energy power generation are screened, historical numerical weather forecast data are selected according to the characteristic indicators, and a multidimensional spatiotemporal characteristic index data sample sequence is constructed, finally forming a multidimensional spatiotemporal characteristic index data sample sequence with data dimensions of time, longitude, dimension, and index. A long-term power prediction model was constructed, and trained using time series data of new energy power generation samples and multi-dimensional spatiotemporal characteristic index data samples as training sample sets. The long-term power prediction model is a deep neural network model based on 2DCNN-GRU, and the training process of this model is as follows: The model is trained using time series of new energy power generation samples and multidimensional spatiotemporal characteristic index data samples with the same time scale. The mean squared error is used as the loss function, and the Adam gradient descent method is used for iterative optimization until the loss function converges or the maximum number of iterations is reached, thus completing the training of the model. The term "long-term" is 30-60 days. A daily average power prediction model is constructed and trained using the time series of new energy power generation samples as the training sample set; the training process of the daily average power prediction model is as follows: Based on the time series of new energy power generation samples, a time series of average daily power generation samples of new energy is constructed. The model is trained using a training sample set consisting of time series of new energy power generation samples and time series of daily average new energy power generation samples. Using mean squared error as the loss function, the Adam gradient descent method is used for iterative optimization until the loss function converges numerically or the maximum number of iterations is reached, thus completing the training of the model. The structure of the daily average power prediction model includes an input layer, an output layer, and a hidden layer. There are four hidden layers, with 96, 64, 32, and 16 neurons in each layer, respectively. Each layer uses the ReLU function as the activation function. At the same time, a combination of BatchNormalization and Dropout terms is selected to prevent the model from overfitting prematurely. The acquired numerical weather forecast data for the target area over a future time period is input into a trained long-term power generation prediction model to obtain the predicted long-term renewable energy power generation. This long-term renewable energy power generation is then input into a daily average power prediction model to obtain the predicted long-term daily average renewable energy power generation. The long-term power prediction model outputs a long-term renewable energy power generation prediction result at the same time scale as the input future numerical weather forecast data. The daily average power prediction model outputs a long-term daily average renewable energy power generation prediction result based on the input long-term renewable energy power generation prediction value.

2. The method for predicting long-term daily average power generation of new energy sources based on deep learning as described in claim 1, characterized in that, After obtaining historical numerical weather forecast data corresponding to the characteristic indicators affecting the power generation of new energy sources, the obtained data is preprocessed. The preprocessing includes data splicing, outlier handling, missing value handling, and normalization of the time series data of each indicator.

3. A long-term daily average power generation prediction system for new energy sources based on deep learning, as described in any one of claims 1-2, characterized in that, include: The data acquisition module is used to acquire historical renewable energy power generation data and historical numerical weather forecast data for the target area; The data processing module is used to construct a time series of new energy power generation samples and a multi-dimensional spatiotemporal characteristic index data sample series based on the historical new energy power generation data and historical numerical weather forecast data of the target area. The module for building and training a long-term power prediction model is used to build a long-term power prediction model and train it using the time series of new energy power generation samples and the multi-dimensional spatiotemporal characteristic index data sample series as training sample sets. The module for building and training the daily average power prediction model is used to build the daily average power prediction model and train it using the time series of new energy power generation samples as the training sample set. The power prediction module is used to input the acquired numerical weather forecast data of the target area for a future time period into the trained long-term power generation prediction model to obtain the predicted long-term renewable energy power generation, and then input the long-term renewable energy power generation into the daily average power prediction model to obtain the predicted long-term renewable energy daily average power generation.

4. An electronic device, characterized in that: It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the processor executes the computer instructions, it completes the steps of a deep learning-based long-term daily average power generation prediction method for new energy sources as described in any one of claims 1-2.

5. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps of a long-term new energy daily average power generation prediction method based on deep learning as described in any one of claims 1-2.

Images