Hybrid neural network-based water-wind-solar power generation power joint prediction method and device
Through the GRU-CNN model based on hybrid neural network, combined with multi-source meteorological observation and historical power generation data, a joint prediction model of water, wind, and light power generation power is constructed, which solves the problems that have not been considered for the impact between water, wind and light energy in the existing technology, and improves the impact of water, wind and light energy in the existing technology. The accuracy of water wind and light power generation power prediction.
Patent Information
- Application Number
- CN202410914270.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-09
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-07-09
AI Technical Summary
The existing power prediction methods are aimed at hydropower, wind power and photovoltaic power generation alone, and fail to effectively consider the impact between water, wind and light energy sources, resulting in insufficient accuracy in water, wind and photovoltaic power generation power prediction.
The GRU-CNN model based on hybrid neural network is adopted, combined with multi-source meteorological observation data and historical power generation data, and through heterogeneous information encoding, spatiotemporal and spatial feature fusion and attention mechanism, a joint prediction model of water and wind light power generation power is constructed to make real-time predictions.
Through data assimilation, the initial field is improved, and the space-time correlation and load curve of water-wind and light power generation power are integrated, which improves the accuracy of water-wind and light power generation power prediction and enhances the ability to capture the correlation characteristics between water, wind and light energy.
Smart Images

Figure CN118983772A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the technical field of power generation prediction, and in particular to a method and device for joint prediction of water, wind and solar power generation based on a hybrid neural network. Background Art
[0002] Vigorously developing clean energy such as water, wind, and light is a major strategic measure to ensure future energy security and respond to global warming. Wind energy and solar energy are the new energy sources with the greatest prospects for large-scale development, but they are easily affected by meteorological factors, and their output power has strong random fluctuations and is difficult to predict. With the continuous advancement of the construction of wind and light power generation bases. Large-scale direct grid connection of wind and photovoltaic power will bring great pressure to the peak, peak-shaving and stable operation of the power system. Hydropower units have the characteristics of rapid start and stop, flexible operation, large output change amplitude, and fast response to load changes. They are ideal peak-shaving power sources. Through the natural complementarity of resources and the flexibility of hydropower, the aggregation of water, wind, and light energy to form a multi-energy complementary power generation system is an effective way to reduce the impact of new energy grid connection and increase the pass rate of basin resources.
[0003] The current power prediction methods all predict the power of hydropower, wind power and photovoltaic power generation separately, without considering the impact of water, wind and light energy. Therefore, how to comprehensively consider the impact of water, wind and light on each other and make a joint power prediction for hydropower, wind and photovoltaic power generation is an issue that needs to be considered to further improve the accuracy of hydropower, wind power and photovoltaic power prediction.
[0004] Therefore, a combined prediction method for hydropower, wind and solar power generation is needed to improve the accuracy of hydropower, wind and solar power generation prediction. Summary of the invention
[0005] In order to solve the above problems, the present invention provides a method and device for joint prediction of water, wind and solar power generation based on a hybrid neural network.
[0006] First, a hybrid neural network-based combined prediction method for hydropower, wind power, and solar power generation includes:
[0007] Obtain multi-source meteorological observation datasets for the target area;
[0008] Obtain the historical power generation data set of hydropower, wind and solar power stations in the target area;
[0009] Based on multi-source meteorological observation data sets and historical power generation data sets, a GRU-CNN hydro-wind-solar power generation joint prediction model was established and trained;
[0010] The meteorological forecast data of the target area in the predicted time period is obtained in real time, and transmitted to the trained GRU-CNN water-wind-solar power joint prediction model to calculate the hydropower, wind power and photovoltaic power of the water-wind-solar power stations in the target area in the predicted time period.
[0011] Furthermore, the multi-source meteorological observation data set includes: meteorological observation data of hydropower stations, wind power stations, satellite meteorological observation data, radar meteorological observation data, ground meteorological observation data, aircraft meteorological observation data and sounding meteorological observation data.
[0012] Furthermore, obtaining a multi-source meteorological observation data set for the target area also includes:
[0013] Numerical forecast assimilation is performed on multi-source meteorological observation datasets in the target area.
[0014] Furthermore, numerical forecast assimilation is performed on the multi-source meteorological observation data set of the target area, including:
[0015] Assimilate and process the initial fields of the multi-source meteorological observation data set to obtain the multi-source meteorological observation data set after data processing;
[0016] Use the WRF regional model for forecasting and obtain numerical forecast data;
[0017] The numerical forecast data is post-corrected based on the multi-source meteorological observation data set after data processing through machine learning algorithms.
[0018] Furthermore, the initial fields of multi-source meteorological observation datasets are assimilated and processed, including:
[0019] The following quality control is performed on the multi-source meteorological observation dataset, and the meteorological observation dataset after quality control is obtained:
[0020] Repeatability check, validity check, continuity check, extreme value check, position consistency check and spatiotemporal consistency check.
[0021] Furthermore, a repeatability check is used to determine whether there are multiple sets of similar detection data within the time, longitude, latitude and altitude limits, and to remove duplicate data.
[0022] Furthermore, validity check is used to determine the current message availability based on the observed data.
[0023] Furthermore, the continuity check is used to check whether the observed data caused by the instrument failure remains unchanged or changes less than a threshold. When the difference between adjacent message data is less than a given criterion for three or more consecutive times, the instrument is judged to have failed the continuity check.
[0024] Furthermore, the extreme value check is used to detect whether the observation values of the ground regional stations of the observed elements are within the extreme value range under the corresponding latitude and altitude conditions.
[0025] Specifically, the threshold range of meteorological observations usually depends on a variety of factors, including geographical location, season, terrain, altitude, etc.;
[0026] For example, temperature: Temperatures can range widely, depending on geographic location and season. For example, in the Arctic, winter temperatures can drop to tens of degrees below zero, while in tropical regions, summer temperatures can reach over 40 degrees Celsius.
[0027] Humidity: Humidity usually ranges from 0% (completely dry) to 100% (saturated). However, in actual observations, humidity rarely reaches these two extreme values due to various factors.
[0028] Air pressure: Air pressure is usually expressed in hectopascals (hPa). Under standard atmospheric conditions, the air pressure at sea level is about 1013.25 hPa. As the altitude increases, the air pressure gradually decreases.
[0029] Wind speed: Wind speed is usually expressed in meters per second (m / s) or kilometers per hour (km / h). In calm weather conditions, wind speeds can be very low, close to zero, or in extreme weather conditions such as storms or hurricanes, wind speeds can reach hundreds of kilometers per hour.
[0030] Precipitation: The range of precipitation depends on the geographical location and season. In arid areas, precipitation may be very low or even zero for many years; while in tropical rainforest areas, precipitation may be very abundant, reaching tens of millimeters or more per day.
[0031] Furthermore, the location consistency check is used to eliminate data quality problems caused by time collection and positioning errors. The previous and subsequent messages are used to determine whether the current message remains consistent in time, longitude and latitude, and altitude changes.
[0032] Furthermore, the time-space consistency check is used to detect whether the message remains consistent during the process of changing with time and space.
[0033] Specifically, time consistency check:
[0034] Definition: Temporal consistency means that the state or information of the same entity or resource should remain consistent at multiple different time points or time periods.
[0035] Application: In message transmission, time consistency check can ensure that the timestamp and time representation of the message are consistent between different nodes. For example, in a joint test system, for nodes m and n, at the same time T, the state or information of entity resource A should be consistent between m and n.
[0036] Detection method: You can ensure time consistency by comparing the timestamps of messages received on different nodes, or by using a unified time standard (such as UTC time). If you find that the timestamps differ too much or the time representation is inconsistent, it may mean that there is a problem with the message during transmission, thus affecting its reliability.
[0037] Spatial consistency check:
[0038] Definition: Spatial consistency means that the state or information of the same entity or resource should remain consistent at multiple different spatial locations or nodes.
[0039] Application: In message transmission, spatial consistency check can ensure that the content, structure and other information of the message remain consistent when it is transmitted on different network nodes. For example, in a joint test system, the four-dimensional space-time coordinates of entity resource A between nodes m and n should be the same.
[0040] Detection method: The consistency of the space can be ensured by comparing the content, structure and other information of the messages received at different nodes. If the message is found to have been changed or damaged during transmission, it may mean that there is a problem in the message transmission process, thus affecting its reliability.
[0041] Furthermore, the numerical forecast data is post-corrected based on the multi-source meteorological observation data set after data processing through machine learning algorithms, including:
[0042] Perform feature extraction, feature preprocessing, feature classification construction and feature combination on numerical forecast data;
[0043] The numerical forecast data are post-corrected through machine learning algorithms based on the feature combination of numerical forecast data and the multi-source meteorological observation data set after data processing.
[0044] Furthermore, the GRU-CNN hydro-wind-solar power generation joint prediction model includes: a heterogeneous information encoding layer, a spatiotemporal feature fusion layer, and a power decoding prediction layer connected in sequence;
[0045] The heterogeneous information encoding layer is used to encode heterogeneous information in the multi-source meteorological observation data and historical power generation data of the hydropower station, wind power station and photovoltaic power station, and generate a spatial matrix of hydropower, a spatial matrix of wind power and a spatial matrix of photovoltaic power;
[0046] The spatiotemporal feature fusion layer is used to generate the spatiotemporal feature matrix of water energy, the spatiotemporal feature matrix of wind energy and the spatiotemporal feature matrix of light energy based on the spatial matrix of water energy, the spatial matrix of wind energy and the spatial matrix of light energy;
[0047] The power decoding prediction layer is based on the spatiotemporal feature matrix of water energy, the spatiotemporal feature matrix of wind energy, and the spatiotemporal feature matrix of light energy. It is used to perform upsampling operations through deconvolution, and then decodes the deconvolution features based on GRU, and outputs the data through the fully connected layer.
[0048] Specifically, the heterogeneous information encoding layer includes wind and solar historical information encoding, hydropower historical information encoding and load historical information encoding; the spatiotemporal feature fusion layer includes GRU temporal feature extraction, attention mechanism and CNN spatiotemporal fusion; the power decoding prediction layer is used for GRU decoding and DENSE sequence matching.
[0049] Furthermore, the heterogeneous information includes historical hydroelectric power, historical water level, historical wind power, historical wind speed, historical photovoltaic power, historical irradiance and load data of the hydropower, wind and solar power stations in the target area.
[0050] Furthermore, the heterogeneous information encoding layer includes a wind power related data encoder, a hydropower related data encoder, a photovoltaic related data encoder and a load data encoder;
[0051] The input of the wind power related data encoder includes the wind power generation power at time tn, the observed wind speed of the hydro-wind-solar power station at time tn, and the forecast meteorological elements of the hydro-wind-solar power station from time t to time tn;
[0052] The input of the hydropower related data encoder includes the hydropower generation power at time tn, the observed runoff of the hydro-wind-solar power station at time tn, and the forecast meteorological elements of the hydro-wind-solar power station from time t to time tn;
[0053] The input of the photovoltaic related data encoder includes the photovoltaic power generation at time tn, the observed irradiance of the water, wind and solar power station at time tn, and the forecast meteorological elements of the water, wind and solar power station from time t to time tn;
[0054] The input of the load data encoder includes the load at time tn and the load at time tnm, where t, n, and m are natural numbers.
[0055] Furthermore, a spatial matrix of water energy, a spatial matrix of wind energy, and a spatial matrix of light energy are generated, including:
[0056] Based on heterogeneous information, the spatial characteristics of water energy, wind energy and light energy are obtained through convolution extraction;
[0057] According to the spatial characteristics of water energy, wind energy and light energy, the corresponding spatial matrix of water energy, the spatial matrix of wind energy and the spatial matrix of light energy are generated.
[0058] Specifically, the convolution layer extracts features from the input data through a convolution operation. The convolution operation is a mathematical operation that generates new spatial features of water energy, wind energy, and light energy by sliding a convolution kernel (also called a filter) on the input data and calculating the dot product between the convolution kernel and the local area of the input data.
[0059] The rows of the spatial matrix represent the spatial characteristics of the energy corresponding to a location.
[0060] Furthermore, based on the spatial matrix of water energy, the spatial matrix of wind energy and the spatial matrix of light energy, the spatiotemporal characteristic matrix of water energy, the spatiotemporal characteristic matrix of wind energy and the spatiotemporal characteristic matrix of light energy are generated, including:
[0061] Based on heterogeneous information, the time matrix of water energy, the time matrix of wind energy and the time matrix of light energy are extracted through GRU;
[0062] Based on the time matrix of water energy, the time matrix of wind energy, the time matrix of light energy, the space matrix of water energy, the space matrix of wind energy and the space matrix of light energy, the spatiotemporal characteristic matrix of water energy, the spatiotemporal characteristic matrix of wind energy and the spatiotemporal characteristic matrix of light energy are generated.
[0063] Specifically, the model automatically extracts GRU (Gated Recurrent Unit), a recurrent neural network (RNN) structure that is particularly suitable for processing sequence data. In terms of feature extraction, GRU can learn and capture long-term dependencies and periodic information in sequence data through its internal mechanism, thereby extracting features that are useful for specific tasks. The time matrix of water energy, wind energy, and light energy will be input into GRU to obtain.
[0064] Furthermore, the spatiotemporal feature fusion layer adopts an attention mechanism to set a weight for all hidden states in the encoder, then weighted sum each hidden state according to the set weight, and then input all hidden states after weighted summation into the decoder.
[0065] In the second aspect, a combined prediction device for water, wind and solar power generation based on a hybrid neural network includes: a data collection unit, a model building unit and a prediction unit;
[0066] A data collection unit, used to obtain a multi-source meteorological observation data set for a target area;
[0067] The data collection unit is also used to obtain a historical power generation data set of the hydro-wind-solar power stations in the target area;
[0068] A model building unit is used to build and train a GRU-CNN hydro-wind-solar power generation joint prediction model based on a multi-source meteorological observation data set and a historical power generation data set;
[0069] The prediction unit is used to obtain the meteorological forecast data of the target area in the time period to be predicted in real time, and transmit it to the trained GRU-CNN water-wind-solar power joint prediction model to predict the hydropower, wind power and photovoltaic power of the water-wind-solar power station in the target area in the time period to be predicted.
[0070] In a third aspect, an electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus;
[0071] a memory storing a computer program;
[0072] The processor is used to implement the above-mentioned water-wind-solar power generation joint prediction method based on hybrid neural network when executing the computer program stored in the memory.
[0073] In a fourth aspect, a computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned hybrid neural network-based joint prediction method for water-wind-solar power generation.
[0074] The present disclosure has at least the following beneficial effects:
[0075] The present invention improves the initial field through data assimilation to improve the numerical forecast effect, integrates the spatiotemporal correlation and load curve of water, wind and solar power generation, constructs an improved GRU-CNN water, wind and solar power generation joint prediction model, realizes the joint prediction of water, wind and solar power generation, and improves the accuracy of water, wind and solar power generation prediction.
[0076] Other features and advantages of the present disclosure will be described in the following description, and partly become apparent from the description, or be understood by implementing the present disclosure. The purpose and other advantages of the present disclosure can be realized and obtained by the structures indicated in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0077] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0078] Figure 1 This is a flowchart of the prediction method according to an embodiment of the present disclosure;
[0079] Figure 2 This is a schematic diagram of the process of numerical forecast assimilation;
[0080] Figure 3 It is a structural diagram of the joint prediction model;
[0081] Figure 4 This is a schematic diagram of the structure of the prediction device according to the embodiment of the present disclosure;
[0082] Figure 5 A schematic diagram of the electronic device structure. DETAILED DESCRIPTION
[0083] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.
[0084] like Figure 1 As shown, the combined prediction method of hydropower, wind power and solar power generation based on hybrid neural network includes:
[0085] S101, obtaining a multi-source meteorological observation dataset of a target area;
[0086] S102, obtaining a historical power generation data set of hydropower, wind and solar power stations in a target area;
[0087] S103, based on the multi-source meteorological observation data set and the historical power generation data set, a GRU-CNN hydro-wind-solar power generation joint prediction model is established and trained;
[0088] S104, real-time acquisition of weather forecast data for the target area in the time period to be predicted, and transmission to the trained GRU-CNN hydro-wind-solar power joint prediction model to predict the hydropower, wind power and photovoltaic power of the hydro-wind-solar power stations in the target area in the time period to be predicted.
[0089] The specific implementation is as follows:
[0090] S101, obtaining a multi-source meteorological observation dataset of a target area.
[0091] In some embodiments, the multi-source meteorological observation data set of the target area includes at least hydro-wind-solar power station observation data, satellite meteorological observation data, radar meteorological observation data, ground meteorological observation data, aircraft meteorological observation data and sounding meteorological observation data.
[0092] like Figure 2 As shown in Figure 1, numerical forecast assimilation is performed on the multi-source meteorological observation datasets in the target area.
[0093] S102, obtaining a historical power generation data set of hydropower, wind and solar power stations in a target area.
[0094] In some embodiments, the historical power generation data set of the hydro-wind-solar power station may include historical hydroelectric power generation, historical wind power generation, historical photovoltaic power generation, historical total power generation and load.
[0095] S103, establish an improved GRU-CNN joint prediction model for hydropower, wind power and solar power generation.
[0096] Figure 3 is a schematic diagram of the structure of the improved GRU-CNN water-wind-solar power generation joint prediction model shown in some embodiments of this specification, such as Figure 3 In some embodiments, the improved GRU-CNN hydro-wind-solar power generation joint prediction model at least includes:
[0097] The heterogeneous information encoding layer is used to encode the heterogeneous information in the multi-source meteorological observation data set and the historical power generation data set assimilated by the numerical forecast of the hydropower station, the wind power station and the light power station, and generate a spatial matrix of water energy, a spatial matrix of wind energy and a spatial matrix of light energy, wherein the heterogeneous information includes the historical hydropower generation power, historical water level, historical wind power generation power, historical wind speed, historical photovoltaic power generation power, historical irradiance and load data of the hydropower station in the target area, and the heterogeneous information encoding layer may include a wind power related data encoder, a hydropower related data encoder, a photovoltaic related data encoder and a load data encoder, wherein the input of the wind power related data encoder may include The wind power generation power at time tn, the observed wind speed of the hydro-wind-solar power station at time tn, and the forecast meteorological elements of the hydro-wind-solar power station from time t to time tn. The input of the hydropower related data encoder may include the hydropower generation power at time tn, the observed runoff of the hydro-wind-solar power station at time tn, and the forecast meteorological elements of the hydro-wind-solar power station from time t to time tn. The input of the photovoltaic related data encoder may include the photovoltaic power generation power at time tn, the observed irradiance of the hydro-wind-solar power station at time tn, and the forecast meteorological elements of the hydro-wind-solar power station from time t to time tn. The input of the load data encoder may include the load at time tn and the load at time tnm.
[0098] The spatiotemporal feature fusion layer is used to generate the spatiotemporal feature matrix of water energy, the spatiotemporal feature matrix of wind energy and the spatiotemporal feature matrix of light energy based on the spatial matrix of water energy, the spatial matrix of wind energy and the spatial matrix of light energy;
[0099] The power decoding prediction layer is used to perform upsampling operations through deconvolution to unify the data dimension, decode the deconvolution features based on GRU, and output the data through the fully connected layer so that the output prediction time meets business needs, such as ultra-short-term, short-term, medium-term and long-term business needs.
[0100] The process from the input layer to the hidden layer can be understood as an encoding process, that is, using a set of data of a lower dimension to characterize the original data. Due to the large differences in different forms of energy such as water, wind, and light, encoding this heterogeneous information can not only reduce the dimension of the data and compress the information, but also better extract representative features and their relationships, laying a solid foundation for subsequent neural network simulation.
[0101] In some embodiments, generating a spatial matrix of water energy, a spatial matrix of wind energy, and a spatial matrix of light energy includes:
[0102] The spatial characteristics of different water, wind and solar power stations for each energy source are extracted through convolution;
[0103] Based on the spatial characteristics of different water, wind and solar power stations for each energy source, a spatial matrix of water energy, a spatial matrix of wind energy and a spatial matrix of solar energy are generated.
[0104] Specifically, the wind power related data encoder, hydropower related data encoder, photovoltaic related data encoder and load data encoder use CNN (Convolutional Neural Networks, CNN) to convolutionally extract the spatial characteristics of different stations for each energy source, thereby forming feature relationship extraction at the spatial level of each station, and obtaining the spatial matrices of the three types of energy: water, wind and photovoltaic.
[0105] In some embodiments, the spatiotemporal feature fusion layer generates a spatiotemporal feature matrix of water energy, a spatiotemporal feature matrix of wind energy, and a spatiotemporal feature matrix of light energy based on the spatial matrix of water energy, the spatial matrix of wind energy, and the spatial matrix of light energy:
[0106] Extract the time matrix of water energy, the time matrix of wind energy and the time matrix of light energy through the GRU layer;
[0107] Based on the time matrix of water energy, the time matrix of wind energy, the time matrix of light energy, the space matrix of water energy, the space matrix of wind energy and the space matrix of light energy, the spatiotemporal characteristic matrix of water energy, the spatiotemporal characteristic matrix of wind energy and the spatiotemporal characteristic matrix of light energy are generated.
[0108] like Figure 3 As shown in the figure, the attention mechanism is introduced into the spatiotemporal feature fusion layer. Because in the time series problem, the importance of each time to the subsequent prediction is different. At the same time, in the joint total power prediction, the contribution of different stations to the total power is also different. Therefore, the attention mechanism layer is introduced. The attention mechanism layer passes all the hidden state information in the encoder to the decoder. If there are N hidden states in the encoder, the information of these N hidden states needs to be given to the decoder. Before all the information is passed to the decoder, a weight needs to be set for each of the N hidden states, and then each hidden state is weighted and summed according to the set weight, and then all the hidden states after the weighted sum are input into the decoder. In this way, the attention to important time and important station information is improved through the weight mechanism, thereby improving the accuracy of the joint prediction.
[0109] Through the above CNN+GRU+Attention operations, the fusion of spatiotemporal features among the three energy sources is completed. On this basis, considering the impact of the power load on the so-called "source-load interaction" on the power generation side in the new power system, the characteristics of the regional power load are further superimposed in the spatiotemporal fusion matrix. And continue to further extract the two-dimensional matrix features through the CNN layer. The purpose of this step is to realize the secondary fusion of the spatiotemporal features on the power generation side, and also hope to capture the correlation characteristics between the source and the load.
[0110] Recurrent Neural Network (RNN) is a commonly used deep learning algorithm in the field of time series prediction. However, its network structure makes it easy to have gradient vanishing and gradient exploding problems in long-term memory and back propagation. The gated recurrent unit (GRU) improves this problem through the "gate" mechanism. The network structure is similar to the long short-term memory (LSTM) neural network. The difference is that GRU combines the forget gate and input gate in LSTM into a single update gate, and mixes the cell state and hidden state, making the structure simpler and more efficient, and has good applicability for capturing time series characteristics.
[0111] GRU processes the hidden state h of the previous moment through two multiplication gates: reset gate and update gate t-1 and the external input information x at the current moment t , thereby updating the hidden state h at the current moment t To implement model training, the calculation rules are as follows:
[0112] z t=σ(W(z)xt+U(z)h t一1 )
[0113] r t =σ(W(r)x t +U(r)h t一1 )
[0114] h't=tanh(Wx t +r t ⊙Uh t一1 )
[0115]
[0116] Among them, W (z) , W (r) , W, U (z) , U is the response input weight matrix; h't is the current memory content; ⊙ is the Hadamard operation; σ is the Sigmoid activation function,
[0117] The core idea of the Attention Model (AM) is to introduce attention weights on the input sequence to evaluate the importance of different input information, which helps the model to pay more attention to the input information that is more closely related to the expected output and enhance the accuracy of the model.
[0118] The introduction of the attention mechanism can further improve the GRU network's ability to extract time series features. However, from the results of the spatiotemporal correlation characteristics analysis, it can be seen that the output of wind, solar and hydropower stations in the region is not only related to the historical output of each station, but also has a strong correlation with the historical output of other similar stations in the region. There are significant differences in the output between heterogeneous energy sources. Ignoring the spatial characteristics of the output between stations and the differences in heterogeneous energy during data processing and modeling may limit the further improvement of prediction accuracy. Therefore, the proposed improved GRU-CNN model, as shown in the figure below, takes into account the differences in the data dimensions of the input and output modules, and in order to better explore the mapping relationship between the input sequence and the output sequence.
[0119] S103 also includes training an improved GRU-CNN hydro-wind-solar power generation joint prediction model using a multi-source meteorological observation data set and a historical power generation data set assimilated by numerical forecasts of hydro-wind-solar power stations.
[0120] In specific implementation, the training process is as follows:
[0121] Model training:
[0122] The model is trained using the training set. During the training process, the model learns how to extract features from the input data and make predictions or decisions.
[0123] Training typically involves multiple epochs, where the model goes through the entire training set once in each epoch.
[0124] During the training process, it is also necessary to select a suitable optimization algorithm (such as gradient descent, stochastic gradient descent, etc.) to update the parameters of the model in order to minimize the loss function.
[0125] Hyperparameter Tuning:
[0126] Hyperparameters are parameters that need to be set before model training, such as learning rate, batch size, number of iterations, etc.
[0127] Use the validation set to tune the model’s hyperparameters to find the best model configuration [3].
[0128] Model Evaluation:
[0129] Use the test set to evaluate the performance of the model. Evaluation metrics can include accuracy, recall, precision, F1 score, etc., depending on the nature of the problem.
[0130] The generalization ability of the model is evaluated by comparing the performance of the model on the test set with the actual performance.
[0131] Model optimization:
[0132] Based on the performance of the model on the validation set and test set, the model can be optimized, such as changing the model structure, adjusting hyperparameters, using more complex features, etc.
[0133] The loss function is a conventional loss function, such as mean square error loss, cross entropy loss, etc.
[0134] S104, obtaining weather forecast data for the target area in the time period to be predicted.
[0135] By improving the GRU-CNN joint prediction model of hydropower, wind power and photovoltaic power generation, based on the meteorological forecast data of the target area in the prediction period, the hydropower, wind power and photovoltaic power generation of the water, wind and photovoltaic power stations in the target area are predicted in the prediction period.
[0136] Figure 2 is a schematic diagram of a process for assimilating numerical forecasts of multi-source meteorological observation data sets according to some embodiments of this specification, such as Figure 2 As shown, in some embodiments, performing numerical forecast assimilation on a multi-source meteorological observation data set of a target area includes the following steps:
[0137] S201, assimilate and process the initial field of the multi-source meteorological observation data set to obtain the multi-source meteorological observation data set after data processing.
[0138] The multi-source meteorological observation data set is tested for repeatability, validity, continuity, and spatiotemporal consistency, and the observation data after quality control and the corresponding quality labels are provided, including:
[0139] Repeatability check: mainly determines whether there are multiple sets of similar detection data when the time, longitude, latitude and altitude are relatively close.
[0140] Validity check: Its purpose is to determine the current message availability based on observed data.
[0141] Continuity check: Check that the observed data caused by instrument failure remains unchanged or changes very little. The difference between adjacent message data is required to be greater than the given criterion. If the difference between adjacent message data is less than the given criterion for 3 or more consecutive times, the instrument is considered to have failed the continuity check.
[0142] Extreme value check: The observation values of the ground regional stations of the observation elements should be within a reasonable range under the corresponding latitude and altitude conditions.
[0143] Position consistency check: It is mainly used to eliminate data quality problems caused by time collection and positioning errors. The previous and next messages can be used to determine whether the current message is consistent in time, longitude and latitude, and altitude changes.
[0144] Time and space consistency check: Time and space consistency check can explore the reliability of messages as they change over time and space.
[0145] Suspicious data check: Suspicious data refers to message data that is not applicable to the above quality control methods. For example, there are only one or two messages on the same flight, or there are too many error messages on the same flight, which makes it impossible to perform quality control normally. Such data is marked with a suspicious identifier to facilitate further processing by manual or other means.
[0146] Based on the position of the simulated area station in the model calculated by the model observation operator, the underlying surface information is collected and compared with the actual situation, and the typical differences are marked, including:
[0147] The position of the simulated area station in the model calculated according to the model observation operator;
[0148] Collect underlying surface information and compare it with the actual situation;
[0149] Those with typical differences (e.g., differences between water and land) are marked.
[0150] For all observation stations, especially those in harsh environmental areas, the observation errors caused by equipment being exposed to sunlight and rain, sand and dust erosion, and erroneous observation data caused by obstacles within the observation field of view are detected, and the observation data with obvious no meteorological significance is detected and eliminated. The continuity of the data, background consistency, etc. are checked, and the deviation of the observation data is corrected. Specifically, it includes:
[0151] Station extreme value check: Station extreme value refers to the maximum and minimum values of a certain element of the detected station that have appeared in history.
[0152] Temporal consistency check: The purpose of the temporal consistency check is to examine the temporal rate of change of the observed information or observed elements and to identify undesirable sudden changes.
[0153] Spatial consistency check: Spatial consistency check is a method of quality control of observation data of a station by fully utilizing the relationship between the observation data of the inspected station and multiple neighboring stations at the same time.
[0154] Background field consistency check: compare the difference between the observed data and the background field (referred to as observation residual) with its judgment criteria.
[0155] Satellite data needs to be converted into PrepBUFR or BUFR (Binary Universal Form for the Representation of meteorological data) format. In BUFR format, in addition to the brightness temperature of each satellite observation channel, data grid longitude and latitude, solar altitude angle, solar azimuth, satellite azimuth, brightness temperature variance, clear sky ratio and other data are also required. Among them, brightness temperature variance and clear sky ratio need to be calculated based on the number of diluted grid points. The FY-2F observation data can be converted into BUFR format through the BUFR program attached to the GSI module.
[0156] The resolution of the original data of the FY-2F satellite is as low as 5km (infrared channel), which is higher than the resolution of GSI assimilation (0.06°). Through sparse processing, the error correlation between adjacent observations can be reduced. Referring to the BUFR data information of the GOES satellite, 9*9 pixels are taken as an observation unit, and the brightness temperature, longitude and latitude, solar altitude and other information are taken as the value of the center point, and the brightness temperature variance and clear sky coverage of the observation unit are calculated. Therefore, the resolution of the BUFR data becomes 40-60km.
[0157] Under clear sky conditions, the infrared band only considers the emission and absorption of the atmosphere and the surface, and does not consider scattering and other effects. The radiation transfer equation can be written as:
[0158]
[0159] Where I is the radiation intensity received at the top of the atmosphere, I s is the surface emission, τ represents the optical thickness from the surface to the atmosphere in the vertical direction, is the vertical optical thickness of the entire atmosphere, μ is the cosine of the zenith angle, and B is the radiation intensity emitted by the atmosphere. The first term of the radiation transfer equation is the part of the radiation emitted by the surface that is absorbed by the entire atmosphere and reaches the top of the atmosphere, and the second term is the contribution of the radiation emitted by the atmosphere itself. Among them, the atmospheric emission part, after step-by-step integration, the entire equation can be rewritten as:
[0160]
[0161] in, The change of transmittance with height is called weight function. It can be regarded as the weight of each layer of the atmosphere in the radiation contribution of the entire atmosphere. To some extent, the height of the peak of the weight function corresponds to the height that the satellite can "see". Most of the contribution of satellite observation information comes from the height where the weight function is larger. The distribution of the weight function is not only related to the band, but also to the gas composition, content and atmospheric state profile in the atmosphere.
[0162] Since infrared channel radiation information is very sensitive to clouds, cloud detection is an important part of the quality control of infrared channel radiation data. The FY-2F nominal disk data contains cloud classification information calculated by the DPC system of the Satellite Meteorological Center during non-encrypted observations. In order to remove cloud-contaminated data as much as possible and avoid mixing pattern information with observation field information, three independent cloud detection algorithms that do not rely on pattern calculations are used to perform cloud detection separately by giving empirical thresholds. The detection method is as follows: When the i grid point satisfies and That is, the pixel is judged to be clouded. Among them, γ represents the vertical decrease rate of atmospheric temperature, and its value is 7K / 1000m. is the standard deviation of terrain height, is the standard deviation of IR1 brightness temperature (3*3 pixels centered at i). Represents the brightness temperature of the i-th pixel of IR1 and IR4 channels respectively, is the maximum brightness temperature of the IR1 channel in the 3*3 pixel point. The empirical thresholds for land and ocean are different. For the IR1IR4T cloud detection method, the empirical threshold ε3 is also different during the day and at night because the IR4 near-infrared is easily affected by solar radiation during the day. It is defined that when the solar altitude angle at a certain point is greater than 0, the empirical threshold is ε 31 , and when the solar altitude angle is less than 0, the empirical threshold is ε 32 , empirical thresholds ε1, ε2, ε 31 , ε 32 As shown in Table 1.
[0163] Table 1
[0164]
[0165] Radar assimilation is also an important part of data assimilation. It checks and controls the quality problems of radar data caused by terrain obstruction, environmental noise, etc., and processes the clutter caused by the ground, data shortages caused by other reasons, clear sky echoes and other quality problems through fuzzy logic and other methods, and provides radar data with higher quality after quality control. Interpolation and extension of data shortage: It can be processed through the continuity characteristics of meteorological data, and interpolation is performed through data obtained from different time periods of the same unit point. Processing of abnormal point echoes: The echo of Doppler radar has a certain threshold. Under normal circumstances, the echo will not exceed this threshold. However, the data of abnormal points exceeds the echo threshold in many cases. Therefore, this can be used as a basis for determining abnormal points. For the processing of abnormal points, the two-dimensional median filtering method is often used. Eliminate clear sky echoes: Sometimes Doppler radar will also generate echoes when detecting cloudless areas in the atmosphere. Processing of terrain obstruction: Most of the echoes generated by terrain obstruction are fixed and unchanged. The terrain echoes are identified and processed to eliminate the echoes generated by terrain obstruction in the radar data. The elevation, azimuth and slant range of the grid point in the Cartesian coordinate system are calculated by the longitude, latitude and altitude of the grid point in the spherical coordinate system. Then, according to the calculated elevation, azimuth and slant range in the radar spherical coordinate system, radial interpolation and azimuth interpolation are performed by the nearest neighbor method, and radial interpolation is performed by linear interpolation in the vertical direction to assign values to the grid point to obtain the analysis value on the grid point. Construct three-dimensional gridded reflectivity data to build data for assimilation system analysis. Carry out rapid cycle analysis based on numerical forecast results. Based on the Bayesian prior estimation model and the Gaussian probability distribution model, establish the GSI assimilation module to assimilate the collected conventional / unconventional observation data. Improve the quality of the analysis field through variable scale iteration technology, direct reflectivity assimilation technology, weak wind field constraint technology, cloud analysis false observation and particle information constraint assimilation technology, and convective scale control variable technology. Provide a better initial field for the next forecast.
[0166] The forecast samples of background error covariance are calculated by NMC method. The forecast samples are obtained by different start-up times.
[0167] By comparing the effects of different control variables on assimilation, we select the best one and build a suitable background error covariance matrix that meets the scale characteristics of the simulated regional climatology. At the same time, we conduct variable analysis on the final statistical background error covariance matrix to find out its height variation in different models.
[0168] Variable conversion: Establish a computational model to convert model prediction variables into control variables.
[0169] Latitude averaging: The differences in latitude of the control variables are averaged through filtering and other methods.
[0170] Decorrelation between variables: Through dynamics and statistical balance, the correlation between background errors between variables is removed and converted into uncorrelated control variables.
[0171] Vertical transformation: Eliminate the vertical correlation within the control variable based on the empirical orthogonal function (EOF) decomposition, and calculate the background error eigenvalue and eigenvector.
[0172] Horizontal transformation: Recursive filtering is used to calculate the horizontal characteristic length scale of the control variable to show the horizontal characteristics of the background error field.
[0173] After strict quality control, the collected conventional / non-conventional observation data need to be converted into a specific format and assimilated into the GSI assimilation system.
[0174] Select conventional / unconventional observation data that have undergone strict quality control during the forecast or simulation period, and select observation data in the 1.5 h time domain before and after the assimilation time.
[0175] The selected observation data may be stored in various formats, such as ASCII, BUFR, and MADIS. Write a program to process all observation data into LITTLE_R format.
[0176] Run the OBSPROC data preprocessing program in the GSI assimilation system to convert the observation data in LITTLE_R format into PREBUFR format for assimilation.
[0177] For different types of observation data, under the consideration of different weather backgrounds, observation operators are established that can reasonably map the relationship between the physical quantities of the atmospheric state and the observed physical quantities. Taking the establishment of near-surface observation operators for ground observation data as an example, ground observation data are greatly affected by topography and landforms, and there is a certain height difference between the general model terrain and the actual observation station terrain. How to effectively solve the problem of height difference between model terrain and observation station terrain has become a basic problem in ground data assimilation. Then considering establishing ground observation operators based on the dynamic and thermal constraints of the near-surface boundary layer is an effective way to solve the above problems.
[0178] The boundary layer dynamic and thermal processes are incorporated, and the height difference between the model terrain and the observation station terrain is considered to establish a new regional near-surface observation operator.
[0179] The established near-surface observation operator is rigorously checked and tested, including the accuracy check of the tangent mode and adjoint mode of the model (adjoint check and gradient check), and the test results are verified to ensure that the new near-surface observation operator can better describe the role of near-surface observations in assimilation.
[0180] Different height differences between model terrain and observed terrain will result in different forecasting effects, so it is necessary to determine the optimal critical height difference.
[0181] Subsequent assimilation experiments are carried out through the newly established observation operator to ensure that the three-dimensional variational analysis system under the new ground observation operator can correctly reflect the interaction relationship between the wind field and other variables.
[0182] For a certain type of observation data and a certain type of forecast object, an integer multiple of the model grid horizontal radius dx is selected as the horizontal influence radius to conduct an assimilation experiment;
[0183] Increase or decrease the horizontal influence radius and conduct assimilation experiments.
[0184] Assimilate several times, compare the analysis fields obtained after assimilation under different horizontal influence radii, and compare them with the observed fields and background fields before assimilation. Evaluate the assimilation effects of different horizontal influence radii through incremental or deviation analysis, and select the experiment with the best assimilation effect by comparison to determine the selection range of the horizontal influence radius.
[0185] The vertical impact scale is changed to conduct assimilation experiments, and the assimilation effects are compared to determine the selection range of the vertical impact radius.
[0186] Based on the Bayesian prior estimation model and Gaussian probability distribution model, a 3DVAR assimilation module is established to assimilate the collected conventional / unconventional observation data.
[0187] The meteorological initial field processed by WPS or the meteorological field predicted by WRF is used as the background field for three-dimensional variational assimilation;
[0188] Select a statistically reasonable background error covariance matrix, link the preprocessed observation data, background data, and background error covariance with specific names, run the GSI assimilation system, and obtain the assimilated analysis field at the corresponding moment as the background field for continued forecasting at the next moment.
[0189] According to the set assimilation time window, multiple cycles of assimilation are performed to continuously improve the forecast effect.
[0190] In the assimilation analysis process, for the same kind of observation data, for different forecast objects, in different iterative cycles, the influencing scale factor is changed to fully extract the effective information of the observation data at different scales.
[0191] For a certain type of observation data or a certain forecast object, at the initial analysis moment, a value within the previously determined horizontal or vertical influence radius is selected as the horizontal or vertical influence scale.
[0192] With the development of weather systems, the impact scales of observational data at different analysis times may not be the same. The analysis field obtained after the first assimilation is used as the background field for the next forecast. The forecast is continued until the next assimilation time. The same horizontal or vertical impact radius as the last assimilation is selected for assimilation experiments. At the same time, the impact radius is increased or reduced in a cyclic form to conduct assimilation experiments. The assimilation effects under different impact scales are compared, and the impact scale corresponding to the group of experiments with the analysis field closest to the observation is selected as the horizontal or vertical impact scale of assimilation at that moment.
[0193] For each subsequent analysis moment, the optimal horizontal or vertical influence radius is found for assimilation at that moment, so as to fully extract the information of the observation data at different scales.
[0194] For variable scale assimilation analysis, multiple iterations are required. Through multi-threading and parallel computing technology, the calculation efficiency of the algorithm is improved, and the analysis efficiency of large amounts of data is accelerated, so that the system can meet the needs of short-term rapid analysis.
[0195] The literature was consulted to find a mathematical model for structural optimization design based on system reliability.
[0196] The optimization criterion method used in the past to solve the optimization model is improved.
[0197] Through the solution process of the example, the good effect of the improved optimization criterion method is verified, ensuring that the iterative convergence speed is accelerated.
[0198] Directly use a multi-level observation operator containing multiple air particle phases without inverting the reflectivity to avoid errors introduced in the inversion process and the problem that the warm cloud parameterization scheme cannot correctly reflect the ice phase process when used as an observation operator. Use background temperature as the basis for particle classification so that the assimilated hydrometeors can more accurately reflect actual observations.
[0199] Thermodynamic variables updated after assimilating observation data include water vapor mixing ratio, perturbation pressure, perturbation potential temperature, perturbation potential height, etc., which do not include the rain mixing ratio which is closely related to the radar basic reflectivity factor. Therefore, in order to enable the echo data of Doppler weather radar to be directly assimilated, the analysis variable of rain mixing ratio is introduced into the calculation in the assimilation framework.
[0200] The echo intensity and rain mixing ratio are linked based on the relationship between radar reflectivity Z (dBZ) and rain mixing ratio qr.
[0201] The rain water mixing ratio is linked to the total liquid water mixing ratio by introducing the warm rain scheme as a constraint in the GSI assimilation system.
[0202] Based on the above relationship, direct assimilation of radar echo intensity is achieved.
[0203] In the GSI objective function, a divergence constraint term is added to analyze the flow field. This term will serve as a weak constraint term to provide a balance relationship constraint for the wind field analysis results and improve the wind field analysis results.
[0204] The principle of GSI assimilation is to solve the analysis field corresponding to the minimum value of the cost function composed of background field, observation field, analysis field, observation operator, background error covariance, observation error covariance, etc. The assimilation test is carried out using the original cost function without any modification.
[0205] A small term related to the divergence wind is added to the original objective function to perform assimilation experiments with the balance constraint of the wind field.
[0206] Analyze and compare the assimilation results of the original scheme and after changing the objective function to ensure that the added constraints are beneficial to the analysis results of the wind farm. On this basis, continuously improve the constraints until the analysis results of the wind farm reach the best.
[0207] Construct a cloud analysis module based on cloud physics initialization. Cloud physics initialization is to extract information about clouds (mainly cumulus and stratus) from ground truth, sounding, satellite and radar observation data, and improve the three-dimensional structure of clouds, so as to adjust the atmospheric humidity, temperature distribution, vertical and horizontal wind fields, so that the initial state of the model is closer to the real state of the atmosphere. Through an empirical function, the relative humidity in the initial field is converted into the cloud coverage on the corresponding grid to obtain the three-dimensional cloud coverage background field. The cloud base height and cloud coverage information in the mesoscale observation data are added to obtain a comprehensive horizontal cloud field analysis.
[0208] The cloud top is obtained from satellite infrared data, while radar reflectivity is mainly used to obtain information about clouds in the middle troposphere. Visible light satellite cloud images are used to estimate the total cloud cover and calibrate overestimation of clouds. Infrared satellite cloud image data are added to obtain cloud top information. The three-dimensional cloud coverage obtained from the first two steps is added to the initial temperature field to achieve correction of cloud analysis, so that the cloud top brightness temperature of the grid analysis is consistent with the observation. Radar echo data is added, and the reflectivity factor is first interpolated to the model grid points. The reflectivity factor value on the grid point is compared with the threshold within the radar scanning range. If the reflectivity factor is lower than the threshold, it is considered as clear sky. If it is higher than the threshold, the cloud base and cloud top are corrected. Satellite visible light cloud images are added to estimate the total cloud cover and calibrate overestimation of clouds. The three-dimensional cloud cover is comprehensively corrected to obtain the final model three-dimensional cloud distribution.
[0209] Through cloud analysis, multiple variables can be obtained in areas with cloud distribution. Unlike traditional cloud analysis, these variables will enter the 3DVAR assimilation as false observations. Cloud analysis information will become a constraint information for model assimilation analysis and enter the assimilation system. Through this analysis method, the lack of cloud and rain information in the model, as well as the heat, humidity and other information in the cloud and rain area can be better supplemented. At the same time, in 3DVAR, due to the constraints of the model background field and background error covariance matrix, the analysis variables will be more coordinated. Through the cloud analysis module, a strictly corrected three-dimensional cloud distribution map of the model is obtained. The cloud distribution area is extracted to obtain variables such as humidity, cloud water, rain water, and temperature. These variables are assimilated into the 3DVAR assimilation system as a constraint information for assimilation analysis as false observations, thereby supplementing the lack of cloud and rain information in the model, as well as the heat, humidity and other information in the cloud and rain area.
[0210] The control variables and analysis variables are unified as u, v, w, t, q, and the stream function and potential function are no longer used as control variables. Such analysis results can better display the subtle characteristics of small and medium scales. For any forecast object, the environmental field and small and medium scale characteristics are analyzed at each analysis time. For convective-scale weather systems, in the GSI assimilation system, the control variables and analysis variables are unified as u, v, w, t, q. Analyze the assimilation results to see whether the analysis results show the subtle characteristics of small and medium scales better.
[0211] Starting from the cold start at 1200UTC every day, the three-dimensional variational assimilation is updated every 3 hours. The assimilation analysis is carried out simultaneously in the three regions of the model. The assimilation includes multi-source observation data (including conventional ground, ship, buoy, automatic station, aircraft message, sounding and other observation data) in the 1.5h assimilation window before and after the analysis time, as well as unconventional observation data such as satellite and radar. Then an 18h forecast is made until the next cold start at 1200UTC. The start of the numerical forecast model requires a driving field, and the three-hour forecast needs to update the driving field at any time. This module is responsible for real-time monitoring of the update of the driving data, collecting the driving data, and using it to start the numerical forecast model. The shell script is used to regularly monitor the update of global forecast fields such as ECMWF to obtain the latest driving data information. The download software is started through the shell script to automatically download the driving data, and the data is stored and backed up by category and date to facilitate future data calls. According to business needs, the spatial range and model resolution of the model area are determined. For large-scale fine forecasts, consider whether to set up nested domains and determine the appropriate projection method according to business requirements. Set the start and end time of the model simulation. If the system is started every 3 hours, the time setting in the model is updated through the shell script. The startup of the model requires the underlying surface information as the lower boundary condition. This module interpolates the global terrain static data available to the model into the simulation area. Read the simulation area information in the model preprocessing module, calculate the longitude and latitude and the scale factor of each grid point in the map. According to the interpolation method of each variable, variables such as soil type, land passage, terrain height, monthly vegetation cover and monthly albedo are interpolated to the model grid.
[0212] The numerical forecast model requires specific variable data and data format to start. When the output results of the global model or the reanalysis data are used as the driving field of the mesoscale model, due to its special storage format and its various variables, it cannot directly drive the numerical model. Therefore, this module is needed to decode and extract the driving field data. Select a coding table suitable for the driving field. The commonly used data released by NCEP and ECMWF are in GRIB format, and both formats have their own coding tables, namely Vtable files. Before extracting the driving field, you first need to select the applicable Vtable. Driving variable extraction. Determine which fields need to be extracted from the GRIB file through the determined coding table, and write the extracted files into intermediate files in a transitional format.
[0213] The extracted meteorological elements are horizontally interpolated to the simulation area determined in the model preprocessing module and written into a data format that can be directly absorbed by the model. Read TBL files. This module controls how to interpolate meteorological elements through TBL files. The TBL file provides an interval for each meteorological element, within which the interpolation method of the element field is determined. The interpolated meteorological elements are written. Usually, the interpolated meteorological elements are written into a data format that can be directly absorbed by the model, usually in NetCDF format, to facilitate visualization by visualization software.
[0214] Through horizontal and vertical interpolation, the global model background field data is processed onto the regional numerical forecast grid to produce lateral boundary conditions for the regional numerical forecast. The driving field data is read to prepare the soil field for the model (usually interpolated to the required height), and the soil classification, land passage, soil temperature and surface temperature are verified to be consistent with each other. After verifying the soil field, the upper air variables are interpolated onto the model calculation surface, that is, the set vertical layer, to generate initial conditions and lateral boundary conditions.
[0215] Focusing on the occurrence and development of small and medium-scale high-impact weather systems, the model settings are the same as above. Through multiple groups of sensitivity tests, the parameterization scheme suitable for the simulation area is compared. Read the initial conditions and lateral boundary conditions generated in the lateral boundary processing module, start the model to generate simulation results. The simulation results are collated and backed up.
[0216] S202: Use the WRF regional model to make forecasts and obtain numerical forecast data.
[0217] S203, post-correcting the numerical forecast data based on the multi-source meteorological observation data set after data processing by using a machine learning algorithm.
[0218] In some embodiments, S203 may specifically include:
[0219] Perform feature extraction, feature preprocessing, feature classification construction and feature combination on numerical forecast data;
[0220] The numerical forecast data are post-corrected through machine learning algorithms based on the feature combination of numerical forecast data and the multi-source meteorological observation data set after data processing.
[0221] Specifically, the role of GSI and WRF not only improves the temporal and spatial resolution of numerical forecasts, but also further improves the forecast accuracy of the model. However, on this basis, the machine learning method is used to make more accurate meteorological element forecasts for the target site in order to lay a good foundation for the following power forecast. The process includes the following steps:
[0222] Download historical observation data and perform data cleaning on the historical observation data, including valid data screening, target threshold judgment, date uniqueness judgment, date supplementation, and missing data processing;
[0223] Download historical NWP numerical forecast data and perform feature engineering on NWP data, including feature extraction, feature preprocessing, feature classification construction, and feature combination;
[0224] Input the features and cleaned observation data into the model for learning. Different models have different learning results. Pick the best model and save the model parameters;
[0225] Directly input NWP data into the previously trained model and output the revised single-station numerical forecast product.
[0226] In some embodiments, four machine learning models, including decision tree (DT), random forest, GBDT (GradientBoosting Decision Tree) and XGBoost, are used to select the best.
[0227] like Figure 4 As shown, the water-wind-solar power generation combined prediction device based on hybrid neural network includes: a data collection unit 401, a model building unit 402 and a prediction unit 403;
[0228] A data collection unit 401 is used to obtain a multi-source meteorological observation data set of a target area;
[0229] The data collection unit 401 is also used to obtain a historical power generation data set of the hydro-wind-solar power stations in the target area;
[0230] A model building unit 402 is used to build and train a GRU-CNN hydro-wind-solar power generation joint prediction model based on a multi-source meteorological observation data set and a historical power generation data set;
[0231] The prediction unit 403 is used to obtain the meteorological forecast data of the target area in the time period to be predicted in real time, and transmit it to the trained GRU-CNN water-wind-solar power joint prediction model to predict the hydropower, wind power and photovoltaic power of the water-wind-solar power station in the target area in the time period to be predicted.
[0232] like Figure 5 As shown, the present disclosure provides an electronic device, including a processor 501, a communication interface 502, a memory 503 and a communication bus 504, wherein the processor 501, the communication interface 502 and the memory 503 communicate with each other through the communication bus 504;
[0233] Memory 503, storing computer programs;
[0234] The processor 501 is configured to implement the above method when executing the computer program stored in the memory 503 .
[0235] The present disclosure provides a computer-readable storage medium storing a computer program, and the computer program implements the above method when executed by a processor.
[0236] The computer-readable storage medium may be included in the device / apparatus described in the above embodiment; or it may exist independently without being assembled into the device / apparatus. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.
[0237] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.
[0238] Although the present disclosure has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure.
Claims
1. A joint prediction method for hydropower, wind power and solar power generation based on hybrid neural network. It is characterized in that include: Obtain multi-source meteorological observation datasets for the target area; Obtain the historical power generation data set of hydropower, wind and solar power stations in the target area; Based on multi-source meteorological observation data sets and historical power generation data sets, a GRU-CNN hydro-wind-solar power generation joint prediction model was established and trained; The meteorological forecast data of the target area in the predicted time period is obtained in real time, and transmitted to the trained GRU-CNN water-wind-solar power joint prediction model to calculate the hydropower, wind power and photovoltaic power of the water-wind-solar power stations in the target area in the predicted time period.
2. According to the method for joint prediction of water, wind and solar power generation based on hybrid neural network of claim 1, It is characterized in that Multi-source meteorological observation data sets, including: meteorological observation data of hydropower stations, wind power stations, satellite meteorological observation data, radar meteorological observation data, ground meteorological observation data, aircraft meteorological observation data and sounding meteorological observation data.
3. According to the combined prediction method of water, wind and solar power generation based on hybrid neural network of claim 1, It is characterized in that Acquire a multi-source meteorological observation dataset for the target area, including: Numerical forecast assimilation is performed on multi-source meteorological observation datasets in the target area.
4. According to the method for joint prediction of water, wind and solar power generation based on hybrid neural network of claim 3, It is characterized in that Perform numerical forecast assimilation on multi-source meteorological observation datasets in the target area, including: Assimilate and process the initial fields of the multi-source meteorological observation data set to obtain the multi-source meteorological observation data set after data processing; Use the WRF regional model for forecasting and obtain numerical forecast data; The numerical forecast data is post-corrected based on the multi-source meteorological observation data set after data processing through machine learning algorithms.
5. According to the method for joint prediction of water, wind and solar power generation based on hybrid neural network of claim 4, It is characterized in that Assimilation and data processing of the initial fields of multi-source meteorological observation datasets, including: The following quality control is performed on the multi-source meteorological observation dataset, and the meteorological observation dataset after quality control is obtained: Repeatability check, validity check, continuity check, extreme value check, position consistency check and spatiotemporal consistency check.
6. According to the method for joint prediction of water, wind and solar power generation based on hybrid neural network of claim 5, It is characterized in that The duplication check is used to determine whether there are multiple sets of detection data within the time, longitude, latitude and altitude limits, and to remove duplicate data.
7. According to the method for joint prediction of water, wind and solar power generation based on hybrid neural network of claim 5, It is characterized in that The validity check is used to determine the availability of the current message based on the observed data.
8. According to the method for joint prediction of water, wind and solar power generation based on hybrid neural network of claim 5, It is characterized in that The continuity check is used to check whether the observed data caused by instrument failure remains unchanged or changes less than the threshold. When the difference between adjacent message data is less than the given criterion for three or more consecutive times, the instrument is judged to have failed the continuity check.
9. According to the method for joint prediction of water, wind and solar power generation based on hybrid neural network of claim 5, It is characterized in that The extreme value check is used to detect whether the observation values of the ground regional station of the observed element are within the extreme value range under the corresponding latitude and altitude conditions.
10. According to the method for joint prediction of water, wind and solar power generation based on hybrid neural network of claim 5, It is characterized in that The location consistency check is used to eliminate data quality problems caused by time collection and positioning errors. It uses the previous and next messages to determine whether the current message remains consistent in time, longitude and latitude, and altitude changes.
11. According to the method for joint prediction of water, wind and solar power generation based on hybrid neural network of claim 5, It is characterized in that The time-space consistency check is used to detect whether the message remains consistent as it changes over time and space.
12. The method for joint prediction of water, wind and solar power generation based on hybrid neural network according to claim 4, It is characterized in that The numerical forecast data is post-corrected based on the multi-source meteorological observation data set after data processing through machine learning algorithms, including: Perform feature extraction, feature preprocessing, feature classification construction and feature combination on numerical forecast data; The numerical forecast data are post-corrected through machine learning algorithms based on the feature combination of numerical forecast data and the multi-source meteorological observation data set after data processing.
13. According to the method for joint prediction of water, wind and solar power generation based on hybrid neural network of claim 1, It is characterized in that The GRU-CNN hydro-wind-solar power generation joint prediction model includes: a heterogeneous information encoding layer, a spatiotemporal feature fusion layer, and a power decoding prediction layer connected in sequence; The heterogeneous information encoding layer is used to encode heterogeneous information in the multi-source meteorological observation data and historical power generation data of the hydropower station, wind power station and photovoltaic power station, and generate a spatial matrix of hydropower, a spatial matrix of wind power and a spatial matrix of photovoltaic power; The spatiotemporal feature fusion layer is used to generate the spatiotemporal feature matrix of water energy, the spatiotemporal feature matrix of wind energy and the spatiotemporal feature matrix of light energy based on the spatial matrix of water energy, the spatial matrix of wind energy and the spatial matrix of light energy; The power decoding prediction layer is based on the spatiotemporal feature matrix of water energy, the spatiotemporal feature matrix of wind energy, and the spatiotemporal feature matrix of light energy. It is used to perform upsampling operations through deconvolution, and then decodes the deconvolution features based on GRU, and outputs the data through the fully connected layer.
14. The method for joint prediction of water, wind and solar power generation based on hybrid neural network according to claim 13, It is characterized in that The heterogeneous information includes the historical hydroelectric power, historical water level, historical wind power, historical wind speed, historical photovoltaic power, historical irradiance and load data of the hydropower, wind and solar power stations in the target area.
15. The method for joint prediction of water, wind and solar power generation based on hybrid neural network according to claim 14, It is characterized in that The heterogeneous information encoding layer includes a wind power related data encoder, a hydropower related data encoder, a photovoltaic related data encoder and a load data encoder; The input of the wind power related data encoder includes the wind power generation power at time tn, the observed wind speed of the hydro-wind-solar power station at time tn, and the forecast meteorological elements of the hydro-wind-solar power station from time t to time tn; The input of the hydropower related data encoder includes the hydropower generation power at time tn, the observed runoff of the hydro-wind-solar power station at time tn, and the forecast meteorological elements of the hydro-wind-solar power station from time t to time tn; The input of the photovoltaic related data encoder includes the photovoltaic power generation at time tn, the observed irradiance of the water, wind and solar power station at time tn, and the forecast meteorological elements of the water, wind and solar power station from time t to time tn; The input of the load data encoder includes the load at time tn and the load at time tnm, where t, n, and m are natural numbers.
16. The method for joint prediction of water, wind and solar power generation based on hybrid neural network according to claim 13, It is characterized in that Generate a spatial matrix of water energy, a spatial matrix of wind energy, and a spatial matrix of light energy, including: Based on heterogeneous information, the spatial characteristics of water energy, wind energy and light energy are obtained through convolution extraction; According to the spatial characteristics of water energy, wind energy and light energy, the corresponding spatial matrix of water energy, the spatial matrix of wind energy and the spatial matrix of light energy are generated.
17. The method for joint prediction of water, wind and solar power generation based on hybrid neural network according to claim 13, It is characterized in that Based on the spatial matrix of water energy, the spatial matrix of wind energy and the spatial matrix of light energy, the spatiotemporal characteristic matrix of water energy, the spatiotemporal characteristic matrix of wind energy and the spatiotemporal characteristic matrix of light energy are generated, including: Based on heterogeneous information, the time matrix of water energy, the time matrix of wind energy and the time matrix of light energy are extracted through GRU; Based on the time matrix of water energy, the time matrix of wind energy, the time matrix of light energy, the space matrix of water energy, the space matrix of wind energy and the space matrix of light energy, the spatiotemporal characteristic matrix of water energy, the spatiotemporal characteristic matrix of wind energy and the spatiotemporal characteristic matrix of light energy are generated.
18. The method for joint prediction of water, wind and solar power generation based on hybrid neural network according to claim 13, It is characterized in that The spatiotemporal feature fusion layer adopts the attention mechanism to set a weight for all the hidden states in the encoder, then weighted sum each hidden state according to the set weight, and then input all the hidden states after weighted sum into the decoder.
19. A combined prediction device for hydropower, wind power and solar power generation based on a hybrid neural network. It is characterized in that include: data collection unit, model building unit and prediction unit; A data collection unit, used to obtain a multi-source meteorological observation data set for a target area; The data collection unit is also used to obtain a historical power generation data set of the hydro-wind-solar power stations in the target area; A model building unit is used to build and train a GRU-CNN hydro-wind-solar power generation joint prediction model based on a multi-source meteorological observation data set and a historical power generation data set; The prediction unit is used to obtain the meteorological forecast data of the target area in the time period to be predicted in real time, and transmit it to the trained GRU-CNN water-wind-solar power joint prediction model to predict the hydropower, wind power and photovoltaic power of the water-wind-solar power station in the target area in the time period to be predicted.
20. An electronic device, It is characterized in that It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; a memory storing a computer program; The processor is used to implement the water-wind-solar power generation joint prediction method based on a hybrid neural network according to any one of claims 1 to 18 when executing the computer program stored in the memory.
21. A computer-readable storage medium storing a computer program, It is characterized in that When the computer program is executed by a processor, the method for joint prediction of water, wind and solar power generation based on a hybrid neural network of any one of claims 1 to 18 is implemented.
Citation Information
Patent Citations
Ultra-short-term wind power plant power prediction method combined with meteorological factors
CN112733462A
Construction method and system of wind-solar power combined prediction model and prediction method
CN115271191A
Offshore wind power prediction method and device based on multi-scale space-time diagram transformation network
CN117114192A
Wind-solar combined prediction model construction method and system based on data coupling
CN117763958A
Method for identifying convective cloud cluster characteristic parameters based on Himawar8 satellite data
CN117975289A
Cited By
Machine learning correction forecasting method for wind and light elements of new energy power station in complex terrain
CN120387529A
