CN117154680A - Wind power prediction method based on non-stationary transducer model - Google Patents
Wind power prediction method based on non-stationary transducer model Download PDFInfo
- Publication number
- CN117154680A CN117154680A CN202310603115.9A CN202310603115A CN117154680A CN 117154680 A CN117154680 A CN 117154680A CN 202310603115 A CN202310603115 A CN 202310603115A CN 117154680 A CN117154680 A CN 117154680A
- Authority
- CN
- China
- Prior art keywords
- data
- wind power
- stationary
- power prediction
- model
- 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
- 238000000034 method Methods 0.000 title claims abstract description 51
- 230000007246 mechanism Effects 0.000 claims abstract description 32
- 238000012545 processing Methods 0.000 claims abstract description 27
- 238000004364 calculation method Methods 0.000 claims abstract description 10
- 239000013598 vector Substances 0.000 claims abstract description 10
- 238000012549 training Methods 0.000 claims description 24
- 238000003860 storage Methods 0.000 claims description 16
- 238000010606 normalization Methods 0.000 claims description 14
- 230000006870 function Effects 0.000 claims description 11
- 230000008569 process Effects 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 9
- 238000013528 artificial neural network Methods 0.000 claims description 8
- 230000002159 abnormal effect Effects 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 7
- 230000015654 memory Effects 0.000 claims description 7
- 238000012952 Resampling Methods 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 5
- 238000011105 stabilization Methods 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims 1
- 238000001556 precipitation Methods 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 12
- 230000005540 biological transmission Effects 0.000 description 4
- 238000013179 statistical model Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000001514 detection method Methods 0.000 description 3
- 230000008034 disappearance Effects 0.000 description 3
- 238000004880 explosion Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 238000002955 isolation Methods 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 238000003491 array Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 239000000835 fiber Substances 0.000 description 2
- 230000014509 gene expression Effects 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 230000001902 propagating effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- RYGMFSIKBFXOCR-UHFFFAOYSA-N Copper Chemical compound [Cu] RYGMFSIKBFXOCR-UHFFFAOYSA-N 0.000 description 1
- 239000009798 Shen-Fu Substances 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 229910052802 copper Inorganic materials 0.000 description 1
- 239000010949 copper Substances 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000001687 destabilization Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000029087 digestion Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 230000001939 inductive effect Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000013450 outlier detection Methods 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000001502 supplementing effect Effects 0.000 description 1
- 239000002352 surface water Substances 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/004—Generation forecast, e.g. methods or systems for forecasting future energy generation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/10—Power transmission or distribution systems management focussing at grid-level, e.g. load flow analysis, node profile computation, meshed network optimisation, active network management or spinning reserve management
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2300/00—Systems for supplying or distributing electric power characterised by decentralized, dispersed, or local generation
- H02J2300/20—The dispersed energy generation being of renewable origin
- H02J2300/28—The renewable source being wind energy
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Power Engineering (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a wind power prediction method based on a non-stationary transducer model. The method comprises the following steps: acquiring weather information and wind power sequence data at a plurality of historical moments; and weather information and wind power sequence data at a plurality of historical moments are used as input, and the trained wind power prediction model is utilized to predict wind power at future moments. The wind power prediction model is constructed based on a transducer model and comprises a position coding module, an encoder and a decoder, wherein the position coding module is used for adding position coding vectors into input time sequence data subjected to standardized processing and embedding layer processing and then transmitting the position coding vectors to the encoder; the output of the decoder is subjected to inverse standardization processing and then is used as the prediction output of the wind power prediction model; the encoder is provided with a non-stationary attention mechanism layer for introducing non-stationary information calculated from the input data into the calculation of the attention mechanism. The method and the device can provide the prediction accuracy and efficiency of wind power.
Description
Technical Field
The invention relates to the technical field of wind power prediction, in particular to a wind power prediction method based on a non-stationary transducer model.
Background
The accurate prediction of wind power is significant for wind farm site selection planning and stable operation of a power system, and can evaluate future comprehensive benefits, reduce standby capacity of the power system and reduce peak shaving pressure of the system. With the continuous increase of the installed scale of the offshore wind power, the provision of an accurate offshore wind power prediction method is particularly important due to the fluctuation and uncertainty of renewable energy sources.
Aiming at the problem of offshore wind power prediction, the prediction models proposed by researchers at present comprise a physical model, a statistical model, an intelligent model and the like. Numerical weather forecast is a typical physical model, and the method is mainly used for mid-term and long-term prediction to solve the problem of wind farm site selection planning. The physical model can carry out physical equation solving according to meteorological data to accurately obtain the predicted power value of the wind power plant, but a large amount of high-precision historical data is needed, and the data collection difficulty and the economic cost are improved. The common statistical model is an autoregressive moving average model, mainly predicts wind power in a future period of time through historical data in a certain period of time, and is effective for short-term power prediction. The statistical model can better capture the linear relationship between the data, but cannot effectively capture the nonlinear relationship between the data.
In recent years, with the development of artificial intelligence technology, on the offshore wind power prediction problem, an intelligent model has become a new research hotspot, and a method based on an artificial neural network has achieved a certain result. By inducing rules among historical wind power data, the intelligent model can avoid constructing a complex physical modeling process compared with a physical model, and can capture linear and nonlinear relations among data compared with a statistical model, so that the intelligent model has stronger self-adaptive capacity and nonlinear expression capacity and becomes a research hotspot of wind power prediction at present.
In the prior art, the offshore wind power prediction method based on the intelligent model mainly utilizes a cyclic neural network (Recurrent Neural Networks, RNN) and a long and short term memory network (Long Short Term Memory, LSTM). The network models can extract characteristic representation of time sequence data with time sequence characteristics through a circulating layer, such as dependency relationship of sequence context and information thereof. However, the training efficiency of the model such as RNN and LSTM is relatively low, gradient disappearance and gradient explosion easily occur during training, and parallelization calculation is difficult. Furthermore, researchers have proposed a Transformer-based offshore wind power prediction model (Lin Zheng, liu Kezhen, shen Fu, zhao Xianping, liang Yuping, dong Min) which can not only effectively improve the accuracy of offshore wind power prediction but also can extract features of global context information, and can effectively avoid gradient disappearance and gradient explosion problems in RNN and LSTM during training, and can perform parallel computation, taking into account the ultra-short-term power prediction model [ J ] of the spatiotemporal characteristics of offshore wind power, power system automation, 2022,46 (23): 59-66. In the field of time sequence prediction, the standardization of the sequence data can effectively improve the prediction capability of the model, but the data processed by the method becomes smoother, so that the transducer model only learns similar information for the characteristics of different attributes, and the application of the transducer in offshore wind power prediction is not facilitated.
In summary, the unique intermittence and uncertainty of renewable energy sources bring challenges to the site selection of offshore wind turbines and the stable operation of the power system. Therefore, the design of a set of method capable of accurately predicting the offshore wind power is particularly urgent.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a wind power prediction method based on a non-stationary transducer model. The method comprises the following steps:
aiming at a target area, weather information and corresponding wind power sequence data at a plurality of historical moments are obtained;
the weather information and the corresponding wind power sequence data at the plurality of historical moments are used as input, and the trained wind power prediction model is utilized to predict the wind power at the future moment;
the wind power prediction model is constructed based on a transducer model and comprises a position coding module, an encoder and a decoder, wherein the position coding module is used for adding position coding vectors into input time sequence data subjected to standardized processing and embedded layer processing and then transmitting the position coding vectors to the encoder; the output of the decoder is subjected to inverse standardization processing and then is used as the prediction output of the wind power prediction model; the encoder is provided with a non-stationary attention mechanism layer for introducing non-stationary information calculated from the input data into the calculation of the attention mechanism.
Compared with the prior art, the wind power prediction model of the non-stationary transducer is provided, and is particularly suitable for offshore wind power prediction, and compared with the existing transducer model, the wind power prediction model is added with three modules, namely a sequence standardization module, a sequence anti-standardization module and a stationary attention mechanism removing module, wherein the sequence standardization module and the sequence anti-standardization module can effectively improve prediction precision, and the stationary attention mechanism removing can improve learning ability of the transducer model on a stationary sequence. In addition, in order to improve the accuracy of the data set, the offshore wind power related data is preprocessed before the offshore wind power prediction model of the non-stationary transducer is trained, so that the accuracy and efficiency of prediction are further improved.
Other features of the present invention and its advantages will become apparent from the following detailed description of exemplary embodiments of the invention, which proceeds with reference to the accompanying drawings.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description, serve to explain the principles of the invention.
FIG. 1 is a flow chart of a wind power prediction method based on a non-stationary transducer model according to one embodiment of the present invention;
FIG. 2 is a schematic diagram of a process for training a wind power prediction model according to one embodiment of the invention;
FIG. 3 is a diagram of a non-stationary transducer model framework in accordance with one embodiment of the present invention.
Detailed Description
Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that: the relative arrangement of the components and steps, numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present invention unless it is specifically stated otherwise.
The following description of at least one exemplary embodiment is merely exemplary in nature and is in no way intended to limit the invention, its application, or uses.
Techniques, methods, and apparatus known to one of ordinary skill in the relevant art may not be discussed in detail, but are intended to be part of the specification where appropriate.
In all examples shown and discussed herein, any specific values should be construed as merely illustrative, and not a limitation. Thus, other examples of exemplary embodiments may have different values.
It should be noted that: like reference numerals and letters denote like items in the following figures, and thus once an item is defined in one figure, no further discussion thereof is necessary in subsequent figures.
In general, the wind power prediction method based on the nonstationary transform model improves the prediction accuracy and efficiency by preprocessing data (such as removing abnormal data, resampling, supplementing missing data, reducing noise and the like) and reintroducing nonstationary information in a data sequence into the attention mechanism calculation of the wind power prediction model. This method is particularly suitable for offshore wind power prediction, but can also be applied to other target areas. Hereinafter, the offshore wind power prediction will be described as an example.
Referring to fig. 1 and 2, the provided wind power prediction method based on the non-stationary converter model comprises the following steps:
in step S110, weather data and wind power of the target area are collected as a sample data set.
The problem of offshore wind power prediction requires predicting the power that an offshore wind turbine can generate in a future period of time based on various meteorological data. Taking offshore wind power sequence data of a certain place as an example, taking every five minutes as a sampling interval to collect a data set, weather data within 1 year and the power of a wind turbine are collected to construct a sample data set. For example, the weather data includes the characteristics of surface air pressure, relative humidity, surface water rate, wind speed, wind direction, air temperature, air density and the like of the altitude at 0m, 100m and 200m, and simultaneously, the power generated by the wind motor at the corresponding moment is obtained.
Step S120, preprocessing the acquired data set and constructing a training set.
In order to achieve both accuracy and efficiency of wind power prediction, the collected data set can be preprocessed. For example, the isolation forest is used for detecting abnormal value data of historical weather data of offshore wind power and power sequences thereof to be used so as to remove the abnormal data. Then, if the sampling density of the data set used is high, dense data is converted into data sets at intervals of fifteen minutes or thirty minutes by resampling to reduce the sampling frequency of the data. After the resampling is finished, if the condition that the data is missing in a certain time period exists, the missing data is complemented by adopting an interpolation method, and a new data set is obtained. Next, discrete wavelet transform is performed on the new data set subjected to outlier detection, resampling, and interpolation to realize noise reduction processing on the new data set. And finally, taking wind power generated by historical weather data and historical moment as input, and taking wind power at future moment as output, constructing a plurality of groups of input-output pairs, and constructing a training set and a testing set. For example, a majority (e.g., 80%) of the dataset is selected as the training set, and the remainder is selected as the test set.
Specifically, the isolated forest algorithm is an unsupervised anomaly detection algorithm based on ensemble learning. According to the method for isolating the forest, an anomaly detection model is built according to the characteristics of outlier points, and the outlier points are defined as isolated outlier points, namely the outlier points occupy less space in the whole data set and are far away from the high-density clusters. The core of the isolation forest algorithm is to construct a plurality of binary trees which are independent of each other, and abnormal value points existing in the data set are detected through the binary trees.
In one embodiment, these binary trees are constructed according to the following steps:
step S1, randomly selecting N sample points to form a sub-data set according to the constructed data set, and distributing the sub-data set to a root node of a binary tree, wherein weather data and wind power at each moment are used as one sample point.
Step S2, randomly selecting a feature from M features of the sub-data set, then selecting a value q between the maximum value and the minimum value of the feature, taking the feature value of the data point as the condition of the binary tree branch being greater than or equal to q, marking the feature value of the data point to the right branch, marking the rest to the left branch, and further dividing the sub-data set. This is equivalent to dividing the data space into two subspaces based on the dimension of the feature and the division point q.
Step S3, recursively executing step S2 to continuously construct new nodes until each data point is completely isolated or reaches the designed maximum limiting depth.
Because the isolation forest needs to generate T binary trees, the T independent binary trees can be obtained by repeatedly executing the steps T times.
Then, for each sample point s, the anomaly score is calculated comprehensively, and the anomaly condition of the sample point s is judged by the anomaly score. In one embodiment, the anomaly score is calculated using the following formula:
where h(s) represents the average path length of the sample points s in all binary trees, c (N) represents the average path length of the tree, and N is the total number of sample points in the data set. h(s) and c (N) are defined as follows:
where T is the number of binary trees, h t (s) represents the path length of the sample point s in the binary tree t.
Wherein H (k) represents a harmonic number estimate:
H(k)≈ln(k)+0.5772156649 (4)
the anomaly score for each sample point s is calculated according to the above equation, and if its score is very close to 1, this sample point must be the anomaly data, and if it is much smaller than, for example, 0.5, it is considered to be the normal data.
In one embodiment, the data set is noise reduced by a discrete wavelet transform, unlike a continuous wavelet transform, in which the scale factor a and the translation parameter b need to be discretized, expressed as:
wherein m and n control the selection of the wavelet function, a 0 And b 0 Typically set to 2 and 1.
The discrete wavelet transform is defined as:
wherein, psi is m，n (x) Is a wavelet function, expressed as:
using the above equation, the signal is subjected to multi-order decomposition, and the coefficient of the high frequency part can be set to 0, and then the noise-reduced signal is reconstructed using inverse wavelet transform, expressed as:
the discrete wavelet transformation can analyze the change of the signal in different time scales and frequency ranges, so that the local characteristics of the signal are effectively extracted, and the influence of signal noise is reduced. The present invention employs a discrete wavelet transform for noise reduction processing in view of the higher computational complexity of the continuous wavelet transform, the faster computational speed of the discrete wavelet transform, and in view of the fact that the acquired signal is discrete and not continuous.
Step S130, training a wind power prediction model by using a training set, wherein the wind power prediction model is constructed based on a transducer model and comprises a data standardization, an inverse standardization and a de-stabilization attention mechanism.
Firstly, a wind power prediction model is built based on a non-stationary transducer model, the non-stationary transducer model is utilized to conduct feature extraction on weather information and offshore wind power, and the relation between the weather information and the offshore wind power is built, so that accuracy of offshore wind power prediction is improved. The nonstationary transform model has strong global context feature extraction capability, high training efficiency, parallelization processing capability, strong robustness on data noise and capability of effectively processing stable data.
Referring to FIG. 3, the wind power prediction model constructed based on the nonstationary transform contains a multi-head attention mechanism, and the encoder-decoder architecture in the seq2seq model is better at learning long-term dependencies and easier to parallelize than RNNs. The model mainly comprises a position coding module, an encoder structure and a decoder structure. The encoder contains a non-stationary attention mechanism layer, residual connection and layer normalization, feed forward neural network. The decoder contains a masked multi-headed attention mechanism layer, residual connection and layer normalization, and a structure corresponding to the encoder structure. Wherein the non-stationary attention mechanism layer is configured to approximate non-stationary information from the input data and re-introduce the non-stationary information into the calculation of the attention mechanism.
Specifically, the position coding module adds a position coding vector to the input time series data subjected to the normalization processing and the embedded layer processing, so that the samples contain different position information after coding, and the coding is defined as follows:
where pos represents the position of the current data in the sample; d represents the dimension of the position-coding vector; p represents the position index of each value, even positions are sine-coded and odd positions are cosine-coded.
The encoder mainly comprises a multi-head attention mechanism layer, residual connection and layer standardization and a feedforward neural network. The sequence data after position coding is X, a transducer adopts a multi-head attention mechanism layer, a query matrix Q, a key matrix K and a value matrix V are obtained through linear mapping, and attention is calculated respectively, wherein the formula is described as follows:
wherein d k Representing the dimensions of the matrix K. For the multi-head attention mechanism, multiple groups of different linear transformations W are adopted j Q 、W j K And W is j V Multi-head attention was obtained, expressed as:
MultiHead(Q,K,V)＝Concat(head 1 ,...,head h )W o (12)
wherein head h ＝Attention(QW h Q ,KW h K ,VW h V ) Concat represents the concatenation of the matrices. The data passing through the multi-head attention mechanism layer is subjected to residual connection and layer standardization, then passes through the feedforward neural network, finally obtains the output of the encoder, and inputs the output into the decoder.
The decoder is similar to the encoder in structure, and specifically includes two multi-head attention mechanism layers, the first is a multi-head attention mechanism layer with a mask, ensuring that position prediction can only rely on data that has already occurred. The input of the decoder is input into the multi-head attention mechanism layer of the decoder together with the output of the encoder after the residual connection and normalization layer through the mask multi-head attention mechanism, and finally the predicted result is output through the network with the same structure as the decoder and is compared with the real data.
Compared with the existing transducer model, the non-stationary transducer model designed by the invention is newly added with three parts, namely data standardization, inverse standardization and destabilization attention mechanisms. Data normalization acts on the raw data and inverse normalization acts on the output of the model, so that processing the data is beneficial to improving the accuracy of the model. Whereas the de-stabilization attention mechanism helps the transducer model better learn the input and output relationships from the over-stabilized sequence.
Data normalization essentially refers to computing the mean and variance of the input sample data for each input sequenceThe mean and variance are calculated according to the following formula:
according to the mean mu X Sum of variancesThe raw data may be normalized, expressed as:
wherein, the ". Is dot product, x j Represents the j-th data in X.
And the inverse normalization process of the output data is expressed as:
y j ＝σ X ⊙(y′ j -μ X ) (16)
wherein y' j Representing the output of the decoder.
The de-stationary attention mechanism approximates non-stationary information from stationary data, and re-introduces the non-stationary information into the calculation of the attention mechanism, so as to solve the problem of data oversstationary, and improve the capability of a transducer model. The de-stationary attention mechanism is mainly to add a transform to the input data in the attention portion of the transducer's encoder, where the transform formula is expressed as:
wherein Q' = (Q-1 μ) Q T )/σ X ，K'＝(K-1μ K T )/σ X ，μ Q Mean value of Q in time dimension, mu K Represents the average value of K in the time dimension.
The de-stabilization factor is then learned directly from the non-stationary X, Q and K statistics by a simple and efficient multi-layer perceptron layer defined as follows:
logτ＝MLP(σ X ,X),Δ＝MLP(μ X ,X)
wherein Δ represents a desmethyl factor, V' = (V-1 μ) V T )/σ X ，μ V Represents the average value of V in the time dimension.
FIG. 2 is a training process for a non-stationary transducer model, wherein the model is trained using a training set and the predictive ability of the model is tested using a test set. The left side of fig. 2 depicts abnormal point detection, deletion, resampling, interpolation processing of the resampled missing data, noise reduction processing of the data set by discrete wavelet transform, and normalization processing. The training of the non-stationary transducer model and the predictive power test of the model on the processed dataset is depicted on the right side of fig. 2.
In one embodiment, in the process of training the non-stationary transducer model, the KL divergence calculation model can be used to calculate the error between the predicted output and the real data, and the calculation formula is expressed as:
wherein q (x i ) Representing an ith load predicted value (or power predicted value) in a predicted output sequence of the transducer model; p (x) i ) Representing q (x) i ) Load values at corresponding times. D (D) KL (P||Q) KL divergence loss function representing both P (X) and Q (X), D KL The smaller (P Q) the closer the two sets of data distributions are, the closer the distribution of Q (X) is approximated by iteratively training the neural network.
Step S140, for weather data and corresponding wind power sequence data of a plurality of historical moments acquired in real time, predicting wind power at future moments by using a trained wind power prediction model.
And after model training is completed, the method can be used for actual wind power prediction. For example, for weather data and corresponding wind power sequence data at a plurality of historical moments collected in real-time, wind power at future moments is predicted using a trained wind power prediction model.
The training process of the wind power prediction model can be performed offline in a server or a cloud, and the real-time wind power prediction can be realized by embedding the trained model into electronic equipment. The electronic device may be a terminal device or a server, where the terminal device includes any terminal device such as a mobile phone, a tablet computer, a Personal Digital Assistant (PDA), a point of sale (POS), a vehicle-mounted computer, and an intelligent wearable device (smart watch, virtual reality glasses, virtual reality helmet, etc.). The server includes, but is not limited to, an application server or a Web server, and may be a stand-alone server or a cluster server or a cloud server, etc.
In summary, the wind power prediction method based on the nonstationary transform model is provided, wherein abnormal value points and noise in a data set can be effectively removed by adopting the isolated forest and discrete wavelet transformation, so that the accuracy and generalization performance of wind power prediction are effectively improved. In addition, compared with other intelligent models, the non-stationary transducer can better extract the characteristics of global context, better construct the relation between weather information and wind power, and avoid gradient explosion and gradient disappearance in the training process. Compared with the existing transducer model, the method can better solve the problem of over-stable sequence. In addition, through computer simulation verification, the wind power prediction method can more accurately predict wind power and can be popularized to all offshore wind power prediction problems. For example, in the aspect of the site selection planning problem of the offshore wind farm, the method can predict the power generation power of the offshore wind farm, provide comments for the construction of the wind farm, improve the comprehensive benefit, and in the aspect of the stable operation problem of the power system, accurately evaluate the wind power in a future time period according to the current meteorological data, so that the power system can better schedule, reduce the peak regulation pressure of the system, ensure that the system can stably operate and improve the digestion capability of the offshore wind power.
The present invention may be a system, method, and/or computer program product. The computer program product may include a computer readable storage medium having computer readable program instructions embodied thereon for causing a processor to implement aspects of the present invention.
The computer readable storage medium may be a tangible device that can hold and store instructions for use by an instruction execution device. The computer readable storage medium may be, for example, but not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium would include the following: portable computer disks, hard disks, random Access Memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static Random Access Memory (SRAM), portable compact disk read-only memory (CD-ROM), digital Versatile Disks (DVD), memory sticks, floppy disks, mechanical coding devices, punch cards or in-groove structures such as punch cards or grooves having instructions stored thereon, and any suitable combination of the foregoing. Computer-readable storage media, as used herein, are not to be construed as transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., optical pulses through fiber optic cables), or electrical signals transmitted through wires.
The computer readable program instructions described herein may be downloaded from a computer readable storage medium to a respective computing/processing device or to an external computer or external storage device over a network, such as the internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmissions, wireless transmissions, routers, firewalls, switches, gateway computers and/or edge servers. The network interface card or network interface in each computing/processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium in the respective computing/processing device.
Computer program instructions for carrying out operations of the present invention may be assembly instructions, instruction Set Architecture (ISA) instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, c++, python, and the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer readable program instructions may be executed entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computer (for example, through the Internet using an Internet service provider). In some embodiments, aspects of the present invention are implemented by personalizing electronic circuitry, such as programmable logic circuitry, field Programmable Gate Arrays (FPGAs), or Programmable Logic Arrays (PLAs), with state information for computer readable program instructions, which can execute the computer readable program instructions.
Various aspects of the present invention are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer-readable program instructions.
These computer readable program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer readable medium having the instructions stored therein includes an article of manufacture including instructions which implement the function/act specified in the flowchart and/or block diagram block or blocks.
The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer, other programmable apparatus or other devices implement the functions/acts specified in the flowchart and/or block diagram block or blocks.
The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions. It is well known to those skilled in the art that implementation by hardware, implementation by software, and implementation by a combination of software and hardware are all equivalent.
The foregoing description of embodiments of the invention has been presented for purposes of illustration and description, and is not intended to be exhaustive or limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the various embodiments described. The terminology used herein was chosen in order to best explain the principles of the embodiments, the practical application, or the technical improvements in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein. The scope of the invention is defined by the appended claims.
Claims (10)
1. A wind power prediction method based on a non-stationary transducer model comprises the following steps:
aiming at a target area, weather information and corresponding wind power sequence data at a plurality of historical moments are obtained;
the weather information and the corresponding wind power sequence data at the plurality of historical moments are used as input, and the trained wind power prediction model is utilized to predict the wind power at the future moment;
the wind power prediction model is constructed based on a transducer model and comprises a position coding module, an encoder and a decoder, wherein the position coding module is used for adding position coding vectors into input time sequence data subjected to standardized processing and embedded layer processing and then transmitting the position coding vectors to the encoder; the output of the decoder is subjected to inverse standardization processing and then is used as the prediction output of the wind power prediction model; the encoder is provided with a non-stationary attention mechanism layer for introducing non-stationary information calculated from the input data into the calculation of the attention mechanism.
2. The method of claim 1, wherein the wind power prediction model is trained according to the steps of:
collecting weather data and wind motor emission power at a plurality of historical moments in a set time period as sample data, wherein the weather data are surface air pressure, relative humidity, surface precipitation rate, wind speed, wind direction, air temperature and air density at different altitudes;
taking weather data and wind power at each moment as a sample point, and removing abnormal sample points from the sample data;
resampling at a set sampling rate if the sampling density of the sample data is above a set threshold;
under the condition that missing data exists in a set time period, the data is complemented by interpolation operation, and a data set is obtained;
forming a plurality of sequences for different types of data in the data set, denoising each sequence by utilizing discrete wavelet transform to obtain denoised sequence data serving as a training set;
and pre-training the wind power prediction model by using the training set by taking the set loss function minimization as an optimization target.
3. The method of claim 2, wherein removing outlier sample points from the sample data comprises:
based on the training set, randomly selecting N sample points to form a sub-data set, and distributing the sub-data set to a root node of a binary tree, wherein N is an integer greater than 2;
randomly selecting a feature from M features of the sub data set, selecting a value q between the maximum value and the minimum value of the feature, marking the feature with the value more than or equal to q to the right branch and the rest to the left branch as the branching condition of the binary tree, so as to divide the child nodes of the binary tree, constructing new nodes through recursion execution, and obtaining T mutually independent binary trees, wherein T is an integer more than 2;
and calculating corresponding anomaly scores for each sample point in the training set by using the obtained T binary trees which are mutually independent so as to judge whether each sample point is anomaly data.
4. A method according to claim 3, wherein the anomaly score is calculated using the formula:
where h(s) represents the average path length of the sample points s in all binary trees, c (N) represents the average path length of all binary trees, N is the total number of sample points in the training set, h t (s) represents the path length of the sample point s in the binary tree t, and H () represents the harmonic number estimate.
5. The method of claim 1, wherein the encoder comprises the non-stationary attention mechanism layer, residual connection and layer normalization, a feed forward neural network, and the decoder comprises a masked multi-headed attention mechanism layer, residual connection and layer normalization, and a structure corresponding to the encoder structure.
6. The method of claim 1, wherein the position encoding module is configured to add a position encoding vector to the output data of the embedded layer, the encoding defined as follows:
where pos denotes the position of the current data in the sample, d denotes the dimension of the position-coding vector, and p denotes the position index of each value.
7. The method of claim 1, wherein for each input sequenceThe normalization process includes:
the mean and variance are calculated for the input sample data, expressed as:
normalization processing is performed according to the calculated mean and variance, expressed as:
the inverse normalization process is expressed as:
y j ＝σ X ⊙(y′ j -μ X )
wherein, the ". As indicated above, the dot product, y' j Representing the output of the decoder.
8. The method of claim 7, wherein the non-stationary attention mechanism layer performs the process of:
the input data is transformed according to the following formula:
wherein Q' = (Q-1 μ) Q T )/σ X ，K′＝(K-1μ K T )/σ X ，μ Q Mean value of Q in time dimension, mu K Representing the average value of K in the time dimension, Q being the query matrix and K being the key matrix;
the calculation of the de-stabilization factor and introduced into the attention mechanism is learned by using a plurality of sensor layers, expressed as:
logτ＝MLP(σ X ，X)，Δ＝MLP(μ X ，X)
wherein,V′＝(V-1μ V T )/σ X ，μ V representing the average value of V in the time dimension, V being the matrix of values, log tau being the output of the multi-layer perceptron.
9. A computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor realizes the steps of the method according to any of claims 1 to 8.
10. A computer device comprising a memory and a processor, on which memory a computer program is stored which can be run on the processor, characterized in that the processor implements the steps of the method according to any of claims 1 to 8 when the computer program is executed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310603115.9A CN117154680A (en) | 2023-05-25 | 2023-05-25 | Wind power prediction method based on non-stationary transducer model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310603115.9A CN117154680A (en) | 2023-05-25 | 2023-05-25 | Wind power prediction method based on non-stationary transducer model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117154680A true CN117154680A (en) | 2023-12-01 |
Family
ID=88910723
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310603115.9A Pending CN117154680A (en) | 2023-05-25 | 2023-05-25 | Wind power prediction method based on non-stationary transducer model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117154680A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117709082A (en) * | 2023-12-04 | 2024-03-15 | 江苏国电南自海吉科技有限公司 | Wind generating set power curve prediction method |
-
2023
- 2023-05-25 CN CN202310603115.9A patent/CN117154680A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117709082A (en) * | 2023-12-04 | 2024-03-15 | 江苏国电南自海吉科技有限公司 | Wind generating set power curve prediction method |
| CN117709082B (en) * | 2023-12-04 | 2024-06-18 | 江苏国电南自海吉科技有限公司 | Wind generating set power curve prediction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Short-term wind speed forecasting based on long short-term memory and improved BP neural network | |
| CN111784041B (en) | Wind power prediction method and system based on graph convolution neural network | |
| CN110163429B (en) | Short-term load prediction method based on similarity day optimization screening | |
| CN111242377B (en) | Short-term wind speed prediction method integrating deep learning and data denoising | |
| Ibrahim et al. | Short‐Time Wind Speed Forecast Using Artificial Learning‐Based Algorithms | |
| CN111091233A (en) | Wind power plant short-term wind power prediction modeling method based on wavelet analysis and multi-model AdaBoost depth network | |
| CN112257263B (en) | Equipment residual life prediction system based on self-attention mechanism | |
| Cao et al. | Multi-step wind power forecasting model using LSTM networks, similar time series and LightGBM | |
| Li et al. | Deep spatio-temporal wind power forecasting | |
| CN113822418A (en) | Wind power plant power prediction method, system, device and storage medium | |
| CN111160620A (en) | Short-term wind power prediction method based on end-to-end memory network | |
| CN115860177A (en) | Photovoltaic power generation power prediction method based on combined machine learning model and application thereof | |
| CN112925824A (en) | Photovoltaic power prediction method and system for extreme weather type | |
| CN114358389A (en) | Short-term power load prediction method combining VMD decomposition and time convolution network | |
| CN117154680A (en) | Wind power prediction method based on non-stationary transducer model | |
| CN117076931A (en) | Time sequence data prediction method and system based on conditional diffusion model | |
| CN114444821A (en) | Integrated learning load prediction method, system and medium for power internet of things | |
| CN116706907B (en) | Photovoltaic power generation prediction method based on fuzzy reasoning and related equipment | |
| CN116307130A (en) | Photovoltaic power generation power short-term prediction method based on improved LSTM combination model | |
| CN113011597B (en) | Deep learning method and device for regression task | |
| CN113962432A (en) | Wind power prediction method and system integrating three-dimensional convolution and light-weight convolution threshold unit | |
| CN116722548B (en) | Photovoltaic power generation prediction method based on time sequence model and related equipment | |
| CN118069717B (en) | Time sequence data characteristic prediction method based on cyclic neural network | |
| CN115860281B (en) | Multi-entity load prediction method and device for energy system based on cross-entity attention | |
| CN117688367B (en) | Wind power generation ultra-short term power prediction method and device based on instant learning |
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 |