CN115860215A - Photovoltaic and wind power generation power prediction method and system - Google Patents
Photovoltaic and wind power generation power prediction method and system Download PDFInfo
- Publication number
- CN115860215A CN115860215A CN202211520752.1A CN202211520752A CN115860215A CN 115860215 A CN115860215 A CN 115860215A CN 202211520752 A CN202211520752 A CN 202211520752A CN 115860215 A CN115860215 A CN 115860215A
- Authority
- CN
- China
- Prior art keywords
- power generation
- photovoltaic
- wind
- power
- prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a method and a system for predicting power generation power of photovoltaic and wind power generation, which relate to the technical field of wind and light power generation and specifically comprise the following steps: constructing a power generation power prediction model based on a convolutional neural network and a bidirectional gated cyclic neural network; acquiring historical data of photovoltaic power generation and wind power generation, extracting features and constructing a feature map; extracting features of the feature map, and representing feature vectors of the photovoltaic and wind power generation power; outputting the characteristic vector as a time sequence, training in the power generation power prediction model, and capturing a long-term training relation in the time sequence; outputting a prediction result; and evaluating the prediction result. The prediction model fully extracts local relevant characteristics of data, and tracks actual photovoltaic power generation and wind power generation power; and excavating long-term dependence in the sequence to make up for the defects of the convolutional neural network, so that high-precision prediction of photovoltaic and wind power is realized.
Description
Technical Field
The invention relates to the technical field of wind and light power generation, in particular to a method and a system for predicting power generation power of photovoltaic wind power generation.
Background
With the storage of fossil fuel energy becoming smaller and the environmental problems caused by the storage becoming more and more serious, the share of photovoltaic power generation and wind power generation in the power market becomes higher and higher. At present, the generated power of wind and light power generation has the characteristics of intermittence and uncertainty, and supply and demand balance is a key condition for maintaining stable operation of a power grid, so that the prediction of the wind and light power generation power is particularly important.
At present, the output power of a photovoltaic power generation system can be predicted by various methods, but physical modeling, traditional statistical modeling and artificial intelligence algorithm modeling are mainly performed aiming at the photovoltaic power generation characteristics so as to achieve the purpose of improving the prediction precision of the output power of the photovoltaic power generation.
Generally, physical modeling refers to reasonably predicting by building a corresponding data model by fully considering physical characteristics of a photovoltaic cell in the process of converting solar energy into electric energy by using a photovoltaic cell panel. The modeling can obtain results only by utilizing characteristic parameters of a photovoltaic system battery pack and various characteristic values in a power grid after photovoltaic power generation is connected, does not completely depend on past historical data of the photovoltaic power generation system, and can be used for generating input variables of a statistical model or other prediction models. The traditional statistical modeling mainly refers to continuous exploration on historical data output by a photovoltaic power generation system, and various interference factors influencing the output power of the photovoltaic power generation system are used as input of a building model, so that a high-efficiency prediction result is obtained through a corresponding statistical prediction model. Because the method is established above the research starting point of historical output data of a power generation system, the method is more widely applied compared with a physical method, but is possibly influenced by nonlinear variables and environmental factors. The method has the advantages that the meaning of the traditional statistical method is supplemented through continuous expansion and innovation, the algorithm structure is continuously and iteratively updated, and common machine learning algorithms and deep learning neural network algorithms are researched more frequently.
Although various methods for predicting the output power of the photovoltaic power generation system exist, the deep learning neural network method has higher prediction performance than methods such as a physical prediction model in a normal state. Although the effect is better in the aspect of photovoltaic power generation output power prediction, a deep learning neural network still has a great number of problems from the perspective of model building and characteristic variable processing and needs to be deeply researched:
at present, the situations of excessive parameters of a prediction model exist, and various data are manually adjusted according to experience, so that the randomness of the parameters is increased, and the prediction precision is influenced to a great extent. The photovoltaic power generation system has uneven output historical data, and great difficulty is caused to the establishment of a high-quality prediction model. Using an inappropriate algorithm to construct the model may result in lower prediction performance.
Therefore, a new photovoltaic and wind power generation power prediction method is needed to be provided, the traditional method is improved, and appropriate historical data are selected to realize high-precision prediction of photovoltaic and wind power.
Disclosure of Invention
Based on the problems, the invention provides a photovoltaic and wind power generation power prediction method.
In order to achieve the purpose, the invention provides the following technical scheme:
the invention provides a photovoltaic and wind power generation power prediction method, which specifically comprises the following steps:
s1, constructing a power generation power prediction model based on a convolutional neural network and a bidirectional gated cyclic neural network;
s2, acquiring historical data of photovoltaic power generation and wind power generation, extracting features, and constructing a feature map;
s3, characterizing eigenvectors of the photovoltaic and wind power generation power;
s4, outputting the characteristic vector as a time sequence, training in the power generation power prediction model, and capturing a long-term training relation in the time sequence;
s5, outputting a prediction result;
and S6, evaluating the prediction result.
Preferably, the historical data at least comprises installed capacity, photovoltaic power generation, wind power generation, light irradiance, illumination intensity and wind speed.
Preferably, the historical data is input through an input layer of the convolutional neural network, local correlation characteristics of the historical data are extracted through the convolutional layer, matrix element multiplication summation is carried out on the input characteristics in the receptive field, and deviation amount is superposed.
Preferably, after the feature extraction is performed on the convolutional layer, the output feature map is transmitted to the pooling layer for feature selection and information filtering; the spatial topology is then lost in the fully-connected layer, expanded into eigenvectors and passed through the excitation function.
Preferably, the method for capturing the long-term training relationship in the time series comprises:
constructing a bidirectional gating cyclic neural network training model, wherein the bidirectional gating cyclic neural network training model is used for realizing existence of an update gate z t And a reset gate r t ；
When no valid information exists in the input time sequence, the gate r is reset t Infinite approach 1, update door z t If the distance is infinitely close to 0, the information of the training cannot be stored;
when valid information exists in the input time sequence and the history information is invalid, the gate r is reset t Infinitely close to 0, update gate z t If the distance is infinitely close to 1, the information of the training is saved, and the historical information is excluded;
when valid information exists in the input time sequence and the history information is valid, the gate r is reset t Infinitely close to 0, update gate z t Infinitely close to 0.5, the information of the training is saved and the history information is kept.
Preferably, the time series of inputs is defined as x t Candidate hidden states
And a hidden state H t The computational expressions for the update gate, reset gate, and hidden state are as follows:
z t ＝(W z X t + z H t-1 + z )
r t ＝(W r X t + r H t-1 )。
preferably, the evaluation method of the prediction result is as follows: the root mean square error eRMSE, the average absolute percentage error eMAPE, the regression coefficient R and the square R2 of the regression coefficient are used as evaluation indexes to evaluate the prediction result,
the specific evaluation mode is as follows:
wherein, P r A representation of the observed value of the power, P p and expressing a power predicted value, wherein N is the total number of predicted future photovoltaic power points.
The invention also provides a photovoltaic and wind power generation power prediction system, which adopts any one of the above prediction methods for photovoltaic and wind power generation power to predict, and comprises the following steps:
the data acquisition module is used for acquiring historical data of photovoltaic power generation and wind power generation;
the characteristic extraction module is used for extracting the characteristics of the characteristic diagram and representing the characteristic vectors of the photovoltaic and wind power generation power; outputting the feature vector as a time series;
the training module is used for training based on the generated power prediction model, capturing a long-term training relation in the time sequence and outputting a prediction result;
and the evaluation module is used for evaluating the prediction result and outputting the evaluation result.
Compared with the prior art, the invention has the following advantages:
the invention provides a method and a system for predicting power of photovoltaic and wind power generation, which are formed by combining a convolutional neural network and a bidirectional gated cyclic neural network, wherein the convolutional neural network is used for fully extracting local relevant characteristics of data so as to improve the local prediction performance of a model, and the addition of a convolutional neural network model enables a model power predicted value to be capable of tracking actual photovoltaic power generation and wind power generation power to the maximum extent; the bidirectional gated cyclic neural network can be transmitted forwards and backwards, and the defect of insufficient data information mining of the unidirectional transmission gated cyclic neural network is overcome. The bidirectional gated cyclic neural network can learn and maximally mine long-term dependence in the sequence to make up for the defects of the convolutional neural network, so that high-precision prediction of photovoltaic and wind power is realized.
Drawings
FIG. 1 is a schematic flow diagram of a photovoltaic and wind power generation power prediction method;
FIG. 2 is a schematic diagram of a power generation power prediction model;
FIG. 3 is a schematic diagram of a bi-directional gated recurrent neural network;
FIG. 4 is a data graph constructed based on historical data;
FIG. 5 is a schematic diagram of a partitioning into training sets and test sets based on a data graph;
FIG. 6 is a graph showing the drop in loss function during model training;
FIG. 7 is a comparison of wind power prediction results and actual observation values on a test set after model training.
Detailed Description
In order to make the purpose and technical solution of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below with reference to the embodiments of the present invention.
Example one
The invention provides a photovoltaic and wind power generation power prediction method, which specifically comprises the following steps as shown in figures 1-3:
s1, constructing a power generation power prediction model based on a convolutional neural network and a bidirectional gated cyclic neural network;
s2, acquiring historical data of photovoltaic power generation and wind power generation, and constructing a characteristic diagram;
s3, extracting the characteristics of the characteristic diagram, and representing the characteristic vectors of the photovoltaic and wind power generation power;
s4, outputting the characteristic vector as a time sequence, training in the power generation power prediction model, and capturing a long-term training relation in the time sequence;
s5, outputting a prediction result;
and S6, evaluating the prediction result.
Among them, convolutional Neural Networks (CNN) are a kind of feed forward Neural Networks (fed forward Neural Networks) containing convolution calculation and having a deep structure, and are one of the representative algorithms of deep learning (deep learning).
The convolutional neural network includes an input layer, a hidden layer, and an output layer.
The input layer may process multidimensional data, and typically, the input layer of a one-dimensional convolutional neural network receives a one-dimensional or two-dimensional array, where the one-dimensional array is typically a time or frequency spectrum sample; the two-dimensional array may include a plurality of channels; an input layer of the two-dimensional convolutional neural network receives a two-dimensional or three-dimensional array; the input layer of the three-dimensional convolutional neural network receives a four-dimensional array. The input features of the convolutional neural network require normalization processing. Specifically, before inputting the learning data into the convolutional neural network, the input data needs to be normalized in the channel or time/frequency dimension, and if the input data is a pixel, the original pixel values distributed in the input data can be normalized to an interval.
The hidden layer comprises a convolution layer, a pooling layer and a full-link layer 3 common structures, and in some more modern algorithms, there may be complicated structures such as an inclusion module and a residual block (residual block). In a common architecture, convolutional and pooling layers are characteristic of convolutional neural networks. The convolution kernel in the convolutional layer contains weight coefficients, while the pooling layer does not, and therefore in the literature, the pooling layer may not be considered a separate layer. Taking LeNet-5 as an example, the order of 3 types of common structures in the hidden layer is usually: input-convolutional layer-pooling layer-full-link layer-output. The function of the convolution layer is to extract the characteristics of input data, the convolution layer internally comprises a plurality of convolution kernels, and each element forming the convolution kernels corresponds to a weight coefficient and a deviation quantity (bias vector), and is similar to a neuron (neuron) of a feedforward neural network. Each neuron in the convolution layer is connected to a plurality of neurons in a closely located region in the previous layer, the size of which depends on the size of the convolution kernel, and is referred to in the literature as the "receptive field", which means a field analogous to that of visual cortical cells. When the convolution kernel works, the convolution kernel regularly sweeps the input characteristics, matrix element multiplication summation is carried out on the input characteristics in the receptive field, and deviation amount is superposed.
The convolutional layer parameters comprise the size of a convolutional kernel, step length and filling, the size of an output characteristic diagram of the convolutional layer is determined by the convolutional layer parameters, and the convolutional layer parameters are hyper-parameters of a convolutional neural network. Where the convolution kernel size can be specified as an arbitrary value smaller than the input image size, the larger the convolution kernel, the more complex the input features that can be extracted.
The convolution step defines the distance between the positions of the convolution kernels when the convolution kernels sweep the feature map twice, when the convolution step is 1, the convolution kernels sweep the elements of the feature map one by one, and when the step is n, n-1 pixels are skipped in the next scanning. As can be seen from the cross-correlation calculation of convolution kernels, the size of the feature map gradually decreases with the stacking of convolution layers, for example, a 16 × 16 input image outputs a 12 × 12 feature map after passing through a unit step size, unfilled 5 × 5 convolution kernel. To this end, padding is a method of artificially increasing the size of the feature map before it passes through the convolution kernel to offset the effects of size shrinkage in the computation. A common padding method is padding by 0 and repeated boundary value (replication padding). Filling can be divided into four categories according to its number of layers and numbers:
valid padding (valid padding): i.e., no padding is used at all, the convolution kernel only allows access to locations in the signature map that contain the complete receptive field. All pixels of the output are a function of the same number of pixels in the input. The convolution using the effective padding is called "narrow convolution", and the feature size of the output of the narrow convolution is (L-f)/s +1.
Same fill/half fill (same/half fill): only enough padding is done to keep the feature sizes of the output and input the same. The feature map is not reduced in size under the same fill but the portions of the input pixels near the boundary have less effect on the feature map than the middle portion, i.e., there is under-representation of the boundary pixels. The convolution using the same padding is called "equal-length convolution".
Full padding (full padding): the filling is done sufficiently that each pixel is accessed the same number of times in each direction. When the step size is 1, the characteristic diagram size of the full filling output is L + f-1 and is larger than the input value. Convolution using full padding is called "wide convolution"
Arbitrary padding (arbitrary padding): between active and full fill, artificially set fills, are less used.
After the feature extraction is performed on the convolutional layer, the output feature map is transmitted to the pooling layer for feature selection and information filtering. The pooling layer contains a preset pooling function, the function of which is to replace the result of a single point in the feature map with the feature map statistics of its neighboring area. The step of selecting the pooling area by the pooling layer is the same as the step of scanning the characteristic diagram by the convolution kernel, and the pooling size, the step length and the filling are controlled.
The fully-connected layer in the convolutional neural network is equivalent to the hidden layer in the conventional feedforward neural network. The fully-connected layer is located at the last part of the hidden layer of the convolutional neural network and only signals are transmitted to other fully-connected layers. The feature map loses spatial topology in the fully connected layer, is expanded into vectors and passes through the excitation function. From the aspect of characterization learning, the convolutional layer and the pooling layer in the convolutional neural network can extract features of input data, and the fully-connected layer is used for performing nonlinear combination on the extracted features to obtain output, namely the fully-connected layer is not expected to have feature extraction capacity, but is used for trying to complete a learning target by using existing high-order features.
The convolutional neural network is usually a fully-connected layer upstream of the output layer, and thus has the same structure and operation principle as the output layer in the conventional feedforward neural network. As a representative algorithm of deep learning, the convolutional neural network has a characteristic learning capability, that is, a capability of extracting high-order features from input information. In particular, convolutional and pooling layers in a convolutional neural network can respond to translational invariance of input features, i.e., can identify nearby features at spatially distinct locations. Being able to extract translation invariant features is one of the reasons why convolutional neural networks find application in computer vision problems.
A gated recurrent neural network (GRU) includes two gate units, an update gate and a reset gate. The reset gate is a combination of the memory unit and the hidden layer, and is used for controlling the hidden state information at the previous moment to be transferred to the candidate hidden state at the current moment so as to reset the candidate hidden state information at the current moment. The updating gate is the combination of the forgetting gate and the input gate and is used for controlling the hidden state information at the previous moment to be transferred to the hidden state at the current moment so as to update the hidden state information at the current moment.
GRU network is divided into reset gate r t And a refresh gate z t And candidate hidden states
And hidden state H t . The computational expressions for the update gate, reset gate, and hidden state are as follows.
And (4) updating the door: the input is the input X of the current time t And hidden state H at the previous moment t-1 The output of the update gate is z t 。z t Is X t And H t-1 The linear combination of (a) and (b) is input to a sigmoid function to obtain a value of 0 to 1.
z t ＝(W z X t + z H t-1 + z )
Resetting a gate: input X is still the input at the current moment t And hidden state H at the previous moment t-1 The output of the reset gate is r t 。r t Is X t And H t-1 The linear combination of (a) and (b) is input to a sigmoid function to obtain a value of 0 to 1.
r t ＝(W r X t + r H t-1 )
Candidate hidden states: will reset the gate output r t And previous time hidden state H t-1 Element-wise multiplication, with the result of the operation and the current input X t Linear combination is carried out and then input into the tanh function to obtain the current candidate hidden state
Has a value of-1 to 1.
From the above formula, it can be seen that the reset door r t When r is t When =0, reset gate output r at this time t And hidden state H at last moment t-1 The result of the element-wise multiplication is 0. Meaning the hidden state H at the previous moment t-1 Hiding the current candidate
No effect, which is equivalent to discarding the hidden state H at the previous moment t-1 And (4) information. Then the candidate hidden state H at the current time t Input X only with the current time t In this regard, this gives a hidden state to reset. Also because of this, resetting the gate at this time helps to capture short term dependencies. Hidden state update gate output z t And hidden state H at last moment t-1 And a current time candidate hidden state>
Linear combination is performed.
From the above formula, the updated door z can be seen t When z is t 1- t And =0. Hidden state H at the previous moment t-1 Completely gives the hidden state H of the current moment t The hidden state at the current moment is completely reserved. At this time, if there is a long-term dependency relationship, the hidden state information can be transmitted and retained all the time. Because of this, the GRU can capture longer-term dependencies and is also the most critical part of the GRU network. Then the update gate helps to capture long term dependencies.
The basic idea of the bidirectional gating recurrent neural network (BiGRU) is as follows: for each training sequence, two RNN models are established in the forward and reverse directions, and hidden layer nodes of the two models are connected to the same nodeAnd (5) outputting the layer. The data processing method may provide complete historical and future information for each point in time in the output layer input sequence. Thus, the BiGRU network has the ability to learn the relationship between past and future photovoltaic impact factors and current photovoltaic power, which helps to extract features of the photovoltaic data. In the forward layer, we compute the forward direction from time step 1 to time step t and obtain and save the output s of each forward hidden layer t . In the reverse layer, reverse calculation is performed from the current time step t to the last time t-1, and the output s of each time of the reverse hidden layer is obtained and saved t '. Finally, a final output o is obtained by combining the output results of the forward layer and the reverse layer at a time t 。
Based on a CNN-BiGRU photovoltaic power generation power and wind power prediction model, a basic framework is shown in the figure. The deep learning fusion model mainly comprises a CNN and a bidirectional GRU neural network, wherein the CNN is used for fully extracting local relevant characteristics of data so as to improve the local prediction performance of the model, and the addition of the CNN enables a model power prediction value to be capable of tracking actual photovoltaic power generation and wind power generation power to the maximum extent; the bidirectional GRU neural network can be transmitted forwards and backwards, and the defect that data information is not sufficient in mining through the unidirectional GRU neural network is overcome. The bidirectional GRU network can learn and maximally mine long-term dependencies in the sequence to make up for the CNN deficiency. The Dropout strategy is employed to avoid overfitting of the model and to improve the generalization capability of the model.
Firstly, constructing a characteristic diagram of photovoltaic historical data and wind-power data in a sliding window mode, inputting the characteristic diagram into a CNN, and extracting characteristic vectors representing dynamic changes of photovoltaic and wind-power by utilizing a convolution layer and a pooling layer of the characteristic diagram; and then converting the output vector into a time sequence and inputting the time sequence into a bidirectional GRU network so as to further capture a long-term dependence relationship in the time sequence, thereby realizing high-precision prediction of photovoltaic and wind power.
The evaluation method of the prediction result comprises the following steps: using root mean square error e RMSE Mean absolute percentage error e MAPE Regression coefficient R, square R of regression coefficient 2 Evaluating the predicted result as an evaluation indexThe price mode is as follows:
wherein, P r A representation of the observed value of the power, P p and expressing the power predicted value, wherein N is the total point number of the predicted future photovoltaic power.
Example two
The photovoltaic and wind power generation power prediction method is provided based on one embodiment and is explained by taking a specific example.
Wind power generation data of a period from 12 months and 1 day in 2021 to 3 months and 1 day in 2022 of a wind farm in a Ma-Huang-Tan wind farm are selected as a research case of the project, and the time resolution of the data is 15 minutes. The wind power data is plotted and shown in fig. 4. The data of the three months of the wind farm of the Ma-Huang Tan-Sha is divided into a training set and a test set, and the proportion of the training set and the test set is divided into 8:2, as shown in FIG. 5. The generated power prediction model constructed by data training of the wind farm of the Ma-Huang beach is adopted, and the loss function of the model in the training process is reduced as shown in FIG. 6. The prediction result of the model on the test set is shown in fig. 7, and the curve of the prediction result is basically overlapped with the curve of the observation value, so that the prediction effect is good.
Using root mean square error e RMSE Mean absolute percentage error e MAPE Regression coefficient R, square R of regression coefficient 2 The prediction results were evaluated as evaluation indexes.
RMSE (Root Mean Square Error) Root Mean Square Error, which measures the deviation between the observed and the true values. The Mean Absolute Percentage Error of MAPE (Mean Absolute Percentage Error) is one of the most popular indicators for evaluating predictive performance.
In the formula: p r Representing observed values of power, P p And expressing the power predicted value, wherein N is the total point number of the predicted future photovoltaic power.
The specific evaluation results are shown in table 1:
TABLE 1 wind power prediction result evaluation
| Evaluation index | Numerical value |
| RMSE | 10.8538 |
| MAPE | 45186.445 |
| R | 0.9887 |
| R 2 | 0.9777 |
EXAMPLE III
The invention also provides a photovoltaic and wind power generation power prediction system, which adopts any one of the above prediction methods for photovoltaic and wind power generation power to predict, and comprises the following steps:
the data acquisition module is used for acquiring historical data of photovoltaic power generation and wind power generation;
the characteristic extraction module is used for extracting characteristics of the characteristic diagram and representing characteristic vectors of photovoltaic and wind power generation power; outputting the feature vector as a time series;
the training module is used for training based on a power generation power prediction model, capturing a long-term training relation in the time sequence and outputting a prediction result;
and the evaluation module is used for evaluating the prediction result and outputting the evaluation result.
The above description is only an embodiment of the present invention, and the present invention is described in detail and specifically, but not to be construed as limiting the scope of the present invention. It should be noted that, for those skilled in the art, various changes and modifications can be made without departing from the spirit of the present invention, and these changes and modifications are within the scope of the present invention.
Claims (8)
1. A photovoltaic and wind power generation power prediction method is characterized by comprising the following steps:
the method specifically comprises the following steps:
s1, constructing a power generation power prediction model based on a convolutional neural network and a bidirectional gated cyclic neural network;
s2, acquiring historical data of photovoltaic power generation and wind power generation, extracting features, and constructing a feature map;
s3, characterizing eigenvectors of the photovoltaic and wind power generation power;
s4, outputting the characteristic vector as a time sequence, training in the power generation power prediction model, and capturing a long-term training relation in the time sequence;
s5, outputting a prediction result;
and S6, evaluating the prediction result.
2. A photovoltaic and wind-powered electricity generation power prediction method as claimed in claim 1, characterized by: the historical data at least comprises installed capacity, photovoltaic power generation, wind power generation, light irradiance, illumination intensity and wind speed.
3. A photovoltaic and wind-powered electricity generation power prediction method as claimed in claim 1, characterized by: and inputting the historical data through an input layer of the convolutional neural network, extracting local relevant characteristics of the historical data through the convolutional layer, and performing matrix element multiplication summation on the input characteristics in a receptive field and superposing deviation values.
4. A photovoltaic and wind-powered electricity generation power prediction method as claimed in claim 3, characterized by: after the feature extraction is carried out on the convolutional layer, the output feature graph is transmitted to the pooling layer for feature selection and information filtering; the spatial topology is then lost in the fully-connected layer, expanded into eigenvectors and passed through the excitation function.
5. A photovoltaic and wind-powered electricity generation power prediction method as claimed in claim 1, characterized by: the method for capturing the long-term training relationship in the time sequence comprises the following steps:
constructing a two-way gated recurrent neural network training model that subsists an update gate z t And a reset gate r t ；
When no valid information exists in the input time sequence, the gate r is reset t Infinite approach 1, update door z t If the distance is infinitely close to 0, the information of the training cannot be stored;
when valid information exists in the input time sequence and the history information is invalid, the gate r is reset t Infinitely close to 0, update gate z t If the distance is infinitely close to 1, the information of the training is saved, and the historical information is excluded;
when there is valid information in the input time series and the history information is valid, the gate r is reset t Infinitely close to 0, update gate z t Infinitely close to 0.5, the information of the training is saved and the historical information is preserved.
6. Photovoltaic and wind-powered electricity generation according to claim 5A method of power prediction, characterized by: defining the time series of inputs as x t Candidate hidden states
And a hidden state H t The computational expressions for the update gate, reset gate, and hidden state are as follows:
z t ＝σ(W z X t +U z H t-1 +b z )
r t ＝σ(W r X t +U r H t-1 )。
7. a photovoltaic and wind-powered electricity generation power prediction method as claimed in claim 1, characterized by: the evaluation method of the prediction result comprises the following steps: using root mean square error e RMSE Mean absolute percentage error e MAPE The regression coefficient R and the square R2 of the regression coefficient are used as evaluation indexes to evaluate the prediction result,
the specific evaluation mode is as follows:
wherein, P r Representing the observed value of power, P p And expressing a power predicted value, wherein N is the total number of predicted future photovoltaic power points.
8. A photovoltaic and wind power generation power prediction system is characterized in that: the photovoltaic and wind power generation power prediction method according to any one of claims 1 to 7 is adopted for prediction, and comprises the following steps:
the data acquisition module is used for acquiring historical data of photovoltaic power generation and wind power generation;
the characteristic extraction module is used for extracting characteristics of the characteristic diagram and representing characteristic vectors of photovoltaic and wind power generation power; outputting the feature vector as a time series;
the training module is used for training based on a power generation power prediction model, capturing a long-term training relation in the time sequence and outputting a prediction result;
and the evaluation module is used for evaluating the prediction result and outputting the evaluation result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211520752.1A CN115860215A (en) | 2022-11-29 | 2022-11-29 | Photovoltaic and wind power generation power prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211520752.1A CN115860215A (en) | 2022-11-29 | 2022-11-29 | Photovoltaic and wind power generation power prediction method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115860215A true CN115860215A (en) | 2023-03-28 |
Family
ID=85668331
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211520752.1A Pending CN115860215A (en) | 2022-11-29 | 2022-11-29 | Photovoltaic and wind power generation power prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115860215A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116911467A (en) * | 2023-09-12 | 2023-10-20 | 浙江华云电力工程设计咨询有限公司 | Renewable energy output prediction method, device and storage medium |
| CN117040030A (en) * | 2023-10-10 | 2023-11-10 | 国网浙江宁波市鄞州区供电有限公司 | New energy consumption capacity risk management and control method and system |
-
2022
- 2022-11-29 CN CN202211520752.1A patent/CN115860215A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116911467A (en) * | 2023-09-12 | 2023-10-20 | 浙江华云电力工程设计咨询有限公司 | Renewable energy output prediction method, device and storage medium |
| CN117040030A (en) * | 2023-10-10 | 2023-11-10 | 国网浙江宁波市鄞州区供电有限公司 | New energy consumption capacity risk management and control method and system |
| CN117040030B (en) * | 2023-10-10 | 2024-04-02 | 国网浙江宁波市鄞州区供电有限公司 | New energy consumption capacity risk management and control method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110059878B (en) | Photovoltaic power generation power prediction model based on CNN LSTM and construction method thereof | |
| Ma et al. | A hybrid attention-based deep learning approach for wind power prediction | |
| Golovko et al. | Convolutional neural network based solar photovoltaic panel detection in satellite photos | |
| CN115860215A (en) | Photovoltaic and wind power generation power prediction method and system | |
| Saeed et al. | Hybrid bidirectional LSTM model for short-term wind speed interval prediction | |
| CN113053115A (en) | Traffic prediction method based on multi-scale graph convolution network model | |
| CN112561058A (en) | Short-term photovoltaic power prediction method based on Stacking-ensemble learning | |
| Mu et al. | ENSO forecasting over multiple time horizons using ConvLSTM network and rolling mechanism | |
| CN109272332B (en) | Client loss prediction method based on recurrent neural network | |
| CN114422382B (en) | Network flow prediction method, computer device, product and storage medium | |
| CN113255995A (en) | Air pollution prediction method | |
| CN112183742A (en) | Neural network hybrid quantization method based on progressive quantization and Hessian information | |
| CN116345555A (en) | CNN-ISCA-LSTM model-based short-term photovoltaic power generation power prediction method | |
| CN115640901A (en) | Small sample load prediction method based on hybrid neural network and generation countermeasure | |
| CN116844041A (en) | Cultivated land extraction method based on bidirectional convolution time self-attention mechanism | |
| CN116384583A (en) | Photovoltaic power prediction method based on multiple neural networks | |
| CN115641720A (en) | Traffic prediction method and system based on space-time fusion graph neural network | |
| Al-Hajj et al. | Multi-level stacking of long short term memory recurrent models for time series forecasting of solar radiation | |
| CN114006411A (en) | Wind power prediction method and system based on LSTM-CNN combined model | |
| CN110717374A (en) | Hyperspectral remote sensing image classification method based on improved multilayer perceptron | |
| CN115828726A (en) | Machine learning-based solar cross-season heat storage and supply system optimization control method | |
| Dai et al. | Multimodal deep learning water level forecasting model for multiscale drought alert in Feiyun River basin | |
| Wan et al. | Turbine location wind speed forecast using convolutional neural network | |
| CN113947243A (en) | Method for predicting photovoltaic power generation capacity based on DCLSTNet neural network model | |
| CN114372418A (en) | Wind power space-time situation description model establishing method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |