Wind power plant cluster joint power prediction method and system considering privacy protection

Abstract

The invention discloses a wind power plant cluster joint power prediction method and system considering privacy protection, and relates to the technical field of new energy power system and artificial intelligence crossing, and the method comprises the steps: obtaining the historical time series data of a wind power plant, inputting the historical time series data of the wind power plant into a pre-established TCN-AE model, and obtaining a TCN-AE model; encoder outputs to obtain an encrypted time sequence feature; and extracting the encrypted time sequence features by using a self-attention mechanism to obtain wind power plant features, inputting the wind power plant features into a pre-established prediction network model based on a marble loss function, and outputting to obtain a wind power plant cluster joint power prediction result.

Description

Technical Field

[0001] The present invention relates to the cross - technical field of new - energy power systems and artificial intelligence, and specifically to a method and system for combined power prediction of wind farm clusters considering privacy protection. Background Technique

[0002] In recent years, as an important part of renewable energy, the penetration rate of wind power in the power system has been continuously increasing. The output of wind power has significant intermittency and volatility, bringing new challenges to the safe and stable operation of the power grid. In order to improve the dispatchability and grid - connection ability of wind power, wind power prediction technology has received extensive attention.

[0003] Wind power prediction methods can be divided into single - station prediction methods and multi - station collaborative prediction methods. Single - station prediction methods usually rely on the historical operation data of a single wind farm and predict the future wind power through time - series models or neural - network models. This type of method is relatively common in actual deployment, but it cannot utilize the spatio - temporal correlation between multiple wind farms, and there are certain limitations in prediction accuracy.

[0004] To further improve the prediction accuracy, multi - station collaborative prediction methods have been proposed. This type of method realizes the joint modeling and prediction of regional wind power by introducing the observation information of multiple wind farms. However, the existing collaborative prediction methods generally face the following problems:

[0005] Firstly, there are problems such as different data owners and high data privacy requirements among wind farms, resulting in difficulties in directly sharing the original data, which limits the popularization and application of collaborative modeling.

[0006] Secondly, when traditional collaborative modeling methods deal with the spatio - temporal dependence relationship of multiple wind farms, they mainly adopt static correlation coefficients or fixed - weight methods based on graph structures, and it is difficult to dynamically capture the non - linear changes of spatio - temporal features, resulting in limited feature - fusion effects.

[0007] Thirdly, existing wind power prediction methods mostly focus on point prediction, lack the modeling of the uncertainty of prediction results, and it is difficult to meet the requirements of confidence intervals in dispatch decision - making.

[0008] Fourthly, some studies have introduced the federated learning mechanism to realize model training under data isolation, but its dependence on the gradient - exchange process still has the risk of gradient inversion, and it will introduce a large communication overhead and system complexity in the collaborative application of multiple wind farms. Summary of the Invention

[0009] To solve the deficiencies mentioned in the above background technique, the purpose of the present invention is to provide a method and system for combined power prediction of wind farm clusters considering privacy protection.

[0010] In a first aspect, the object of the present invention can be achieved by the following technical solutions: A method for combined power prediction of a wind farm cluster considering privacy protection, the method comprising the following steps:

[0011] Obtain the historical time series data of the wind farm, and input the historical time series data of the wind farm into a pre-established TCN-AE model. The Encoder outputs encrypted time series features;

[0012] Use the self-attention mechanism to extract the encrypted time series features to obtain the wind farm features, and input the wind farm features into a pre-established prediction network model based on the pinball loss function, and output the combined power prediction result of the wind farm cluster.

[0013] Combined with the first aspect, in some implementation manners of the first aspect, the method further includes: The historical time series data X of the wind farm i ={x i,t ∣t = 1, 2, …, T}, where represents the observed data of the i-th wind farm at time t, T represents the sequence length, and d is the feature dimension.

[0014] Combined with the first aspect, in some implementation manners of the first aspect, the method further includes: The process of inputting the historical time series data of the wind farm into a pre-established TCN-AE model, and the Encoder outputs encrypted time series features:

[0015] Perform encrypted feature extraction on X i to obtain the encrypted data feature representation h i , satisfying:

[0016]

[0017] where Encoder TCN is an encryptor based on TCN, and Decoder TCN is a decryptor based on TCN.

[0018] And minimize the reconstruction error L rec :

[0019]

[0020] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: The temporal convolutional network TCN of the pre-established TCN-AE model includes causal convolution and dilated convolution structures. Among them, causal convolution is used to ensure that the prediction moment only depends on historical information, satisfying causality in the time direction. The convolution kernel of causal convolution only acts on the input features of the current and past moments; dilated convolution captures the dynamic correlation relationship within a long time span by inserting a hole dilation step in the convolution operation, so that the receptive field expands exponentially. The sequence is modeled using the following formula:

[0021]

[0022] Among them, represents the convolution output of the l-th layer, K is the convolution kernel size, is the convolution kernel weight, σ(·) is the activation function, is the bias of the l-th layer, and L is the total number of layers of the TCN.

[0023] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: The process of using the self-attention mechanism to extract the encrypted temporal features to obtain the wind farm features:

[0024] Input the encrypted features h of each wind farm i into the self-attention module. Based on the self-attention mechanism, perform weighted interaction with other wind farm features to generate a joint feature vector s i :

[0025]

[0026] Among them, is the wind farm set, q i , k j are the query and key vectors respectively, α ij is the attention score, and exp() is the exponential function of nature.

[0027] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: Inputting the wind farm features into the pre-established prediction network model based on the pinball loss function, and the output obtains the joint power prediction result of the wind farm cluster, including:

[0028] Input s i into the prediction network for probability prediction, and output the prediction result

[0029]

[0030] Among them, f pb is the improved pinball loss function model, p represents the predicted value, is a set of confidence quantiles.

[0031] Combined with the first aspect, in some implementations of the first aspect, the method further includes: the pre-established prediction network model based on the pinball loss function is trained based on the improved pinball loss function, and the improved loss is:

[0032]

[0033] where is the upper bound of the probability prediction, is the lower bound of the probability prediction, γ is the weight coefficient, ρ τ (u) = max(τu, (τ - 1)u).

[0034] In a second aspect, to achieve the above object, the present invention discloses a wind farm cluster combined power prediction system considering privacy protection, including:

[0035] A time series extraction module, configured to obtain the historical time series data of the wind farm, input the historical time series data of the wind farm into the pre-established TCN-AE model, and the Encoder outputs the encrypted time series features;

[0036] A power prediction module, configured to use the self-attention mechanism to extract the encrypted time series features to obtain the wind farm features, input the wind farm features into the pre-established prediction network model based on the pinball loss function, and output the combined power prediction result of the wind farm cluster.

[0037] In another aspect of the present invention, to achieve the above object, a terminal device is disclosed, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The computer program stored in the memory is capable of running on the processor. When the processor loads and executes the computer program, it adopts a method for predicting the combined power of a wind farm cluster considering privacy protection as described above.

[0038] In still another aspect of the present invention, to achieve the above object, a computer-readable storage medium is disclosed. The computer-readable storage medium stores a computer program. When the computer program is loaded and executed by a processor, it adopts a method for predicting the combined power of a wind farm cluster considering privacy protection as described above.

[0039] Advantages of the present invention:

[0040] The present invention can fully utilize spatio-temporal correlation on the premise of protecting the privacy of the original data of each wind farm, realize collaborative wind power prediction among multiple wind farms, and effectively improve the prediction accuracy and system robustness. By introducing a local feature extraction mechanism based on a temporal convolutional network-autoencoder, the present invention avoids the risk of original data leakage; uses a self-attention mechanism to dynamically model the spatio-temporal dependence relationship among wind farms, enhancing the collaborative modeling ability; combines quantile output with an improved pinball loss function to construct a probability prediction model with the ability to output prediction intervals, improving the ability to characterize prediction uncertainty; at the same time, through a joint optimization strategy, feature reconstruction and prediction targets are co-trained, effectively improving the generalization ability and stability of the model. Compared with traditional centralized modeling or static dependence fusion methods, the present invention takes into account privacy protection, modeling ability and prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0042] Figure 1 It is a schematic flow chart of the method of the present invention;

[0043] Figure 2 It is a schematic diagram of the framework of the privacy-preserving wind power collaborative prediction method of the selective spatio-temporal dependence extraction and encrypted sharing mechanism of the present invention;

[0044] Figure 3 It is a schematic diagram of the comparison of the probability prediction effects of the method of the present invention and other methods;

[0045] Figure 4 It is a schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0047] Embodiment 1:

[0048] As Figure 1 shown, a method for joint power prediction of a wind farm cluster considering privacy protection includes the following steps:

[0049] S101: Obtain the historical time series data of the wind farm, and input the historical time series data of the wind farm into the pre-established TCN-AE model. The Encoder outputs the encrypted time series features;

[0050] Construct the time series data set X of the local wind farm i =[x i,1 ,x i,2 ,…,x i,T , where represents the observation feature of the i-th wind farm at time t, T represents the sequence length, and d is the feature dimension;

[0051] The process of inputting the historical time series data of the wind farm into the pre-established TCN-AE model and the Encoder outputting the encrypted time series features:

[0052] Extract encrypted feature extraction from X i to obtain the encrypted data feature representation h i , satisfying:

[0053]

[0054] where Encoder TCN is an encryptor based on TCN, and Decoder TCN is a decryptor based on TCN.

[0055] And minimize the reconstruction error L rec :

[0056]

[0057] The temporal convolutional network TCN of the pre-established TCN-AE model includes causal convolution and dilated convolution structures. Among them, causal convolution is used to ensure that the prediction moment only depends on historical information and satisfies causality in the time direction. The convolution kernel of causal convolution only acts on the input features of the current and past moments; dilated convolution captures the dynamic correlation relationship within a long time span by inserting a hole dilation step in the convolution operation, so that the receptive field expands exponentially. The sequence is modeled using the following formula:

[0058]

[0059] where represents the output of the l-th layer of convolution, K is the convolution kernel size, is the convolution kernel weight, σ(·) is the activation function, is the bias of the l-th layer, and L is the total number of layers of TCN.

[0060] Adopt a localization feature extraction method (TCN-AE) that combines a temporal convolutional network (TCN) and an autoencoder (AE) to encrypt the original data of the wind farm into a low-dimensional feature representation, sharing only the encrypted features instead of the original data to achieve data privacy protection. Avoid directly transmitting the original data to eliminate the risk of privacy leakage from the source, while retaining the temporal dynamic correlation.

[0061] S102: Use the self-attention mechanism to extract the encrypted temporal features to obtain the wind farm features, and input the wind farm features into a pre-established prediction network model based on the pinball loss function to output the combined power prediction result of the wind farm cluster.

[0062] The process of using the self-attention mechanism to extract the encrypted temporal features to obtain the wind farm features:

[0063] Input the encrypted features h of each wind farm i into the self-attention module. Based on the self-attention mechanism, perform weighted interaction with the features of other wind farms to generate a joint feature vector s i :

[0064]

[0065] where, is the wind farm set, q i , k j are the query and key vectors respectively, α ij is the attention score, and exp() is the natural exponential function.

[0066] Through the self-attention mechanism (Self-Attention), dynamically calculate the attention weights between the encrypted features of different wind farms, and selectively aggregate the spatio-temporal dependence relationships among multiple wind farms. Overcome the limitations of traditional static weights or fixed graph structures, adaptively capture non-linear spatio-temporal correlations, and improve the collaborative modeling accuracy.

[0067] The process of inputting the wind farm features into a pre-established prediction network model based on the pinball loss function to output the combined power prediction result of the wind farm cluster includes:

[0068] Input s i into the prediction network for probability prediction and output the prediction result:

[0069]

[0070] where, f pb is the improved pinball loss function model, p represents the predicted value, is a set of confidence quantiles.

[0071] The pre-established prediction network model based on the pinball loss function is trained based on the improved pinball loss function, and the improved loss is as follows:

[0072]

[0073] where is the upper bound of probability prediction, is the lower bound of probability prediction, γ is the weight coefficient, and ρ τ (u) = max(τu, (τ - 1)u).

[0074] Based on the traditional quantile loss (Pinball loss), an interval width penalty term is introduced to optimize the compactness of the prediction interval and support the output of multiple confidence levels.

[0075] On the premise of ensuring the interval coverage probability (PICP), the prediction interval width (PINAW) is significantly reduced, and the uncertainty modeling ability is improved (as shown in Table 1, the CWC index is reduced by 14.5% - 26.6%).

[0076] The joint fine-tuning mechanism is optimized using the following weighted loss function:

[0077] L total = λ·(L rec - L ae ) + L pre

[0078] It is carried out locally in the wind farm without cross-site transmission of raw data, model parameters or gradients, ensuring that the original information is not leaked during the prediction process.

[0079] Through the joint loss function, the feature reconstruction error and the prediction error are optimized simultaneously, realizing the coordinated improvement of feature encryption and prediction performance. The generalization ability of the model is enhanced, avoiding overfitting or feature degradation caused by a single optimization target (as shown in experiments, the reconstruction error is reduced by 12%).

[0080] Specifically, the solution of the present invention is further elaborated through the following embodiments:

[0081] The method of the present invention has been verified for performance on the real datasets of multiple wind farms. The data used are the wind farm stations on the "one map" platform of the State Grid of a certain province: JH, DT, RH, HH, ZGS, RQ, FH, RD, TH, KS, a total of ten wind farm stations, covering different geographical locations and climate conditions, and having strong representativeness. The model takes the 24-hour wind power sequence as the input, and the wind power of the next hour is the prediction target. The prediction results cover two forms: point value output and interval prediction.

[0082] Table 1 summarizes the performance comparison between the traditional probability regression method and the method of the present invention on the Huahai (Zhu) Wind Farm dataset in the interval prediction task. The comparison models include the original Pinball loss method (PB) and the Bayesian Ridge Regression model (Bayes). Figure 3 It shows the comparison of probability predictions of various methods. Each model takes the first 24-hour data of wind power as input and outputs the prediction intervals at 90%, 80%, and 70% confidence levels for the next hour. The evaluation metrics adopted include: Prediction Interval Coverage Probability (PICP), Prediction Interval Width (PINAW), and Coverage Width-based Criterion (CWC).

[0083] Combining Table 1 and Figure 3 From the results, it can be seen that in the probability prediction task, while maintaining a high coverage rate, the method of the present invention can generate more compact prediction intervals, demonstrating better interval prediction performance. At the 90% confidence level, the Prediction Interval Coverage Probability (PICP) of the method of the present invention is 90.90%, which is closer to the set confidence level compared to the PB method (91.72%) and the Bayes method (92.54%), indicating that the prediction intervals generated by this method have higher accuracy and credibility. At the same time, the Coverage Width-based Criterion (CWC) of the method of the present invention is 0.2604, which is approximately 14.5% and 23.9% lower than that of the PB method (0.3046) and the Bayes method (0.3422) respectively, significantly improving the compactness of the interval while maintaining reliable coverage.

[0084] In addition, at the 80% and 70% confidence levels, the method of the present invention also shows consistent advantages, with a higher degree of matching in PICP and CL, and the CWC index is improved by approximately 11.5% and 16.4% compared to the PB method, and by approximately 11.3% and 26.6% compared to the Bayes method. This result further verifies that the method of the present invention has good generalization ability and stability at multiple confidence levels.

[0085] The experimental results fully show that the method of the present invention takes into account the sharpness and coverage of the prediction interval in the interval prediction task, can effectively improve the uncertainty modeling level in wind power prediction, and is applicable to the scheduling optimization task that requires fine control of the risk range.

[0086] Performance comparison of the proposed method in Table 1 with Bayes and PB methods

[0087]

[0088] Embodiment 2: Second aspect, as Figure 4 shown, to achieve the above object, the present invention discloses a wind farm cluster combined power prediction system considering privacy protection, including:

[0089] A time series extraction module 11, configured to obtain historical time series data of a wind farm, input the historical time series data of the wind farm into a pre-established TCN-AE model, and the Encoder outputs encrypted time series features;

[0090] A power prediction module 12, configured to use a self-attention mechanism to extract wind farm features from the encrypted time series features, input the wind farm features into a pre-established prediction network model based on a pinball loss function, and output a combined power prediction result of the wind farm cluster.

[0091] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors, and a memory for storing one or more computer programs; the program includes program instructions, and the processor is configured to execute the program instructions stored in the memory. The processor may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is used to implement one or more instructions, specifically for loading and executing one or more instructions in the computer storage medium to implement the above method.

[0092] It should be further noted that, based on the same inventive concept, the present invention also provides a computer storage medium, on which a computer program is stored, and the computer program, when run by a processor, executes the above method. The storage medium may adopt any combination of one or more computer-readable media. The computer-readable media may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the computer-readable storage medium include: an electrical connection with one or more wires, 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), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program may be used by or in combination with an instruction execution system, apparatus, or device.

[0093] In the description of this specification, the descriptions referring to the terms "one embodiment", "example", "specific example", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples.

[0094] The above shows and describes the basic principles, main features, and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited by the above embodiments, and the above embodiments and the descriptions in the specification only illustrate the principles of the present disclosure. Without departing from the spirit and scope of the present disclosure, the present disclosure will have various changes and improvements, and these changes and improvements all fall within the scope of the present disclosure claimed.

Claims

1. A wind farm cluster joint power prediction method taking privacy protection into account, characterized in that: The method comprises the following steps: Obtain the historical time series data of the wind farm, input the historical time series data of the wind farm into the pre-established TCN-AE model, and the Encoder outputs the encrypted time series features; The encrypted time series features are extracted using the self-attention mechanism to obtain wind farm features, which are then input into a pre-established prediction network model based on the pinball loss function to output the joint power prediction results of the wind farm cluster.

2. A wind farm cluster joint power prediction method taking privacy protection into account according to claim 1, characterized in that: The wind farm historical time series data X i ={x i,t |t=1,2,…,T}, where represents the observed data of the i-th wind farm at time t, T represents the sequence length, and d is the feature dimension.

3. A wind farm cluster joint power prediction method taking privacy protection into account according to claim 1, characterized in that: The process of inputting the historical time series data of the wind farm into the pre-established TCN-AE model and outputting the encrypted time series features by the Encoder is as follows: X i Perform encryption feature extraction to obtain the encrypted data feature representation h i ,satisfy: The Encoder TCN It is a TCN-based encryptor, Decoder TCN It is a TCN-based decryptor; And minimize the reconstruction error L rec :

4. A wind farm cluster joint power prediction method taking privacy protection into account according to claim 3, characterized in that: The temporal convolutional network TCN of the pre-established TCN-AE model includes causal convolution and dilated convolution structures, wherein the causal convolution is used to ensure that the prediction moment depends only on historical information and satisfies the causality in the time direction, and the convolution kernel of the causal convolution only acts on the input features of the current and past moments; the dilated convolution inserts a hole expansion step in the convolution operation to expand the receptive field exponentially to capture the dynamic correlation relationship within a long time span, and the sequence is modeled using the following formula: in, represents the output of the lth layer of convolution, K is the size of the convolution kernel, is the convolution kernel weight, σ(·) is the activation function, is the bias of the lth layer, and L is the total number of layers of TCN.

5. The method for wind farm cluster joint power prediction taking privacy protection into account according to claim 1, characterized in that: The process of extracting encrypted time series features using the self-attention mechanism to obtain wind farm features: The encrypted features h of each wind farm i Input into the self-attention module, based on the self-attention mechanism, weighted interaction is performed with other wind farm features to generate a joint feature vector s i : in, is the wind farm collection, q i ,k j are query and key vectors respectively, α ij is the attention score, and exp() is the natural exponential function.

6. A wind farm cluster joint power prediction method taking privacy protection into account according to claim 1, characterized in that: The wind farm characteristics are input into a pre-established prediction network model based on a pinball loss function, and the wind farm cluster joint power prediction result is output, including: Will s i Input the prediction network for probability prediction and output the prediction result Among them, f pb To improve the pinball loss function model, p represents the predicted value, is a set of confidence quantiles.

7. A wind farm cluster joint power prediction method taking privacy protection into account according to claim 6, characterized in that: The pre-established prediction network model based on the pinball loss function is trained based on the improved pinball loss function, and the improved loss is: in, is the upper bound of the probability prediction, is the lower bound of probability prediction, γ is the weight coefficient, ρ τ (u) = max(τu, (τ-1)u).

8. A wind farm cluster joint power prediction system taking privacy protection into account, characterized in that: include: The time series extraction module is used to obtain the historical time series data of the wind farm, input the historical time series data of the wind farm into the pre-established TCN-AE model, and the encoder outputs the encrypted time series features; The power prediction module is used to extract the encrypted time series features using the self-attention mechanism to obtain the wind farm features, input the wind farm features into a pre-established prediction network model based on the pinball loss function, and output the joint power prediction results of the wind farm cluster.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: The memory stores a computer program that can be run on the processor. When the processor loads and executes the computer program, a wind farm cluster joint power prediction method taking privacy protection into account as described in any one of claims 1 to 7 is adopted.

10. A computer-readable storage medium having a computer program stored therein, characterized in that: When the computer program is loaded and executed by the processor, a wind farm cluster joint power prediction method taking privacy protection into account as described in any one of claims 1 to 7 is adopted.

Images