A wind turbine multi-source data fusion and power prediction method

Abstract

This invention discloses a method for multi-source data fusion and power prediction of wind turbine units, belonging to the field of power prediction technology. It addresses the technical problems of poor multi-source data fusion and power prediction analysis in existing solutions. By filtering and retaining key features through mutual information, redundant information can be effectively reduced. Dynamic weighting via an attention mechanism makes the fused features more adaptable to real-time operating conditions. Differential weight adjustments for wind speed, terrain, and equipment status ensure high relevance of the fused features even in complex scenarios. Spatial feature extraction captures local spatial correlations, temporal features capture temporal dependencies, and random forests handle nonlinear mappings, solving the problem of insufficient generalization ability of single models. The attention mechanism automatically assigns weights to different time steps and features, avoiding complex mathematical processes such as matrix operations and noise covariance estimation.

Description

Technical Field

[0001] This invention relates to the field of wind turbine power prediction technology, specifically to a method for multi-source data fusion and power prediction of wind turbines. Background Technology

[0002] Multi-source data fusion and power prediction for wind turbines are key technologies in the intelligent operation and management of modern wind farms. They aim to improve the accuracy of wind power generation prediction, thereby optimizing grid dispatch, enhancing wind power absorption capacity, and improving operational economy.

[0003] Existing technical solutions, when implemented, mostly rely on only a single type of feature, such as environmental features like wind speed and direction, without integrating key operating parameters such as equipment health status and terrain complexity. This results in feature sequences failing to fully reflect the operating characteristics of wind turbines under complex conditions, and the initial prediction model input information being one-sided. Furthermore, existing technologies often employ fixed rules or traditional static correction methods (such as fixed-proportion correction based on historical average deviations), without considering the impact of dynamic changes in operating conditions on prediction errors. This leads to a significant decrease in correction effectiveness under extreme operating conditions, large fluctuations in prediction accuracy, poor performance in multi-source data fusion processing, and inadequate power prediction processing and analysis. Summary of the Invention

[0004] The purpose of this invention is to provide a method for multi-source data fusion and power prediction of wind turbines, which solves the technical problems of poor multi-source data fusion processing and poor power prediction processing and analysis in existing technical solutions.

[0005] The objective of this invention can be achieved through the following technical solutions: A method for multi-source data fusion and power prediction of wind turbine units includes: S1: Collect multi-source raw data of wind turbine under different operating conditions. The multi-source raw data includes meteorological data, equipment operation data and environmental data. After preprocessing the multi-source raw data, dynamically weighted fusion of the preprocessed multi-source data is performed by an improved mutual information-attention fusion algorithm to generate a fusion feature sequence. S2: Using the generated fusion feature sequence as input, construct a CNN-LSTM-random forest hybrid prediction model based on spatiotemporal attention mechanism, train the hybrid prediction model, and output the preliminary predicted value of wind turbine power. S3: By fusing key operating parameters in the feature sequence, the operating conditions of the wind turbine are classified. For each type of operating condition, an error correction rule base based on historical data statistics and expert experience is established. During real-time operation, the obtained preliminary prediction value is dynamically adjusted according to the correction strategy in the error correction rule base matched with the current operating condition to obtain the final power prediction value.

[0006] Preferably, monitoring equipment is deployed under different operating conditions, and the operating conditions are divided and predefined, including wind speed level, terrain type and equipment health status; wind speed level includes low wind speed, medium wind speed and high wind speed; terrain type includes plain, mountain and sea; equipment health status includes normal, sub-healthy and fault warning. Collect multi-source raw data from wind turbines, including meteorological data, equipment operation data, and environmental data.

[0007] Preferably, when dynamically weighting and fusing the preprocessed multi-source data using the improved mutual information-attention fusion algorithm, the mutual information value between each data source and the power output of the wind turbine is calculated, and key features are screened; features with a mutual information value > 0.6 are retained. When performing dynamic weighting of the attention mechanism, an attention network is constructed, and the weights of each data source are dynamically adjusted according to real-time operating conditions. The weighted multi-source data are then spliced ​​together according to timestamps to generate a fused feature sequence.

[0008] Preferably, when constructing the CNN-LSTM-random forest hybrid prediction model with spatiotemporal attention mechanism, it includes a spatial feature extraction module, a temporal feature capture module, a feature weighting fusion module, and a random forest module; The spatial feature extraction module takes a fused feature sequence as input and outputs local spatial correlation features. The temporal feature capture module takes as input the local spatial correlation features output by the spatial feature extraction module and outputs temporal dependency features. The input to the feature weighted fusion module is local spatial correlation features and temporal dependency features, and the output is weighted fused features; The input to the random forest module is weighted fusion features, and the output is the prediction value of a single tree. All the prediction values ​​of the single trees are integrated to obtain the integrated prediction value. Finally, the preliminary prediction value is directly obtained from the integrated prediction value output by the random forest.

[0009] Preferably, when training the constructed hybrid prediction model, the input data is the generated fusion feature sequence, the label data is the historical actual power data, and the division ratio is 70% for the training set, 15% for the validation set, and 15% for the test set, divided in chronological order; Bayesian optimization is used to adjust the key hyperparameters, and the optimization objective is to minimize the RMSE of the validation set.

[0010] Preferably, the fused features of the test set are input into the trained hybrid model to output preliminary power prediction values.

[0011] Preferably, the working condition categories are divided based on the three core dimensions in the fused feature sequence to form a three-dimensional working condition classification matrix; The three core dimensions are wind speed range, equipment health status, and ambient temperature; the corresponding classification criteria are wind speed characteristics, equipment health characteristics, and ambient temperature characteristics.

[0012] Preferably, for 27 types of working conditions, correction rules are constructed based on historical data and expert experience to form an error correction rule base: Obtain historical running data for the test dataset, including fused feature sequences, preliminary power predictions, and actual power. For each type of working condition, calculate the average deviation rate and deviation direction under that condition.

[0013] Preferably, when analyzing the direction of deviation, if the average deviation rate is greater than 0, the preliminary power prediction value is determined to be generally too high. If the average deviation rate is less than 0, the preliminary power prediction value is determined to be generally too low.

[0014] Preferably, the feature is that, when completing the current working condition identification and error correction based on the real-time fused feature sequence, three key parameters are extracted from the real-time fused feature sequence: real-time wind speed, real-time equipment health status, and real-time ambient temperature at the current moment. The three key parameters are compared with the classification criteria of the corresponding parameters to determine the current working condition category; Based on the matched operating condition category, the corresponding preset correction rule is called from the rule base to adjust and correct the preliminary power prediction value; The corrected result is output as the final power prediction value.

[0015] Compared to existing solutions, the beneficial effects achieved by this invention are: This invention retains key features through mutual information filtering, which can effectively reduce redundant information. The attention mechanism and dynamic weighting make the fused features more adaptable to real-time operating conditions. The differentiated weighting for wind speed, terrain, and equipment status ensures that the fused features maintain high correlation in complex scenarios, such as typhoons or equipment aging, which can effectively improve the reliability of the average mutual information value of power output.

[0016] This invention captures local spatial correlations, such as wind speed-pitch angle coupling, through spatial feature extraction; captures temporal dependencies, such as short-term power trends, through temporal features; and processes nonlinear mappings, such as power saturation under extreme wind speeds, through random forests, which can solve the problem of insufficient generalization ability of a single model. Through the attention mechanism, it can automatically allocate weights for different time steps and features, which can effectively improve prediction accuracy compared to fixed weight fusion.

[0017] This invention avoids complex mathematical processes such as matrix operations and noise covariance estimation. Its core logic is classification-table lookup-correction, which can be implemented through simple programming, such as Python dictionary matching or Excel rule tables, effectively reducing the development difficulty of wind farm control systems. Correction rules directly correspond to specific operating conditions and expert experience; for example, under high wind speed and poor operating conditions, the rule needs to be reduced by 7.5%. Maintenance personnel can intuitively understand the basis for correction and quickly locate whether the corresponding rule in the rule base needs optimization when deviations occur. Rules are customized for 27 types of operating conditions, covering low / high wind speed, equipment age, and temperature, demonstrating strong adaptability. Compared to a single correction strategy, it can effectively improve prediction accuracy under extreme conditions. The operating condition matching and rule calling process is time-efficient, requiring only the extraction of 3 parameters and one table lookup, meeting the time requirements for real-time wind turbine control. Attached Figure Description

[0018] The present invention will now be further described with reference to the accompanying drawings.

[0019] Figure 1 This is a flowchart of Embodiment 1 of the present invention.

[0020] Figure 2 This is a flowchart of Embodiment 2 of the present invention.

[0021] Figure 3 A schematic diagram of the structure of a computer device for implementing embodiments of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1: As Figure 1 As shown, this invention is a method for multi-source data fusion and power prediction of wind turbine units, comprising: S1: Collect multi-source raw data from wind turbines under different operating conditions. The multi-source raw data includes meteorological data, equipment operation data, and environmental data. After preprocessing the multi-source raw data, dynamically weighted fusion is performed on the preprocessed multi-source data using an improved mutual information-attention fusion algorithm to generate a fusion feature sequence. Specific steps include: Deploy monitoring equipment under different operating conditions, and classify and predefine the operating conditions, including wind speed level, terrain type and equipment health status; Wind speed rating: 3-7 m / s, corresponding to low wind speed; 7-12 m / s, corresponding to medium wind speed; >12 m / s, corresponding to high wind speed; Terrain types include plains, mountains, and sea; Equipment health status includes normal, sub-health, and fault warning; for example, sub-health corresponds to abnormal gearbox temperature but no alarm is triggered; fault warning corresponds to blade damage warning signal. Collect multi-source raw data from wind turbines, including meteorological data, equipment operation data, and environmental data; Meteorological data were collected using lidar anemometers to measure wind speed and direction; ambient temperature and humidity were collected using temperature and humidity sensors. Equipment operating data, such as generator speed, pitch angle, gearbox temperature, and converter output current, are collected through the SCADA system; the SCADA system is the existing data acquisition and monitoring system. Environmental data were collected using a barometric pressure sensor and a turbulence intensity monitor to measure air pressure and atmospheric turbulence intensity. Data was collected continuously for 30 days under each working condition, with a daily collection time of 24 hours, and multi-source data was synchronized via GPS timestamps. Preprocessing the collected multi-source raw data includes, but is not limited to, outlier removal, missing value imputation and data standardization. These are all existing conventional technical methods, and the specific implementation steps will not be elaborated here. When performing dynamic weighted fusion of preprocessed multi-source data using an improved mutual information-attention fusion algorithm, the mutual information value between each data source and the power output of the wind turbine is calculated to screen key features. The mutual information formula is as follows: Where X is a candidate feature, such as wind speed or rotational speed, and Y is power output; p(x,y) is the joint probability distribution of X and Y; p(x) and p(y) are marginal probability distributions. Features with mutual information values ​​I(X;Y) > 0.6, such as wind speed, rotational speed, propeller pitch angle, and air density, are retained. When performing dynamic weighting of the attention mechanism, an attention network is constructed, and the weights of each data source are dynamically adjusted according to real-time conditions. The weight calculation formula is as follows: ;in, Let m be the weight of the m-th data type at time t; m is the terrain type; t is the time variable. The attention score is calculated by using a multilayer perceptron (MLP) to process the feature sequence. Obtained by performing a nonlinear transformation; The rules for dynamic adjustment are as follows: At high wind speeds, the weight of meteorological data The value ranges from 0.6 to 0.7; Weighting of equipment operating data at low wind speeds The value ranges from 0.5 to 0.6; In complex terrains, such as mountains and sea, environmental data weighting The value ranges from 0.3 to 0.4; The weighted multi-source data is concatenated according to timestamps to generate a fused feature sequence: ;in, These are preprocessed feature sequences for meteorological, equipment, and environmental data, respectively. Fusion feature sequence The dimension is [T×D], where T is the number of time steps and D is the total dimension of the features; for example, if there are 20 features in 3 classes of data, then D=20.

[0024] In this embodiment of the invention, key features are retained through mutual information filtering, which can effectively reduce redundant information. The dynamic weighting of the attention mechanism makes the fused features more adaptable to real-time operating conditions. The differentiated weight adjustment for wind speed, terrain, and equipment status enables the fused features to maintain high correlation in complex scenarios, such as typhoons or equipment aging, which can effectively improve the reliability of the average mutual information value of power output.

[0025] S2: Using the generated fused feature sequence as input, construct a CNN-LSTM-random forest hybrid prediction model based on a spatiotemporal attention mechanism, train the hybrid prediction model, and output a preliminary predicted value of the wind turbine power; the specific steps include: When constructing a CNN-LSTM-random forest hybrid prediction model with a spatiotemporal attention mechanism, it includes a spatial feature extraction module, a temporal feature capture module, a feature weighted fusion module, and a random forest module. The network structure of the spatial feature extraction module is as follows: Convolutional layer 1: 3×3 convolutional kernels, number 32, stride 1, ReLU activation function; Pooling layer 1: 2×2 max pooling, output dimension 50×16; Convolutional layer 2: 3×3 convolutional kernels, 64 in number, stride 1, ReLU activation function; Pooling layer 2: 2×2 max pooling, output dimension 25×32; The input is a fused feature sequence. The output is the local spatial correlation feature C(t), with a dimension of 25×32, and the relevant expression is: ;in, This is the first layer of convolution operations; This is the first layer pooling operation; This is the second layer of convolution operation; This is the second layer pooling operation; Local spatial correlation features C(t) are extracted through convolution and pooling operations of CNN, which can solve the nonlinear coupling problem of multi-source data; for example, the coupling relationship between wind speed and blade pitch angle, and the nonlinear correlation between temperature and current. Network structure of the temporal feature capture module: A 2-layer LSTM with 128 hidden nodes per layer and a dropout rate of 0.2. Output layer: Fully connected layer with 64 nodes, using the tanh activation function; The input is the local spatial association feature C(t) output by the spatial feature extraction module, and the output is the temporal dependency feature L(t) with a dimension of 25×64. The relevant expression is as follows: ;in, This is the first layer of LSTM; It is a second-layer LSTM; It is a fully connected layer; The time-dependent feature L(t) captures the dynamic features of time series through the memory unit of LSTM, which can solve the problem of short-term trend dependence in power prediction; for example, the short-term trend of power fluctuates within 5 minutes. The input to the feature weighted fusion module is the local spatial correlation feature C(t) and the temporal dependency feature L(t). The expression involved in the weight calculation is as follows: ;in, Let be the attention weight at time step i; where i is the time step index. The similarity score is calculated using the learnable parameter matrix W and the bias b. And output weighted fusion features The dimension is 1×96. For feature splicing; The processing parameters of the random forest module include the number of decision trees, maximum depth, and splitting feature selection; Number of decision trees: 100; Maximum depth: 15; Splitting feature selection: Gini impurity; The input to the random forest module is a weighted fusion feature. The output is the predicted value for a single tree. k is the decision tree index. The predictions of all individual trees are integrated to obtain the integrated prediction value. ; Final preliminary forecast The ensemble prediction value R(t) is obtained directly from the output of the random forest; When training the constructed hybrid prediction model, the input data is the generated fusion feature sequence, and the label data is the historical actual power data. The division ratio is 70% for the training set, 15% for the validation set, and 15% for the test set, divided in chronological order to avoid data leakage. The loss function is: Where L is the total loss function value; N is the number of samples; λ is the regularization coefficient, with a default value of 0.001; and w is the model's learnable parameter. Predict power values ​​for the model; This is the actual power value; For L1 regularization terms; The optimizer is Adam optimizer, with a learning rate of 1e-4, a decay rate of 1e-5, and a batch size of 32. Bayesian optimization is used to adjust key hyperparameters, with the optimization objective being to minimize the RMSE on the validation set. The key hyperparameters include: The number of CNN convolutional kernels is searched within the range of [16, 64], with the optimal values ​​being 32 and 64. The number of hidden nodes in the LSTM layer is within the search range of [64, 256], with an optimal value of 128. The number of trees in the random forest is searched within the range of [50, 200], with an optimal value of 100. The number of hidden nodes in the attention layer MLP has a search range of [32, 128], with an optimal value of 64. The key hyperparameters of the Bayesian optimization were adjusted to existing conventional technical solutions. The specific implementation steps for the adjustment are not detailed here. Fuse features from the test set Input the trained hybrid model and output preliminary power predictions. .

[0026] In this embodiment of the invention, local spatial correlations are captured by spatial feature extraction, such as wind speed-pitch angle coupling; temporal dependencies are captured by temporal features, such as short-term power trends; and nonlinear mappings, such as power saturation under extreme wind speeds, are processed by random forest. This can solve the problem of insufficient generalization ability of a single model. Through the attention mechanism, the weights of different time steps and features can be automatically allocated, which can effectively improve prediction accuracy compared with fixed weight fusion.

[0027] S3: By fusing key operating parameters from the feature sequence, the operating conditions of the wind turbine are classified. For each type of operating condition, an error correction rule base based on historical data statistics and expert experience is established. During real-time operation, the obtained preliminary prediction value is dynamically adjusted according to the correction strategy in the error correction rule base matched to the current operating condition to obtain the final power prediction value. The specific steps include: The working condition categories are divided based on the three core dimensions in the fused feature sequence F(t), forming a three-dimensional working condition classification matrix; The three core dimensions are wind speed range, equipment health status, and ambient temperature; the corresponding classification criteria are wind speed characteristics, equipment health characteristics, and ambient temperature characteristics. Equipment health characteristics can be scored based on parameters such as vibration and temperature, ranging from 0 to 100 points. There are no specific scoring rules, and they can be customized according to the application needs of the actual application scenario. Wind speed characteristics, specifically corresponding to low wind speeds <5, medium wind speeds 5-12, and high wind speeds >12; Equipment health characteristics are specifically categorized as follows: 80-100 points = Good, 60-79 points = Fair, and <60 points = Poor. The ambient temperature characteristics are specifically defined as <0°C for low temperature, 0-35°C for normal temperature, and >35°C for high temperature. The total number of operating condition categories is 3×3×3=27; for example, "low wind speed - good - normal temperature", "high wind speed - poor - high temperature", etc. For 27 types of operating conditions, correction rules were constructed based on historical data and expert experience, forming an error correction rule base: Obtain historical running data from the test dataset, including the fused feature sequence F(t) and preliminary power prediction values. Actual power Y(t); For each type of working condition, calculate the average deviation rate and deviation direction under that condition; The formula for calculating the average deviation rate C is as follows: ;in, The number of historical samples for the k-th type of working condition; For the first time under this working condition Preliminary predicted values ​​for one sample; This is the actual value; When analyzing the direction of deviation, if the average deviation rate is greater than 0, then the preliminary power prediction value is determined. Overall, it is relatively high; If the average deviation rate is less than 0, then the preliminary power prediction value is determined. Overall, it is relatively low; Based on the real-time fusion feature sequence F(t), when completing the current working condition identification and error correction, three key parameters are extracted from the real-time fusion feature sequence F(t): real-time wind speed, real-time equipment health status, and real-time ambient temperature. The three key parameters are compared with the classification criteria of the corresponding parameters to determine the current working condition category; Based on the matched operating condition category, the corresponding preset correction rule is retrieved from the rule base to adjust the initial power prediction value. Adjustments are made; the preset correction rules include setting corresponding correction coefficients for different operating condition categories; for example, "medium wind speed - good - normal temperature" corresponds to a correction coefficient of X=1.0378, and the final power prediction value... ; It should be explained that the specific values ​​of the correction coefficients for different operating conditions can be determined by combining historical operating data with the experience of wind farm operation and maintenance experts. Before formal deployment, simulation tests can also be conducted based on existing simulation testing software, and the correction coefficients can be corrected and determined based on the results of the simulation tests. When handling special cases, if the real-time parameters are close to the classification boundary, for example, wind speed = 4.9 m / s, which is close to the boundary between "low wind speed" and "medium wind speed", then a weighted average correction is adopted: take the correction value of the rules of the two adjacent working conditions, and weight them according to the proportion of distance from the boundary. For example, if the distance from the low wind speed boundary is 0.1 m / s and the distance from the medium wind speed boundary is 0.1 m / s, then each is given a weight of 50%. The corrected result will be used as the final power prediction value. Output.

[0028] In this embodiment of the invention, complex mathematical processes such as matrix operations and noise covariance estimation are avoided. The core logic is classification-table lookup-correction, which can be implemented through simple programming, such as Python dictionary matching and Excel rule tables, effectively reducing the development difficulty of wind farm control systems. The correction rules directly correspond to specific operating conditions and expert experience. For example, under high wind speed-poor operating conditions, the rule needs to be reduced by 7.5%. Operation and maintenance personnel can intuitively understand the basis for correction and quickly locate whether the rules in the rule base corresponding to the operating conditions need to be optimized when deviations occur. Rules are customized for 27 types of operating conditions, covering low / high wind speed, new and old equipment, high / low temperature scenarios, etc., with strong adaptability to operating conditions. Compared with a single correction strategy, it can effectively improve the prediction accuracy under extreme operating conditions. The operating condition matching and rule calling process takes less than 0.5ms, requiring only 3 parameters to be extracted and 1 table lookup, which can meet the time requirements of real-time control of wind turbine units, with a typical control cycle of ≥10ms.

[0029] Example 2: As Figure 2 As shown, it also includes S4: monitoring and analyzing the prediction effect of the final power prediction value through multi-dimensional methods, and dynamically optimizing and managing the existing power prediction scheme based on the analysis results. Specific steps include: When processing and analyzing the single prediction effect of the final power value, the difference between the final power prediction value and the actual power value is calculated, and the difference is matched with a preset difference range. The difference range can be determined according to existing industry standards or customized according to the application requirements of the actual application scenario. The specific value is not limited. If the difference is within the range, the single prediction effect is determined to be normal, and a first effect label is generated. If the difference is not a difference, the single prediction effect is determined to be abnormal, and a second effect label is generated; When processing and analyzing the overall prediction effect of existing power prediction schemes based on the obtained effect labels, the total number of the first effect labels and the total number of the second effect labels are counted respectively, and the abnormal impact value is calculated using a formula. Where n and m are the total number of second effect labels and the total number of first effect labels, respectively; a is the standard value of abnormal influence, which can be determined based on previous simulation test data. If the abnormal impact value is less than 0, the overall prediction effect is judged to be normal, and the existing power prediction scheme is maintained. If the abnormal impact value is greater than or equal to 0, the overall prediction effect is determined to be abnormal, and the existing power prediction scheme will be optimized and managed. The optimization management can add, delete, or modify the existing power prediction scheme, and the specific implementation content is not limited here.

[0030] In this embodiment of the invention, the prediction effect of the final power prediction value is monitored and processed through multi-dimensional means, and the existing power prediction scheme is dynamically optimized and managed based on the analysis results. This achieves subsequent automated monitoring and proactive optimization, which can further improve the diversity and reliability of wind turbine power prediction.

[0031] Example 3: As Figure 3 The diagram shown is a structural schematic of a computer device for implementing a method for multi-source data fusion and power prediction of wind turbine generators, as provided in an embodiment of the present invention.

[0032] Computer equipment may include a processor, memory, and bus, and may also include computer programs stored in memory and capable of running on the processor, such as a wind turbine multi-source data fusion and power prediction program.

[0033] The memory 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, disk, optical disk, etc. In some embodiments, the memory can be an internal storage unit of a computer device, such as a portable hard drive. In other embodiments, the memory can be an external storage device of a computer device, such as a plug-in portable hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc. Furthermore, the memory can include both internal and external storage units of the computer device. The memory can be used not only to store application software and various types of data installed on the computer device, such as the code for a wind turbine multi-source data fusion and power prediction program, but also to temporarily store data that has been output or will be output.

[0034] In some embodiments, the processor may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits packaged with the same or different functions. This includes combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor is the control unit of the computer device, connecting various components of the device via various interfaces and lines. It executes programs or modules stored in memory (e.g., a multi-source data fusion and power prediction program for wind turbines) and calls data stored in memory to perform various functions and process data.

[0035] 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 address bus, data bus, control bus, etc. The bus is configured to enable communication between the memory and at least one processor, etc.

[0036] Figure 3 Only computer equipment with components is shown; those skilled in the art will understand that... Figure 3 The structure shown does not constitute a limitation on the computer device and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0037] For example, although not shown, the computer device may also include a power supply (such as a battery) to power various components. Preferably, the power supply can be logically connected to at least one processor via 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 sources, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The computer device may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be elaborated further here.

[0038] Furthermore, the computer device may also include a network interface. Optionally, the network interface 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 computer device and other computer devices.

[0039] Optionally, the computer device may further include a user interface, which may be a display, an input unit (such as a keyboard), and optionally, a standard wired interface or a 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, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the computer device and to display a visual user interface.

[0040] It should be understood that the above embodiments are for illustrative purposes only and are not limited to this structure in the scope of patent applications.

[0041] A multi-source data fusion and power prediction program for wind turbine generators, stored in the memory of a computer device, is a combination of multiple instructions.

[0042] Specifically, the processor'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.

[0043] Furthermore, if the modules / units integrated into a computer device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, a computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0044] The present invention also provides a computer-readable storage medium storing a computer program that is executed by a processor of a computer device.

[0045] In the several embodiments provided by this invention, it should be understood that the disclosed methods can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for example, the division of modules is merely a logical functional division, and there may be other division methods in actual implementation.

[0046] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0047] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0048] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for multi-source data fusion and power prediction of wind turbine generators, characterized in that, include: S1: Collect multi-source raw data of wind turbine under different operating conditions. The multi-source raw data includes meteorological data, equipment operation data and environmental data. After preprocessing the multi-source raw data, the preprocessed multi-source data is dynamically weighted and fused using an improved mutual information-attention fusion algorithm to generate a fused feature sequence. S2: Using the generated fusion feature sequence as input, construct a CNN-LSTM-random forest hybrid prediction model based on spatiotemporal attention mechanism, train the hybrid prediction model, and output the preliminary predicted value of wind turbine power. S3: By fusing key operating parameters in the feature sequence, the operating conditions of the wind turbine are classified. For each type of operating condition, an error correction rule base based on historical data statistics and expert experience is established. During real-time operation, the obtained preliminary prediction value is dynamically adjusted according to the correction strategy in the error correction rule base matched with the current operating condition to obtain the final power prediction value.

2. The method for multi-source data fusion and power prediction of wind turbine units according to claim 1, characterized in that, The monitoring equipment is deployed under different operating conditions, and the operating conditions are divided and predefined, including wind speed level, terrain type and equipment health status; wind speed level includes low wind speed, medium wind speed and high wind speed; terrain type includes plains, mountains and sea; equipment health status includes normal, sub-healthy and fault warning. Collect multi-source raw data from wind turbines, including meteorological data, equipment operation data, and environmental data.

3. The method for multi-source data fusion and power prediction of wind turbine units according to claim 2, characterized in that, When performing dynamic weighted fusion of preprocessed multi-source data using an improved mutual information-attention fusion algorithm, the mutual information value between each data source and the power output of the wind turbine is calculated to screen key features. Features with mutual information values ​​> 0.6 are retained; When performing dynamic weighting of the attention mechanism, an attention network is constructed, and the weights of each data source are dynamically adjusted according to real-time operating conditions. The weighted multi-source data are then spliced ​​together according to timestamps to generate a fused feature sequence.

4. The method for multi-source data fusion and power prediction of wind turbine units according to claim 3, characterized in that, When constructing a CNN-LSTM-random forest hybrid prediction model with a spatiotemporal attention mechanism, it includes a spatial feature extraction module, a temporal feature capture module, a feature weighted fusion module, and a random forest module. The spatial feature extraction module takes a fused feature sequence as input and outputs local spatial correlation features. The temporal feature capture module takes as input the local spatial correlation features output by the spatial feature extraction module and outputs temporal dependency features. The input to the feature weighted fusion module is local spatial correlation features and temporal dependency features, and the output is weighted fused features; The input to the random forest module is weighted fusion features, and the output is the prediction value of a single tree. All the prediction values ​​of the single trees are integrated to obtain the integrated prediction value. Finally, the preliminary prediction value is directly obtained from the integrated prediction value output by the random forest.

5. The method for multi-source data fusion and power prediction of wind turbine units according to claim 4, characterized in that, When training the constructed hybrid prediction model, the input data is the generated fusion feature sequence, and the label data is the historical actual power data. The division ratio is 70% for the training set, 15% for the validation set, and 15% for the test set, in chronological order. Bayesian optimization is used to adjust the key hyperparameters, and the optimization objective is to minimize the RMSE of the validation set.

6. The method for multi-source data fusion and power prediction of wind turbine units according to claim 5, characterized in that, The fused features from the test set are input into the trained hybrid model, which outputs preliminary power predictions.

7. The method for multi-source data fusion and power prediction of wind turbine units according to claim 6, characterized in that, The working condition categories are divided based on the three core dimensions in the fused feature sequence, forming a three-dimensional working condition classification matrix; The three core dimensions are wind speed range, equipment health status, and ambient temperature; the corresponding classification criteria are wind speed characteristics, equipment health characteristics, and ambient temperature characteristics.

8. The method for multi-source data fusion and power prediction of wind turbine units according to claim 7, characterized in that, For 27 types of operating conditions, correction rules were constructed based on historical data and expert experience, forming an error correction rule base: Obtain historical running data for the test dataset, including fused feature sequences, preliminary power predictions, and actual power. For each type of working condition, calculate the average deviation rate and deviation direction under that condition.

9. A method for multi-source data fusion and power prediction of wind turbine units according to claim 8, characterized in that, When analyzing the direction of deviation, if the average deviation rate is greater than 0, the preliminary power prediction value is determined to be generally too high. If the average deviation rate is less than 0, the preliminary power prediction value is determined to be generally too low.

10. A method for multi-source data fusion and power prediction of wind turbine units according to claim 9, characterized in that, Based on the real-time fusion feature sequence, when completing the current working condition identification and error correction, three key parameters are extracted from the real-time fusion feature sequence: real-time wind speed, real-time equipment health status, and real-time ambient temperature. The three key parameters are compared with the classification criteria of the corresponding parameters to determine the current working condition category; Based on the matched operating condition category, the corresponding preset correction rule is called from the rule base to adjust and correct the preliminary power prediction value; The corrected result is output as the final power prediction value.

Images