A wind turbine ice adhesion prediction method based on time series analysis

Abstract

The present application relates to the technical field of ice adhesion prediction of wind driven generators, and discloses a time series analysis method for ice adhesion prediction of wind driven generators, which comprises the following steps: collecting time series data of ice adhesion indexes of wind driven generators, and performing index data weighting based on the weight of the ice adhesion indexes of wind driven generators; encoding the weighted time series data of the ice adhesion indexes of wind driven generators; optimizing the constructed time series analysis model of the ice adhesion of wind driven generators to obtain an optimal time series analysis model of the ice adhesion of wind driven generators, and obtaining the ice adhesion probability of the wind driven generators by using the optimal time series analysis model of the ice adhesion of wind driven generators. According to the information entropy of different indexes, the present application determines the benchmark index weight of the indexes, and corrects the benchmark index weight according to the correlation between the indexes, wherein the greater the weight of the index capable of effectively detecting the ice adhesion state, the higher the accuracy of subsequent ice adhesion probability prediction is improved, and the ice adhesion probability is calculated based on the continuously updated prior probability and combined with the cross-encoding result of the index data.

Description

Technical Field

[0001] This invention relates to the technical field of wind turbine icing prediction, and more particularly to a time-series analysis-based method for wind turbine icing prediction. Background Technology

[0002] Currently, wind power generation has become one of the main methods of electricity production in my country. It not only possesses strong clean and pollution-free characteristics but also effectively reduces production costs and improves resource and energy utilization. However, in some wind power generation areas, wind turbines gradually become icy over time, leading to a significant decrease in turbine capacity and affecting wind power generation efficiency and operational safety. To address this problem, this invention proposes a time-series analysis-based method for predicting wind turbine icing. By collecting multi-dimensional icing index data and performing time-series analysis, the method predicts the likelihood of wind turbine icing, providing early warning and protection to enhance the reliability and safety of wind turbine operation. Summary of the Invention

[0003] In view of this, the present invention provides a time-series analysis method for predicting icing in wind turbines. The purpose is to: 1) determine the baseline index weights of different indicators based on their information entropy. A higher baseline index weight indicates greater accuracy in classifying wind turbines into non-icing and icing states. Then, fix the indicator with the highest baseline index weight, calculate the correlation between other indicators and the fixed index data, and calibrate the baseline index weights of other indicators. A higher correlation indicates closer information entropy between the indicator and the fixed index, resulting in a smaller calibrated baseline index weight. Finally, assign weights to the collected time-series data of wind turbine icing indicators to obtain representations of different indicators. The more important the time series data of the indicators, the greater the weight of the indicators that can effectively detect the icing state, the higher the accuracy of subsequent icing prediction; 2) By encoding the time series data of each indicator separately, and combining the most obvious encoding results of other indicators to construct a convolution matrix, convolution encoding operation is performed to realize cross-encoding between different indicators. Combined with the prior probability distribution of the icing probability, a wind turbine icing time series analysis model is constructed. By iteratively solving the likelihood function of the probability distribution parameters and sample data, the probability distribution parameters are converged. Based on the continuously updated prior probability, combined with the cross-encoding results of the current indicator data, the icing probability is calculated to realize the wind turbine icing prediction combined with time series data.

[0004] To achieve the above objectives, the present invention provides a time-series analysis-based method for predicting icing in wind turbines, comprising the following steps:

[0005] S1: Collect time-series data of wind turbine icing index, and assign weights to the index data based on the weights of wind turbine icing index to obtain weighted time-series data of wind turbine icing index.

[0006] S2: Encode the weighted wind turbine icing index time series data to obtain the encoded index vector data;

[0007] S3: Construct a time series analysis model for icing of wind turbines. The model takes index vector data as input and icing probability as output.

[0008] S4: Optimize the constructed wind turbine icing time series analysis model to obtain the optimal wind turbine icing time series analysis model, and use the optimal wind turbine icing time series analysis model to obtain the icing probability of the wind turbine.

[0009] As a further improvement of the present invention:

[0010] Optionally, the time-series data of the icing index of the wind turbine in step S1 includes:

[0011] Time-series data on wind turbine icing indicators were collected, including wind turbine blade speed, generator active power, blade temperature, blade humidity, and blade images. The collected time-series data x for wind turbine icing indicators is as follows:

[0012] x = [x1, x2, x3, x4, x5] T

[0013] x i =(x i (t1),x i (t2),...,x i (t n ),...,x i (t N ),i∈[1,5]

[0014] in:

[0015] x i The time series data represents the icing index of the i-th type of wind turbine, where x1 represents the blade speed index, x2 represents the generator active power index, x3 represents the blade temperature index, x4 represents the blade humidity index, and x5 represents the blade image index.

[0016] x i (t n ) represents t n The time series data of the i-th type of wind turbine icing index collected at time t1 represents the initial time of data collection for the wind turbine icing index time series data. N This indicates the cutoff time for collecting time-series data on the icing index of wind turbines; the time interval between adjacent data collection times is Δt.

[0017] T stands for transpose.

[0018] In this embodiment of the invention, a temperature and humidity sensor is used to collect temperature and humidity data of the wind turbine blades, a camera is used to collect image data of the wind turbine blades, a speed sensor is used to collect speed data of the wind turbine blades, and the active power of the wind turbine is monitored in real time.

[0019] Optionally, determining the weight of the icing index for the wind turbine in step S1 includes:

[0020] The weights of different icing performance indicators for wind turbines are determined, and the process for determining the weights is as follows:

[0021] S11: Collect time-series data of wind turbine icing index at M time points. The time-series data of wind turbine icing index at each time point includes time-series data of five wind turbine icing indices. The wind turbine is in different icing states at the M time points collected, including non-icing state and icing state.

[0022] S12: Calculate the information entropy H of the time-series data of wind turbine icing index at M time points:

[0023]

[0024] in:

[0025] M0 represents the number of times the wind turbine was in an icing-free state, and M1 represents the number of times the wind turbine was in an icing state.

[0026] S13: Use the Canny edge detection algorithm to extract the image edges of the binarized wind blade image, and calculate the cosine similarity of any image edge. Mark the wind blade images with a cosine similarity higher than the similarity threshold as the same wind blade image and encode them. The encoding representation results of the same wind blade image are the same.

[0027] S14: Calculate the baseline index weight w for any i-th type of wind turbine icing index. i :

[0028]

[0029] in:

[0030] Ω i Let num represent the set of unique, repeating data items for the i-th type of wind turbine icing index in the time series data of wind turbine icing index collected at M time points. k(i) represents the number of times data k appears in the time series data of the wind turbine icing index at M time points for the i-th type of wind turbine icing index. For the blade image index, the data in the dataset is the encoded representation of the blade image. In this embodiment of the invention, Ω i The dataset retains unique data and unique items of duplicate data. For wind blade image indicators, the dataset retains the encoded representation of wind blade images.

[0031] S15: Select the i-th benchmark index with the highest weight. * Based on the time-series data of the icing index of wind turbines collected at M time points, the corresponding data of other indicators are calculated and compared with the data of the i-th time point. * The correlation between the data corresponding to the various indicators, where the data corresponding to the j-th indicator is related to the data of the i-th indicator. * The correlation between the data corresponding to the various indicators is sim(i * ,j), j,i * ∈[1,5],j≠i * ;

[0032] S16: Update the weight of the j-th indicator, where the update formula is:

[0033] w j ←(1-sim(i * ,j))w j

[0034] S17: Normalize the weights of all updated indicators to obtain the normalized weights of each indicator, where the normalized weight of the i-th indicator is...

[0035] Optionally, in step S1, the index data is weighted based on the icing index weight of the wind turbine, including:

[0036] The index data is weighted based on the icing index of wind turbines. The index data weighting process is as follows:

[0037] The collected data on wind turbine blade speed, generator active power, wind turbine blade temperature, and wind turbine blade humidity were standardized. The standardization formula is as follows:

[0038]

[0039] in:

[0040] x represents r (t n The standardized result of ), r∈[1,4];

[0041] The standardized data is then weighted using the following formula:

[0042]

[0043] in:

[0044] x ′ r (t n )express The result of the empowerment;

[0045] The wind turbine blade image is then converted to grayscale to obtain a grayscale wind turbine blade image, where t n Grayscale image of wind turbine blades at any given moment And on Perform weighting and binarization:

[0046]

[0047] in:

[0048] C(·) represents binarization;

[0049] The weighted time series data of wind turbine icing index x ′ :

[0050] x ′ =[x ′ 1,x ′ 2,x ′ 3,x ′ 4,x ′ 5] T

[0051] x i ′ =(x i ′ (t1),x i ′ (t2),...,x i ′ (t n ),...,x i ′ (t N ),i∈[1,5]

[0052] in:

[0053] x i ′ This represents the weighted time-series data of the icing index of the i-th type of wind turbine.

[0054] Optionally, the S2 step involves encoding the weighted wind turbine icing index time-series data, including:

[0055] The weighted wind turbine icing index time series data are encoded, and the encoding process is as follows:

[0056] S21: Input the weighted wind turbine icing index time series data into the encoding module. The encoding module encodes the weighted wind turbine icing index time series data. The encoding formula for the time series data of the i-th index is:

[0057]

[0058]

[0059] in:

[0060] This represents the encoding matrix for the i-th index;

[0061] c i This represents the encoding and processing result of the time series data collected for the i-th indicator;

[0062] S22: Cross-encode the encoding results of multiple indicator data to obtain the encoded indicator vector data:

[0063] X=[Max(1)*c1,Max(2)*c2,Max(3)*c3,Max(4)*c4,Max(5)*c5] T

[0064] Max(1)=[max(c2),max(c3),max(c4),max(c5)]

[0065] Max(2)=[max(c1),max(c3),max(c4),max(c5)]

[0066] Max(3)=[max(c1),max(c2),max(c4),max(c5)]

[0067] Max(4)=[max(c1),max(c2),max(c3),max(c5)]

[0068] Max(5)=[max(c1),max(c2),max(c3),max(c4)]

[0069] in:

[0070] * indicates a convolution operation;

[0071] max(·) represents selecting the maximum value;

[0072] X represents the encoded indicator vector data.

[0073] Optionally, the construction of the icing time series analysis model for the wind turbine in step S3 includes:

[0074] A wind turbine icing time series analysis model is constructed. The model takes index vector data as input and icing probability as output. The wind turbine icing time series analysis model includes a prior probability layer, an input layer, a probability calculation layer and an output layer.

[0075] The prior probability layer is used to obtain the prior probability of the icing prediction result, the input layer is used to receive index vector data, the probability calculation layer is used to calculate the icing probability of the wind turbine and use it as the output, and the prior probability is updated according to the output result.

[0076] Optionally, in step S4, the constructed wind turbine icing time series analysis model is optimized to obtain the optimal wind turbine icing time series analysis model, including:

[0077] The optimal wind turbine icing time series analysis model was obtained by optimizing the constructed model. The model optimization process is as follows:

[0078] S41: Initialize the probability distribution parameters θ(0) of the wind turbine icing time series analysis model:

[0079] θ(0)=(θ1(0),θ2(0),θ3(0),θ4(0),θ5(0))

[0080] in:

[0081] θ i (0) represents the probability distribution parameter of the i-th type of wind turbine icing index generated during initialization;

[0082] S42: Obtain Q sets of independent indicator vector data, and initialize and generate the icing probability of each set of indicator vector data to form the training set data for the wind turbine icing time series analysis model:

[0083] data={(X q ,p q )|q∈[1,Q]}

[0084] in:

[0085] X q Let p represent the index vector data of the qth group. q This represents the index vector data X generated during initialization. q The icing probability; the icing probability corresponding to the Q group index vector data follows a prior probability distribution p; in this embodiment of the invention, the prior probability distribution p represents the probability distribution of different icing probabilities of the wind turbine in a real environment;

[0086] S43: Set the number of optimizations for the current model to d, with an initial value of 0. Then the probability distribution parameter obtained in the d-th optimization is θ(d).

[0087] S44: Calculate the log-likelihood function of the training data in the training set:

[0088]

[0089] in:

[0090] z represents the probability of stepping on ice following a prior probability distribution p;

[0091] P(z|X q ,θ(d)) represents X q When input into the wind turbine icing time series analysis model based on the probability distribution parameter θ(d), the model outputs the probability that the icing probability conforms to the prior probability distribution p.

[0092] P(X q Z|θ(d)) represents the wind turbine icing time series analysis model based on the probability distribution parameter θ(d), with inputs and outputs X and X respectively. q The probability of z, where z follows a prior probability distribution p;

[0093] S45: Update the probability distribution parameters θ(d+1) obtained from the (d+1)th optimization:

[0094]

[0095] in:

[0096] θ(d+1) represents the probability distribution parameter that maximizes L(p,θ(d));

[0097] S46: If θ(d+1) satisfies the following equation, then stop training and optimization, and construct the optimal wind turbine icing time series analysis model based on θ(d+1):

[0098]

[0099] in:

[0100] ||·|| represents the L1 norm, and ||·||2 represents the L2 norm;

[0101] ε represents the preset threshold for parameter changes.

[0102] Optionally, step S4 involves using an optimal wind turbine icing time series analysis model to obtain the icing probability of the wind turbine, including:

[0103] The icing probability of a wind turbine is obtained using an optimal wind turbine icing time series analysis model. The calculation process for the icing probability is as follows:

[0104] Obtain the prior probability of the current ice-walking prediction result:

[0105]

[0106]

[0107] in:

[0108] F represents the preset baseline number of times, F0 represents the preset number of times the wind turbine is in a non-icing state, and F1 represents the preset number of times the wind turbine is in an icing state; in this embodiment of the invention, F is set to 100, and F0 and F1 are set to 60 and 40 respectively.

[0109] α represents the total number of times the current icing probability is calculated in the optimal wind turbine icing time series analysis model, α0 represents the number of times the icing probability is lower than the probability threshold, and α1 represents the number of times the icing probability is greater than or equal to the probability threshold.

[0110] P0 represents the probability that the wind turbine is in a non-icing state, and P1 represents the probability that the wind turbine is in an icing state.

[0111] Receive indicator vector data X=[Max(1)*c1,Max(2)*c2,Max(3)*c3,Max(4)*c4,Max(5)*

[0112] c5]T;

[0113] The probability of walking on ice, P(X; θ), is calculated as follows:

[0114]

[0115] in:

[0116] exp(·) denotes an exponent with the natural constant as the base;

[0117] θ = (θ1, θ2, θ3, θ4, θ5) represents the probability distribution parameters of the optimal wind turbine icing time series analysis model;

[0118] Update the prior probability based on the ice-walking probability P(X; θ);

[0119] If the probability of icing P(X; θ) is less than the probability threshold, it means that the wind turbine is in a non-icing state. If the probability of icing P(X; θ) is greater than or equal to the probability threshold, it means that the wind turbine is in an icing state.

[0120] To address the above problems, the present invention provides an electronic device, the electronic device comprising:

[0121] Memory, storing at least one instruction;

[0122] Communication interfaces enable communication between electronic devices; and

[0123] The processor executes the instructions stored in the memory to implement the wind turbine icing prediction method based on the time-series analysis described above.

[0124] To address the aforementioned problems, the present invention also provides a computer-readable storage medium storing at least one instruction, which is executed by a processor in an electronic device to implement the wind turbine icing prediction method described above based on time-series analysis.

[0125] Compared with existing technologies, this invention proposes a time-series analysis-based method for predicting icing on wind turbines, which has the following advantages:

[0126] First, this scheme proposes a multi-index weighting method. It collects time-series data on wind turbine icing indicators at M time points, where each time point contains time-series data on five different wind turbine icing indicators. The wind turbines are in different icing states at the M time points, including non-icing and icing states. The information entropy H of the time-series data on wind turbine icing indicators at the M time points is calculated.

[0127]

[0128] Where: M0 represents the number of times the wind turbine is in an icing-free state, and M1 represents the number of times the wind turbine is in an icing state; the Canny edge detection algorithm is used to extract the image edges of the binarized wind turbine blade images, and the cosine similarity of any image edge is calculated. Wind turbine blade images with a cosine similarity higher than the similarity threshold are marked as the same wind turbine blade image and encoded accordingly. The encoding representation of the same wind turbine blade image is identical; the baseline index weight w for any i-th type of icing index of the wind turbine is calculated. i :

[0129]

[0130] Where: Ω i Let num represent the set of unique, repeating data items for the i-th type of wind turbine icing index in the time series data of wind turbine icing index collected at M time points. k (i) represents the number of times data k appears in the time series data of the wind turbine icing index at M time points for the i-th type of wind turbine icing index. For the blade image index, the data in the dataset is the encoded representation of the blade image. The i-th index with the highest weight is selected as the benchmark index.* Based on the time-series data of the icing index of wind turbines collected at M time points, the corresponding data of other indicators are calculated and compared with the data of the i-th time point. * The correlation between the data corresponding to the various indicators, where the data corresponding to the j-th indicator is related to the data of the i-th indicator. * The correlation between the data corresponding to the various indicators is sim(i * ,j), j,i * ∈[1,5],j≠i * Update the weight of the j-th indicator, where the update formula is:

[0131] w j ←(1-sim(i * ,j))w j

[0132] The updated weights of all indicators are normalized to obtain the normalized weights of each indicator, where the normalized weight of the i-th indicator is... This scheme determines the baseline indicator weights based on the information entropy of different indicators. The larger the baseline indicator weight, the higher the accuracy of using that indicator to classify wind turbines into non-iced and iced states. Then, the indicator with the largest baseline indicator weight is fixed, and the correlation between other indicators and the fixed indicator data is calculated. The baseline indicator weights of other indicators are calibrated; the greater the correlation, the closer the information entropy contained in the indicator is to the fixed indicator, and the smaller the calibrated baseline indicator weight. The collected time-series data of wind turbine icing indicators are then weighted to obtain time-series data representing the importance of different indicators. The indicators that can effectively detect icing states have larger weights, improving the accuracy of subsequent icing predictions.

[0133] Meanwhile, this solution proposes a method for calculating the icing probability by combining time-series data. The optimal wind turbine icing time-series analysis model is used to obtain the icing probability of the wind turbine. The calculation process for the icing probability is as follows: Obtain the prior probability of the current icing prediction result:

[0134]

[0135]

[0136] Where: F represents the preset baseline number of times, F0 represents the preset number of times the wind turbine is in a non-icing state, and F1 represents the preset number of times the wind turbine is in an icing state; α represents the total number of times the current icing probability calculation of the optimal wind turbine icing time sequence analysis model is performed, α0 represents the number of times the icing probability is lower than the probability threshold, and α1 represents the number of times the icing probability is greater than or equal to the probability threshold; P0 represents the probability that the wind turbine is in a non-icing state, and P1 represents the probability that the wind turbine is in an icing state; Receive index vector data X = [Max(1)*c1,Max(2)*c2,Max(3)*c3,Max(4)*c4,Max(5)*c5]T; Calculate the icing probability P(X; θ):

[0137]

[0138] Where: exp(·) represents the exponent with the natural constant as the base; θ=(θ1,θ2,θ3,θ4,θ5) represents the probability distribution parameters of the optimal wind turbine icing time series analysis model; the prior probability is updated according to the icing probability P(X; θ). This scheme encodes the time series data of each indicator separately, and constructs a convolution matrix by combining the most obvious encoding results of other indicators, performs convolution encoding operations to achieve cross-encoding between different indicators, and constructs a wind turbine icing time series analysis model by combining the prior probability distribution of the icing probability. By iteratively solving the likelihood function of the probability distribution parameters and sample data, the probability distribution parameters are converged. Based on the continuously updated prior probability, the icing probability is calculated by combining the cross-encoding results of the current indicator data, realizing the prediction of wind turbine icing by combining time series data. Attached Figure Description

[0139] Figure 1 A flowchart illustrating a time-series analysis-based method for predicting icing in wind turbines, as provided in an embodiment of the present invention.

[0140] Figure 2 This is a schematic diagram of the electronic device for predicting icing of wind turbines using time-series analysis, as provided in an embodiment of the present invention.

[0141] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0142] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0143] This application provides a time-series analysis-based method for predicting icing in wind turbines. The execution entity of this time-series analysis-based method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the time-series analysis-based method for predicting icing in wind turbines can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0144] Example 1:

[0145] S1: Collect time-series data of wind turbine icing index, and assign weights to the index data based on the weights of the wind turbine icing index to obtain weighted time-series data of wind turbine icing index.

[0146] The time-series data of wind turbine icing index collected in step S1 includes:

[0147] Time-series data on wind turbine icing indicators were collected, including wind turbine blade speed, generator active power, blade temperature, blade humidity, and blade images. The collected time-series data x for wind turbine icing indicators is as follows:

[0148] x = [x1, x2, x3, x4, x5] T

[0149] x i =(x i (t1),x i (t2),...,x i (t n ),...,x i (t N ),i∈[1,5]

[0150] in:

[0151] x i The time series data represents the icing index of the i-th type of wind turbine, where x1 represents the blade speed index, x2 represents the generator active power index, x3 represents the blade temperature index, x4 represents the blade humidity index, and x5 represents the blade image index.

[0152] x i (t n ) represents t n The time series data of the i-th type of wind turbine icing index collected at time t1 represents the initial time of data collection for the wind turbine icing index time series data. NThis indicates the cutoff time for collecting time-series data on the icing index of wind turbines; the time interval between adjacent data collection times is Δt.

[0153] T stands for transpose.

[0154] The determination of the icing index weights for wind turbines in step S1 includes:

[0155] The weights of different icing performance indicators for wind turbines are determined, and the process for determining the weights is as follows:

[0156] S11: Collect time-series data of wind turbine icing index at M time points. The time-series data of wind turbine icing index at each time point includes time-series data of five wind turbine icing indices. The wind turbine is in different icing states at the M time points collected, including non-icing state and icing state.

[0157] S12: Calculate the information entropy H of the time-series data of wind turbine icing index at M time points:

[0158]

[0159] in:

[0160] M0 represents the number of times the wind turbine was in an icing-free state, and M1 represents the number of times the wind turbine was in an icing state.

[0161] S13: Use the Canny edge detection algorithm to extract the image edges of the binarized wind blade image, and calculate the cosine similarity of any image edge. Mark the wind blade images with a cosine similarity higher than the similarity threshold as the same wind blade image and encode them. The encoding representation results of the same wind blade image are the same.

[0162] S14: Calculate the baseline index weight w for any i-th type of wind turbine icing index. i :

[0163]

[0164] in:

[0165] Ω i Let num represent the set of unique, repeating data items for the i-th type of wind turbine icing index in the time series data of wind turbine icing index collected at M time points. k (i) represents the number of times data k appears in the time series data of the wind turbine icing index at M time points for the i-th type of wind turbine icing index. For the blade image index, the data in the dataset is the encoded representation of the blade image. In this embodiment of the invention, Ω iThe dataset retains unique data and unique items of duplicate data. For wind blade image indicators, the dataset retains the encoded representation of wind blade images.

[0166] S15: Select the i-th benchmark index with the highest weight. * Based on the time-series data of the icing index of wind turbines collected at M time points, the corresponding data of other indicators are calculated and compared with the data of the i-th time point. * The correlation between the data corresponding to the various indicators, where the data corresponding to the j-th indicator is related to the data of the i-th indicator. * The correlation between the data corresponding to the various indicators is sim(i * ,j), j,i * ∈[1,5],j≠i * ;

[0167] S16: Update the weight of the j-th indicator, where the update formula is:

[0168] w j ←(1-sim(i * ,j))w j

[0169] S17: Normalize the weights of all updated indicators to obtain the normalized weights of each indicator, where the normalized weight of the i-th indicator is...

[0170] Step S1 involves assigning weights to the index data based on the icing index weights of the wind turbine, including:

[0171] The index data is weighted based on the icing index of wind turbines. The index data weighting process is as follows:

[0172] The collected data on wind turbine blade speed, generator active power, wind turbine blade temperature, and wind turbine blade humidity were standardized. The standardization formula is as follows:

[0173]

[0174] in:

[0175] x represents r (t n The standardized result of ), r∈[1,4];

[0176] The standardized data is then weighted using the following formula:

[0177]

[0178] in:

[0179] x ′r (t n )express The result of the empowerment;

[0180] The wind turbine blade image is then converted to grayscale to obtain a grayscale wind turbine blade image, where t n Grayscale image of wind turbine blades at any given moment And on Perform weighting and binarization:

[0181]

[0182] in:

[0183] C(·) represents binarization;

[0184] The weighted time series data of wind turbine icing index x ′ :

[0185] x ′ =[x ′ 1,x ′ 2,x ′ 3,x ′ 4,x ′ 5] T

[0186] x i ′ =(x i ′ (t1),x i ′ (t2),...,x i ′ (t n ),...,x i ′ (t N ),i∈[1,5]

[0187] in:

[0188] x i ′ This represents the weighted time-series data of the icing index of the i-th type of wind turbine.

[0189] S2: Encode the weighted wind turbine icing index time series data to obtain the encoded index vector data.

[0190] The S2 step involves encoding the weighted wind turbine icing index time-series data, including:

[0191] The weighted wind turbine icing index time series data are encoded, and the encoding process is as follows:

[0192] S21: Input the weighted wind turbine icing index time series data into the encoding module. The encoding module encodes the weighted wind turbine icing index time series data. The encoding formula for the time series data of the i-th index is:

[0193]

[0194]

[0195] in:

[0196] This represents the encoding matrix for the i-th index;

[0197] c i This represents the encoding and processing result of the time series data collected for the i-th indicator;

[0198] S22: Cross-encode the encoding results of multiple indicator data to obtain the encoded indicator vector data:

[0199] X=[Max(1)*c1,Max(2)*c2,Max(3)*c3,Max(4)*c4,Max(5)*c5] T

[0200] Max(1)=[max(c2),max(c3),max(c4),max(c5)]

[0201] Max(2)=[max(c1),max(c3),max(c4),max(c5)]

[0202] Max(3)=[max(c1),max(c2),max(c4),max(c5)]

[0203] Max(4)=[max(c1),max(c2),max(c3),max(c5)]

[0204] Max(5)=[max(c1),max(c2),max(c3),max(c4)]

[0205] in:

[0206] * indicates a convolution operation;

[0207] max(·) represents selecting the maximum value;

[0208] X represents the encoded indicator vector data.

[0209] S3: Construct a time series analysis model for icing of wind turbines. The model takes index vector data as input and icing probability as output.

[0210] The S3 step involves constructing a wind turbine icing time-series analysis model, including:

[0211] A wind turbine icing time series analysis model is constructed. The model takes index vector data as input and icing probability as output. The wind turbine icing time series analysis model includes a prior probability layer, an input layer, a probability calculation layer and an output layer.

[0212] The prior probability layer is used to obtain the prior probability of the icing prediction result, the input layer is used to receive index vector data, the probability calculation layer is used to calculate the icing probability of the wind turbine and use it as the output, and the prior probability is updated according to the output result.

[0213] S4: Optimize the constructed wind turbine icing time series analysis model to obtain the optimal wind turbine icing time series analysis model, and use the optimal wind turbine icing time series analysis model to obtain the icing probability of the wind turbine.

[0214] In step S4, the constructed wind turbine icing time series analysis model is optimized to obtain the optimal wind turbine icing time series analysis model, including:

[0215] The optimal wind turbine icing time series analysis model was obtained by optimizing the constructed model. The model optimization process is as follows:

[0216] S41: Initialize the probability distribution parameters θ(0) of the wind turbine icing time series analysis model:

[0217] θ(0)=(θ1(0),θ2(0),θ3(0),θ4(0),θ5(0))

[0218] in:

[0219] θ i (0) represents the probability distribution parameter of the i-th type of wind turbine icing index generated during initialization;

[0220] S42: Obtain Q sets of independent indicator vector data, and initialize and generate the icing probability of each set of indicator vector data to form the training set data for the wind turbine icing time series analysis model:

[0221] data={(X q ,p q )|q∈[1,Q]}

[0222] in:

[0223] Xq Let p represent the index vector data of the qth group. q This represents the index vector data X generated during initialization. q The icing probability; the icing probability corresponding to the Q group index vector data follows a prior probability distribution p; in this embodiment of the invention, the prior probability distribution p represents the probability distribution of different icing probabilities of the wind turbine in a real environment;

[0224] S43: Set the number of optimizations for the current model to d, with an initial value of 0. Then the probability distribution parameter obtained in the d-th optimization is θ(d).

[0225] S44: Calculate the log-likelihood function of the training data in the training set:

[0226]

[0227] in:

[0228] z represents the probability of stepping on ice following a prior probability distribution p;

[0229] P(z|X q ,θ(d)) represents X q When input into the wind turbine icing time series analysis model based on the probability distribution parameter θ(d), the model outputs the probability that the icing probability conforms to the prior probability distribution p.

[0230] P(X q Z|θ(d)) represents the wind turbine icing time series analysis model based on the probability distribution parameter θ(d), with inputs and outputs X and X respectively. q The probability of z, where z follows a prior probability distribution p;

[0231] S45: Update the probability distribution parameters θ(d+1) obtained from the (d+1)th optimization:

[0232]

[0233] in:

[0234] θ(d+1) represents the probability distribution parameter that maximizes L(p,θ(d));

[0235] S46: If θ(d+1) satisfies the following equation, then stop training and optimization, and construct the optimal wind turbine icing time series analysis model based on θ(d+1):

[0236]

[0237] in:

[0238] ||·|| represents the L1 norm, and ||·||2 represents the L2 norm;

[0239] ε represents the preset threshold for parameter changes.

[0240] Step S4 utilizes the optimal wind turbine icing time series analysis model to obtain the icing probability of the wind turbine, including:

[0241] The icing probability of a wind turbine is obtained using an optimal wind turbine icing time series analysis model. The calculation process for the icing probability is as follows:

[0242] Obtain the prior probability of the current ice-walking prediction result:

[0243]

[0244]

[0245] in:

[0246] F represents the preset baseline number of times, F0 represents the preset number of times the wind turbine is in a non-icing state, and F1 represents the preset number of times the wind turbine is in an icing state; in this embodiment of the invention, F is set to 100, and F0 and F1 are set to 60 and 40 respectively.

[0247] α represents the total number of times the current icing probability is calculated in the optimal wind turbine icing time series analysis model, α0 represents the number of times the icing probability is lower than the probability threshold, and α1 represents the number of times the icing probability is greater than or equal to the probability threshold.

[0248] P0 represents the probability that the wind turbine is in a non-icing state, and P1 represents the probability that the wind turbine is in an icing state.

[0249] Receive indicator vector data X=[Max(1)*c1,Max(2)*c2,Max(3)*c3,Max(4)*c4,Max(5)*

[0250] c5]T;

[0251] The probability of walking on ice, P(X; θ), is calculated as follows:

[0252]

[0253] in:

[0254] exp(·) denotes an exponent with the natural constant as the base;

[0255] θ = (θ1, θ2, θ3, θ4, θ5) represents the probability distribution parameters of the optimal wind turbine icing time series analysis model;

[0256] Update the prior probability based on the ice-walking probability P(X; θ);

[0257] If the probability of icing P(X; θ) is less than the probability threshold, it means that the wind turbine is in a non-icing state. If the probability of icing P(X; θ) is greater than or equal to the probability threshold, it means that the wind turbine is in an icing state.

[0258] Example 2:

[0259] like Figure 2 The diagram shown is a schematic representation of the electronic device for predicting icing of wind turbines using time-series analysis, according to an embodiment of the present invention.

[0260] The electronic device 1 may include a processor 10, a memory 11, a communication interface 13 and a bus, and may also include a computer program, such as program 12, stored in the memory 11 and executable on the processor 10.

[0261] The memory 11 includes at least one type of readable storage medium, such as flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the electronic device 1, such as a portable hard drive. In other embodiments, the memory 11 can be an external storage device of the electronic device 1, such as a plug-in portable hard drive, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device 1. Furthermore, the memory 11 can include both internal and external storage units of the electronic device 1. The memory 11 can be used not only to store application software and various types of data installed on the electronic device 1, such as the code of program 12, but also to temporarily store data that has been output or will be output.

[0262] In some embodiments, the processor 10 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory 11 (such as program 12 for wind turbine icing prediction to implement time-series analysis) and calls data stored in the memory 11 to perform various functions of the electronic device 1 and process data.

[0263] The communication interface 13 may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, a Bluetooth interface, etc.), which is typically used to establish communication connections between the electronic device 1 and other electronic devices, and to enable communication between internal components of the electronic device.

[0264] The bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. The bus is configured to enable communication between the memory 11 and at least one processor 10, etc.

[0265] Figure 2 Only electronic devices with components are shown; it will be understood by those skilled in the art that... Figure 2 The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0266] For example, although not shown, the electronic device 1 may also include a power supply (such as a battery) to power various components. Preferably, the power supply can be logically connected to the at least one processor 10 through a power management device, thereby enabling functions such as charging management, discharging management, and power consumption management. The power supply may also include one or more DC or AC power supplies, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device 1 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0267] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), or a standard wired or wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device 1 and to display a visual user interface.

[0268] It should be understood that the embodiments described are for illustrative purposes only and are not limited to this structure in the scope of the patent application.

[0269] The program 12 stored in the memory 11 of the electronic device 1 is a combination of multiple instructions, which, when run in the processor 10, can achieve the following:

[0270] Collect time-series data of icing index of wind turbines, and assign weights to the index data based on the weights of the icing index of wind turbines to obtain weighted time-series data of icing index of wind turbines.

[0271] The weighted time-series data of the icing index of the wind turbine were encoded to obtain the encoded index vector data.

[0272] Constructing a time-series analysis model for wind turbine icing;

[0273] The constructed wind turbine icing time series analysis model is optimized to obtain the optimal wind turbine icing time series analysis model, and the icing probability of the wind turbine is obtained by using the optimal wind turbine icing time series analysis model.

[0274] Specifically, the processor 10's implementation method for the above instructions can be found in [reference needed]. Figures 1 to 2 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.

[0275] It should be noted that the sequence numbers of the above embodiments of the present invention are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0276] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0277] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A time-series analysis-based method for predicting icing on wind turbines, characterized in that, The method includes: S1: Collect time-series data of wind turbine icing index, and assign weights to the index data based on the weights of wind turbine icing index to obtain weighted time-series data of wind turbine icing index. Time-series data on wind turbine icing indicators were collected, including wind turbine blade speed, generator active power, blade temperature, blade humidity, and blade images. The collected time-series data x for wind turbine icing indicators is as follows: in: This represents the time-series data of the icing index of the i-th type of wind turbine. This indicates the speed of the wind turbine blades. This indicates the active power index of the generator. This indicates the temperature index of the wind turbine blades. This indicates the humidity level of the wind turbine blades. Indicates wind turbine blade image indicators; express The time-series data of the icing index of the i-th wind turbine collected at time t. This indicates the initial time of data collection for the icing index of the wind turbine. This indicates the cutoff time for collecting time-series data on the icing index of wind turbines; the time interval between adjacent data collection times is... ; T represents transpose; The weights of different icing performance indicators for wind turbines are determined, and the process for determining the weights is as follows: S11: Collect time-series data of wind turbine icing index at M time points. The time-series data of wind turbine icing index at each time point includes time-series data of five wind turbine icing indices. The wind turbine is in different icing states at the M time points collected, including non-icing state and icing state. S12: Calculate the information entropy H of the time-series data of wind turbine icing index at M time points: in: This indicates the number of times the wind turbine was in an ice-free state. This indicates the number of moments during which the wind turbine is in an icing state. S13: Use the Canny edge detection algorithm to extract the image edges of the binarized wind blade image, and calculate the cosine similarity of any image edge. Mark the wind blade images with a cosine similarity higher than the similarity threshold as the same wind blade image and encode them. The encoding representation results of the same wind blade image are the same. S14: Calculate the baseline index weights for any i-th type of wind turbine icing index. : in: This represents the set of data containing unique, repeating data for the i-th type of wind turbine icing index in the time-series data of wind turbine icing index collected at M time points. This represents the number of times data k appears in the time series data of the wind turbine icing index at M time points for the i-th type of wind turbine icing index. For the blade image index, the data in the dataset is the encoded representation of the blade image. S15: Select the benchmark indicator with the highest weight. Based on the time-series data of the wind turbine icing index collected at M time points, calculate the corresponding data of other indicators and compare them with the first indicator. The correlation between the data corresponding to the j-th indicator and the data corresponding to the j-th indicator. The correlation between the data corresponding to the various indicators is , , ; S16: Update the weight of the j-th indicator, where the update formula is: S17: Normalize the weights of all updated indicators to obtain the normalized weights of each indicator, where the normalized weight of the i-th indicator is... ; S2: Encode the weighted wind turbine icing index time series data to obtain the encoded index vector data; S3: Construct a time series analysis model for icing of wind turbines. The model takes index vector data as input and icing probability as output. S4: Optimize the constructed wind turbine icing time series analysis model to obtain the optimal wind turbine icing time series analysis model, and use the optimal wind turbine icing time series analysis model to obtain the icing probability of the wind turbine.

2. The wind turbine icing prediction method based on time series analysis as described in claim 1, characterized in that, Step S1 involves assigning weights to the index data based on the icing index weights of the wind turbine, including: The index data is weighted based on the icing index of wind turbines. The index data weighting process is as follows: The collected data on wind turbine blade speed, generator active power, wind turbine blade temperature, and wind turbine blade humidity were standardized. The standardization formula is as follows: in: express The standardized processing results ; The standardized data is then weighted using the following formula: in: express The result of the empowerment; The wind turbine blade image was then converted to grayscale to obtain a grayscale wind turbine blade image. Grayscale image of wind turbine blades at any given moment and to Perform weighting and binarization: in: This indicates binarization processing; Time series data of wind turbine icing index after weighting : in: This represents the weighted time-series data of the icing index of the i-th type of wind turbine.

3. The wind turbine icing prediction method based on time series analysis as described in claim 2, characterized in that, The S2 step involves encoding the weighted wind turbine icing index time-series data, including: The weighted wind turbine icing index time series data are encoded, and the encoding process is as follows: S21: Input the weighted wind turbine icing index time series data into the encoding module. The encoding module encodes the weighted wind turbine icing index time series data. The encoding formula for the time series data of the i-th index is: in: This represents the encoding matrix for the i-th index; This represents the encoding and processing result of the time series data collected for the i-th indicator; S22: Cross-encode the encoding results of multiple indicator data to obtain the encoded indicator vector data: in: Indicates the convolution operation; This indicates that the maximum value is selected; This represents the encoded indicator vector data.

4. The wind turbine icing prediction method based on time series analysis as described in claim 1, characterized in that, The S3 step involves constructing a wind turbine icing time-series analysis model, including: A wind turbine icing time series analysis model is constructed. The model takes index vector data as input and icing probability as output. The wind turbine icing time series analysis model includes a prior probability layer, an input layer, a probability calculation layer and an output layer. The prior probability layer is used to obtain the prior probability of the icing prediction result, the input layer is used to receive index vector data, the probability calculation layer is used to calculate the icing probability of the wind turbine and use it as the output, and the prior probability is updated according to the output result.

5. The wind turbine icing prediction method based on time series analysis as described in claim 4, characterized in that, In step S4, the constructed wind turbine icing time series analysis model is optimized to obtain the optimal wind turbine icing time series analysis model, including: The optimal wind turbine icing time series analysis model was obtained by optimizing the constructed model. The model optimization process is as follows: S41: Initialize the probability distribution parameters of the wind turbine icing time series analysis model : in: The probability distribution parameter represents the i-th type of wind turbine icing index generated during initialization; S42: Obtain Q sets of independent indicator vector data, and initialize and generate the icing probability of each set of indicator vector data to form the training set data for the wind turbine icing time series analysis model: in: This represents the q-th group of indicator vector data. This represents the index vector data generated during initialization. The probability of walking on ice; the probability of walking on ice corresponding to the Q-group index vector data follows a prior probability distribution p; S43: Set the current model's optimization iteration number to d, with an initial value of 0. Then, the probability distribution parameters obtained from the d-th optimization are: ; S44: Calculate the log-likelihood function of the training data in the training set: in: z represents the probability of stepping on ice following a prior probability distribution p; Indicates will Input to probability distribution parameters In the icing time series analysis model of the wind turbine, the model outputs the probability that the icing probability conforms to the prior probability distribution p. Indicates based on probability distribution parameters The wind turbine icing time series analysis model has the following inputs and outputs: The probability of z follows a prior probability distribution p; S45: Update the probability distribution parameters obtained from the (d+1)th optimization. : in: Indicates that To achieve the maximum probability distribution parameter; S46: If If the following expression is satisfied, then training optimization stops, and based on... The optimal icing time series analysis model for wind turbines was constructed as follows: in: Describing the L1 norm, Represents the L2 norm; This indicates the preset threshold for parameter changes.

6. The wind turbine icing prediction method based on time series analysis as described in claim 1, characterized in that, Step S4 utilizes the optimal wind turbine icing time series analysis model to obtain the icing probability of the wind turbine, including: The icing probability of a wind turbine is obtained using an optimal wind turbine icing time series analysis model. The calculation process for the icing probability is as follows: Obtain the prior probability of the current ice-walking prediction result: in: This indicates the preset baseline number of times. This indicates the preset number of times the wind turbine is in an ice-free state. This indicates the preset number of times the wind turbine is in an icing state; This represents the total number of calculations for the current icing probability in the optimal wind turbine icing time series analysis model. This indicates the number of times the probability of walking on ice is below a probability threshold. Indicates the probability of walking on ice. The number of times the probability threshold is reached; This represents the probability that the wind turbine is in a non-icing state. This represents the probability that a wind turbine is in an icing state. Receive indicator vector data ; in: This represents the encoding and processing result of the time series data collected for the i-th indicator; This indicates that the maximum value is selected; The probability of walking on ice was calculated. : in: Represents an exponent with the natural constant as its base; The probability distribution parameters represent the optimal wind turbine icing time series analysis model; Based on the probability of ice walking Update prior probabilities; If the probability of walking on ice If the probability threshold is less than 1, it indicates that the wind turbine is in a non-icing state; if the icing probability is less than 1, it indicates that the wind turbine is in a non-icing state. The probability threshold indicates that the wind turbine is in an icing state.

Images