A method, system, equipment and medium for predicting port wind speed

Abstract

This invention discloses a method, system, equipment, and medium for port wind speed prediction, relating to the field of port and shipping safety technology. The method includes: collecting port wind speed time-series data; performing variational mode decomposition on the port wind speed time-series data to obtain several mode components with finite bandwidth; obtaining the one-sided spectrum of each mode component through Hilbert transform; transferring the one-sided spectrum of each mode component to baseband; tuning to their respective estimated center frequencies through a mixing exponent; demodulating the signal according to Gaussian smoothing constraints; obtaining the bandwidth of each mode component using the squared 2L norm of the demodulated mode component gradient; independently predicting the bandwidth of each decomposed mode using multiple GRU predictors; and performing regression analysis on the prediction results of each component to obtain the final wind speed prediction value. This invention improves the ability to capture sudden wind speed changes, enhances the synergistic utilization of multi-component prediction results, and provides more reliable data support for port operation safety and efficiency.

Description

Technical Field

[0001] This invention relates to the field of port and shipping safety technology, and in particular to a method, system, equipment and medium for predicting port wind speed. Background Technology

[0002] Wind is a key meteorological factor affecting port operations. Strong winds can cause a series of problems, including navigational obstacles, disruptions to cargo loading and unloading operations, and damage to port facilities. Meteorological wind speed forecasts are typically based on large-scale areas, which limits their accuracy and precision for localized areas like ports. Furthermore, the timeliness of weather forecasts often fails to meet the real-time wind speed information requirements of port operations. Therefore, it is necessary to develop an accurate and reliable port wind speed prediction model to provide timely wind speed information for port operations, enabling port personnel to take appropriate measures and improve port operational efficiency and safety.

[0003] Currently, wind speed prediction methods are mainly classified into three categories: physical models, statistical methods, and machine learning methods. Physical models, represented by numerical weather prediction (NWP), predict wind speed by simulating atmospheric processes, but require a large amount of observational data and computational resources, resulting in high costs and difficulties. Traditional statistical methods build linear or nonlinear models based on historical wind speed data, often leading to unsatisfactory prediction accuracy. Machine learning methods, such as artificial neural networks (ANN), support vector machines (SVM), and random forests (RF), can automatically learn features from data, but due to the strong non-stationarity of wind speed sequences, single-model prediction errors are relatively large. To overcome this deficiency, researchers have proposed several hybrid prediction models: using Empirical Mode Decomposition (EMD) to decompose wind speed into multiple modal components, and then using Long Short-Term Memory (LSTM) networks for prediction; and using Variational Mode Decomposition (VMD) combined with LSTM, verifying that this method outperforms EMD and Adaptive Noise Complete Empirical Mode Decomposition (CEEMDAN) in ultra-short-term wind speed prediction. However, after adopting VMD decomposition, the above methods usually require training K independent LSTM networks, which leads to a significant increase in the number of model parameters. In short-term wind speed prediction (such as the next 10 minutes to 1 hour), there are still problems with poor performance and weak generalization ability, making it difficult to directly apply to the multi-timescale wind speed prediction needs in the complex environment of ports. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of the prior art by providing a port wind speed prediction method, system, device, and medium to solve the problems in the prior art.

[0005] The present invention specifically provides the following technical solution: A method for predicting wind speed in a port includes: Collect time-series wind speed data at the port; Variational mode decomposition is performed on port wind speed time series data to obtain several modal components with finite bandwidth. The one-sided spectrum of each modal component is obtained through Hilbert transform. The one-sided spectrum of each modal component is transferred to the baseband. The transferred one-sided spectrum is tuned to the estimated center frequency of each modal component through a mixing exponent. Based on the center frequency, Gaussian smoothing constraint is used to demodulate the modal components. The bandwidth of each modal component is obtained by the square 2L norm of the gradient of the demodulated modal component. By combining multiple GRU predictors with attention mechanisms to independently predict the bandwidth of each decomposed mode, and by weighting and summing the hidden states at all time steps during prediction, the wind speed prediction results of each modal component are obtained. The wind speed prediction results of each modal component are then input into a deep feedforward neural network for regression analysis to obtain the final wind speed prediction value.

[0006] Preferably, the demodulation of modal components based on the center frequency and using Gaussian smoothing constraints, and obtaining the bandwidth of each modal component using the squared 2L norm of the demodulated modal component gradient, specifically: Using Gaussian smoothness and the gradient squared criterion as Gaussian smoothing constraints, the modal components are demodulated. ; ; Among them, ∂ t This indicates finding the partial derivative. δ ( t ) is the Dirac distribution function. j The imaginary unit, For the decomposition of the first k IMF of modal components, * denotes convolution operation. It is a natural constant. for The center frequency; K The total number of components, This is time-series data on wind speed at the port. For time variables, The squared 2L norm is used; Gaussian smoothness and the gradient squared criterion are used as Gaussian smoothness constraints.

[0007] Preferably, the bandwidth of each decomposition mode is transformed and solved, specifically as follows: via Lagrange multipliers λ and secondary penalty items α The constrained extremum problem is transformed into an unconstrained problem for solution, yielding the bandwidth of each decomposed mode; the specific expression is: ; in, This represents the bandwidth of each decomposed mode after the transformation. Let Lagrange multipliers be the time function of the Lagrange multipliers. It is the time function of the modal components.

[0008] Preferably, the step of tuning each transferred single-sided spectrum to the estimated center frequency of each modal component using a mixing exponent specifically involves: The modal components and their corresponding center frequencies are obtained using the alternating direction multiplier method. The specific expression is as follows: ; ; ; in: , , and Iteration Fourier transform at the next time step; For frequency, Number of iterations Iteration The center frequency at this time, Iteration Fourier transform of the next time modal component Iteration Fourier transform of the Lagrange multipliers at the next time step This is the step size parameter.

[0009] Preferably, the method of independently predicting the bandwidth of each decomposed mode by combining multiple GRU predictors with an attention mechanism, and weighted summing of the hidden states at all time steps during prediction to obtain the wind speed prediction results for each mode component, specifically: Obtain the reset gate using the bandwidth of each decomposition mode. and the update gate The specific expression is: ; ; in, and These are the parameter matrix and bias vector of the reset gate. and It updates the parameter matrix and bias vector of the gate. It was the hidden state from a previous moment. It is the input at the current moment, that is, the bandwidth of each modal component after wind speed decomposition, and it incorporates meteorological factors and temporal characteristics. It is an activation function; The hidden state is obtained and updated based on the reset and update gates, and the specific expression is as follows: ; ; in, and These are the parameter matrix and the bias vector. This indicates the hidden state from the previous moment. And the reset door at the current moment Combined to control the extent of its impact, This indicates the retention of historical information. Indicates new information. Indicates a hidden state. Represents the hyperbolic tangent function. Indicates the updated hidden state; The hidden states at all time steps are weighted and summed using a time-mode attention mechanism, and the wind speed prediction results for each decomposed mode are obtained after transformation through the output layer.

[0010] Preferably, the step of inputting the wind speed prediction results of each modal component into a deep feedforward neural network for regression analysis to obtain the final wind speed prediction value specifically involves: The wind speed prediction results of each modal component are input into a deep feedforward neural network (DFF) for regression analysis. The wind speed prediction values ​​are updated in real time, and the final wind speed prediction value is output. The specific expression is as follows: ; in, x For input data, W Weight matrix, b As a bias term, ReLU is chosen as the activation function.

[0011] This invention provides a port wind speed prediction system, comprising: The data acquisition module is used to collect time-series wind speed data at the port. The decomposition module performs variational mode decomposition on port wind speed time series data to obtain several modal components with finite bandwidth. It then uses Hilbert transform to obtain the one-sided spectrum of each modal component, transfers the one-sided spectrum of each modal component to the baseband, and tunes each transferred one-sided spectrum to the estimated center frequency of each modal component using a mixing exponent. Based on the center frequency, it demodulates the modal components using Gaussian smoothing constraints, and obtains the bandwidth of each modal component using the squared 2L norm of the demodulated modal component gradient. The prediction module is used to independently predict the bandwidth of each decomposed mode by combining multiple GRU predictors with attention mechanism, and to perform weighted summation of the hidden states at all time steps during prediction to obtain the wind speed prediction results of each modal component. The wind speed prediction results of each modal component are then input into a deep feedforward neural network for regression analysis to obtain the final wind speed prediction value.

[0012] The present invention provides a computer device, including a memory and a processor. The memory stores a program, and when the program is executed by the processor, the processor performs the steps of the above-described port wind speed prediction method.

[0013] The present invention provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described port wind speed prediction method.

[0014] Compared with the prior art, the present invention has the following significant advantages: This invention employs a heuristic optimization algorithm to adaptively determine the optimal decomposition parameters for VMD (Vibration Dynamics Decomposition), decomposing the original non-stationary wind speed sequence into a series of stationary subsequences with finite bandwidth. Simultaneously, it constructs multiple GRU predictors incorporating an attention mechanism to replace the LSTM used in existing methods, performing parallel predictions on each subsequence. Regression analysis is used to nonlinearly fuse the multidimensional prediction results, outputting the final wind speed prediction value. Adaptive decomposition enhances the algorithm's adaptability to complex marine environments, while the attention mechanism improves the ability to capture sudden wind speed changes. It also enhances the synergistic utilization of multi-component prediction results, significantly improving the accuracy and stability of multi-timescale wind speed prediction in complex environments such as ports and waterways. It can achieve wind speed trend prediction at multiple timescales, including 1 minute, 5 minutes, and 10 minutes, providing more reliable data support for port operation safety and efficiency. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the VMD-GRU-MLP model structure of the present invention; Figure 2 This is a schematic diagram illustrating the wind speed data decomposition effect of the present invention; wherein Figure 2 (a) represents the time series data of wind speed at the port. Figure 2 (b) represents the decomposed data; Figure 3 The figure shows the experimental results of the GRU model when comparing the prediction performance of the model of the present invention; Figure 4 The figure shows the experimental results of the VMD-GRU model when comparing the prediction performance of the model of the present invention; Figure 5 The experimental results of the VMD-GRU-DFF model are shown in the comparison of the prediction performance of the model in this invention. Figure 6 This is a comparison chart of the 1-minute prediction result and the actual wind speed in the multi-timescale prediction effect of this invention; Figure 7 This is a comparison chart of the 5-minute prediction result and the actual wind speed in the multi-timescale prediction effect of this invention; Figure 8 This is a comparison chart of the 10-minute prediction result and the actual wind speed in the multi-timescale prediction effect of this invention. Figure 9 The present invention provides a flowchart of a port wind speed prediction method. Detailed Implementation

[0016] 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, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0017] Based on existing research methods, a combined prediction model of VMD and GRU-DFF (deep feedforward network) is proposed. After decomposing port wind speed time series data into multiple stationary modal components, the GRU-DFF model is used for prediction to obtain the final predicted wind speed. The effectiveness of the model is verified on a real wind speed sequence dataset.

[0018] like Figure 1 As shown, this embodiment provides a port wind speed prediction method through raw wind speed - VMD decomposition - GRU prediction of each component - DFF fusion - final predicted wind speed. Figure 9 As shown, it includes the following steps: Step 1: Collect port wind speed time series data.

[0019] Data preprocessing: cleaning outliers and normalization.

[0020] Step 2: Perform variational mode decomposition on the port wind speed time series data to obtain several mode components with finite bandwidth. Obtain the one-sided spectrum through Hilbert transform. Transfer the one-sided spectrum of each mode component to the baseband. Tune each transferred one-sided spectrum to the estimated center frequency of each mode component through a mixing exponent. Based on the center frequency, demodulate the mode components using Gaussian smoothing constraints. Obtain the bandwidth of each mode component by the square 2L norm of the gradient of the demodulated mode component.

[0021] VMD decomposition: The original wind speed sequence is decomposed into K modal components using variational mode decomposition, preferably K=7, with a penalty factor α=2000. The comparison between the original wind speed sequence and each modal component after VMD decomposition is as follows: Figure 2 As shown.

[0022] (1) Assuming the original port wind speed time series data It is decomposed into 7 components For each component, a one-sided spectrum is obtained through Hilbert transform.

[0023] (2) For each modal component, the one-sided spectrum of the modal component is transferred to the baseband and tuned to the estimated center frequency by the mixing index.

[0024] (3) The signal is demodulated using Gaussian smoothness and gradient squared criterion as Gaussian smoothness constraints, and the squared 2L norm of its gradient is calculated to obtain the bandwidth of each decomposition mode.

[0025] ; ; In the formula: ∂ t This indicates finding the partial derivative. δ ( t ) is the Dirac distribution function. j The imaginary unit, For the decomposition of the first k IMF of modal components, * denotes convolution operation. It is a natural constant. for The center frequency; K The total number of components, Port wind speed time series data. For time variables, Let be the squared 2L norm. The constrained extremum problem is transformed into an unconstrained problem by using the Lagrange multiplier λ and the quadratic penalty term α, yielding the bandwidth of each decomposed mode, as shown in the following equation:

[0026] (2); in, This represents the bandwidth of each decomposed mode after the transformation. Let Lagrange multipliers be the time function of the Lagrange multipliers. It is the time function of the modal components.

[0027] The modal components and their corresponding center frequencies are obtained using the alternating direction multiplier method, and the update method is as follows: ; ; ; In the formula: , Iterations Fourier transform at the next time step; For frequency, The number of iterations. For iteration The center frequency at this time, For iteration Fourier transform of the next time modal component For iteration Fourier transform of the Lagrange multipliers at the next time step This is the step size parameter.

[0028] Step 3: The bandwidth of each decomposed mode is independently predicted by multiple GRU predictors with attention mechanism. The hidden states of all time steps during prediction are weighted and summed to obtain the wind speed prediction results of each modal component. The wind speed prediction results of each modal component are then input into a deep feedforward neural network for regression analysis to obtain the final wind speed prediction value.

[0029] GRU prediction: Construct a GRU model for each modal component and make independent predictions: (1) Obtain the reset gate using the bandwidth of each decomposition mode. and the update gate The specific expression is: ; ; in, and These are the parameter matrix and bias vector of the reset gate. and It updates the parameter matrix and bias vector of the gate. It was the hidden state from a previous moment. The input is the current moment, which is the bandwidth of each modal component after wind speed decomposition, and introduces meteorological factors (such as the air pressure difference and temperature gradient at the same moment (e.g., obtained from the same meteorological station)) and time series features (minute pointer (time of day) and seasonal dummy variables). It is an activation function.

[0030] (2) Obtain the hidden state based on the reset door and update door, the specific expression is: ; in and These are the parameter matrix and the bias vector. This indicates the hidden state from the previous moment. And the reset door at the current moment Combined to control the extent of its impact.

[0031] (3) Update the hidden state, the specific expression is: ; in( ) indicates the historical information that has been retained. ) indicates new information. Indicates a hidden state. Represents the hyperbolic tangent function. This indicates the updated hidden state.

[0032] A temporal pattern attention mechanism is introduced after the GRU network for each modal component. Instead of simply relying on the hidden state of the last time step of the GRU, the temporal pattern attention mechanism performs a weighted summation of the hidden states of all time steps, allowing the model to focus on the moments in the historical sequence that are most relevant to the current prediction (such as the point of sudden increase in wind speed), and obtains the wind speed prediction results for each decomposed modality after transformation through the output layer.

[0033] Deep Feedforward Neural Network (DFF) Regression Fusion: The wind speed prediction results of each modal component are input into the DFF for regression analysis, and the final wind speed prediction value is output; the specific expression is: ; in, x Input data for this layer, W The weight matrix maps the input layer to the hidden layer, and then back to the output layer. b As the bias term, ReLU is chosen as the activation function. DFF further expands the width of the deep feedforward network by adding sparse autoencoder feature maps. Leveraging its fast incremental learning characteristics, it enables real-time updates of newly arrived wind speed data patterns without the need for frequent retraining of the entire VMD and GRU models.

[0034] Model evaluation: The RMSE and MAPE metrics were used to evaluate the model performance.

[0035] ; ; In the formula, This represents the actual wind speed value. To predict wind speed values.

[0036] The processing procedure in one embodiment is described below: Data sources and processing: Real-time wind speed data was collected using a wind speed monitoring instrument for large-scale lifting machinery in the port, with a sampling interval of 1 minute. The data was cleaned, null and negative values ​​were removed, and the data was normalized to the [0,1] interval.

[0037] VMD decomposition parameter settings: The VMD decomposition is implemented using the vmdpy library in Python. The number of modes K is set to 7, the penalty factor α is set to 2000, and the number of iterations is 500.

[0038] GRU Model Construction: The GRU network was constructed using the TensorFlow framework, with 64 neurons in the hidden layer, tanh as the activation function, mean squared error as the loss function, and Adam as the optimizer.

[0039] DFF fusion layer structure: DFF consists of one input layer, four hidden layers and one output layer, with the number of neurons being 32, 32, 32 and 16 respectively, the activation function being ReLU, and the output layer having linear activation.

[0040] Training and Testing: The data was divided into 80% training set and 20% test set, with 100 training epochs and a batch size of 32. RMSE and MAPE were used to evaluate model performance.

[0041] Results Comparison: Compared with GRU and VMD-GRU models, the prediction curves of the three models (GRU, VMD-GRU, and VMD-GRU-DFF) on the test set are compared as follows: Figure 3 , Figure 4 and Figure 5 As shown, after comparison, the VMD-GRU and VMD-GRU-DFF models have higher accuracy in predicting wind speed after signal decomposition, while the VMD-GRU-DFF model is more accurate in predicting extreme wind speed values. Among them, the VMD-GRU-DFF model of this invention reduces RMSE to 0.37m / s and MAPE to 6.78% on the dataset, and the prediction accuracy is significantly improved.

[0042] To give on-site workers more time to implement windproofing measures, this invention predicts wind speeds after 1 minute, 5 minutes, and 10 minutes, respectively. Figure 6 , Figure 7 and Figure 8 The comparison between the VMD-GRU-DFF model's predictions of wind speed after 1 minute, 5 minutes, and 10 minutes and the actual wind speeds in the dataset is shown. It can be seen that although the VMD-GRU-DFF model cannot accurately predict the wind speed value after 10 minutes, it can still predict the trend of wind speed change.

[0043] Based on the above methods and content, this invention also proposes a port wind speed prediction system, comprising: The system comprises the following modules: a data acquisition module for collecting port wind speed time-series data; a decomposition module for performing variational mode decomposition on the port wind speed time-series data to obtain several mode components with finite bandwidths, obtaining the one-sided spectrum of each mode component through Hilbert transform, transferring the one-sided spectrum of each mode component to baseband, and tuning each transferred one-sided spectrum to the estimated center frequency of each mode component through a mixing exponent; demodulating the mode components based on the center frequency using Gaussian smoothing constraints, and obtaining the bandwidth of each mode component by the square 2L norm of the demodulated mode component gradient; and a prediction module for independently predicting the bandwidth of each decomposed mode by using multiple GRU predictors combined with an attention mechanism, weighting and summing the hidden states at all time steps during prediction to obtain the wind speed prediction results for each mode component, and inputting the wind speed prediction results of each mode component into a deep feedforward neural network for regression analysis to obtain the final wind speed prediction value.

[0044] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a program, and when the program is executed by the processor, the processor performs the steps of a port wind speed prediction method.

[0045] According to the disclosed embodiments, the computer device can communicate with one or more external devices (e.g., keyboard, pointing device, Bluetooth communication, etc.) or with any device that enables the computing device to communicate with one or more other computing devices (e.g., router, demodulator, etc.).

[0046] The present invention also provides a storage medium storing a computer program thereon, characterized in that the computer program, when executed by a processor, implements the steps of a port wind speed prediction method.

[0047] According to the disclosed embodiments, the storage medium can be a non-volatile computer-readable storage medium, such as, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, the storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0048] The above description, in conjunction with specific preferred embodiments, provides a more detailed explanation of the present invention. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered to fall within the scope of protection of the present invention.

Claims

1. A method for predicting port wind speed, characterized in that, include: Collect time-series wind speed data at the port; Variational mode decomposition is performed on port wind speed time series data to obtain several modal components with finite bandwidth. The one-sided spectrum of each modal component is obtained through Hilbert transform. The one-sided spectrum of each modal component is transferred to the baseband. The transferred one-sided spectrum is tuned to the estimated center frequency of each modal component through a mixing exponent. Based on the center frequency, Gaussian smoothing constraint is used to demodulate the modal components. The bandwidth of each modal component is obtained by the square 2L norm of the gradient of the demodulated modal component. By combining multiple GRU predictors with attention mechanisms to independently predict the bandwidth of each decomposed mode, and by weighting and summing the hidden states at all time steps during prediction, the wind speed prediction results of each modal component are obtained. The wind speed prediction results of each modal component are then input into a deep feedforward neural network for regression analysis to obtain the final wind speed prediction value.

2. The port wind speed prediction method as described in claim 1, characterized in that, The modal components are demodulated using Gaussian smoothing constraints based on the center frequency. The bandwidth of each modal component is obtained by the square 2L norm of the gradient of the demodulated modal component, specifically expressed as follows: ； ； Among them, ∂ t This indicates finding the partial derivative. δ ( t ) is the Dirac distribution function. j The imaginary unit, For the decomposition of the first k IMF of modal components, * denotes convolution operation. It is a natural constant. for The center frequency; K The total number of components, This is time-series data on wind speed at the port. For time variables, The squared 2L norm is used; Gaussian smoothness and the gradient squared criterion are used as Gaussian smoothness constraints.

3. The port wind speed prediction method as described in claim 2, characterized in that, The bandwidth of each decomposed mode is transformed and solved, specifically as follows: via Lagrange multipliers λ and secondary penalty items α The constrained extremum problem is transformed into an unconstrained problem for solution, yielding the bandwidth of each decomposed mode; the specific expression is: ； in, This represents the bandwidth of each decomposed mode after the transformation. Let Lagrange multipliers be the time function of the Lagrange multipliers. It is the time function of the modal components.

4. The port wind speed prediction method as described in claim 3, characterized in that, The process of tuning each transferred single-sided spectrum to the center frequency estimated for each modal component using a mixing exponent is specifically as follows: The modal components and their corresponding center frequencies are obtained using the alternating direction multiplier method. The specific expression is as follows: ； ； ； in: , and They are respectively , and Iteration Fourier transform at the next time step; For frequency, The number of iterations. For iteration The center frequency at this time, For iteration Fourier transform of the next time modal component For iteration Fourier transform of the Lagrange multipliers at the next time step This is the step size parameter.

5. The port wind speed prediction method as described in claim 1, characterized in that, The method involves independently predicting the bandwidth of each decomposed mode by combining multiple GRU predictors with an attention mechanism, and weighted summing of the hidden states at all time steps during prediction to obtain the wind speed prediction results for each mode component. Specifically: Obtain the reset gate using the bandwidth of each decomposition mode. and the update gate The specific expression is: ； ； in, and These are the parameter matrix and bias vector of the reset gate. and It updates the parameter matrix and bias vector of the gate. It was the hidden state from a previous moment. It is the input at the current moment, that is, the bandwidth of each modal component after wind speed decomposition, and it incorporates meteorological factors and temporal characteristics. It is an activation function; The hidden state is obtained and updated based on the reset and update gates; the specific expression is: ； ； in, and These are the parameter matrix and the bias vector. This indicates the hidden state from the previous moment. And the reset door at the current moment Combined to control the extent of its impact, This indicates the retention of historical information. Indicates new information. Indicates a hidden state. Represents the hyperbolic tangent function. Indicates the updated hidden state; The hidden states at all time steps are weighted and summed using a time-mode attention mechanism, and the wind speed prediction results for each decomposed mode are obtained after transformation through the output layer.

6. The port wind speed prediction method as described in claim 5, characterized in that, The wind speed prediction results of each modal component are input into a deep feedforward neural network for regression analysis to obtain the final wind speed prediction value, specifically as follows: The wind speed prediction results of each modal component are input into a deep feedforward neural network (DFF) for regression analysis. The wind speed prediction values ​​are updated in real time, and the final wind speed prediction value is output. The specific expression is as follows: ； in, x For input data, W Weight matrix, b As a bias term, ReLU is chosen as the activation function.

7. A port wind speed prediction system, characterized in that, include: The data acquisition module is used to collect time-series wind speed data at the port. The decomposition module performs variational mode decomposition on port wind speed time series data to obtain several modal components with finite bandwidth. It then uses Hilbert transform to obtain the one-sided spectrum of each modal component, transfers the one-sided spectrum of each modal component to the baseband, and tunes each transferred one-sided spectrum to the estimated center frequency of each modal component using a mixing exponent. Based on the center frequency, it demodulates the modal components using Gaussian smoothing constraints, and obtains the bandwidth of each modal component using the squared 2L norm of the demodulated modal component gradient. The prediction module is used to independently predict the bandwidth of each decomposed mode by combining multiple GRU predictors with attention mechanism, and to perform weighted summation of the hidden states at all time steps during prediction to obtain the wind speed prediction results of each modal component. The wind speed prediction results of each modal component are then input into a deep feedforward neural network for regression analysis to obtain the final wind speed prediction value.

8. A computer device, characterized in that, The device includes a memory and a processor, wherein the memory stores a program that, when executed by the processor, causes the processor to perform the steps of a port wind speed prediction method as described in any one of claims 1 to 6.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the port wind speed prediction method according to any one of claims 1 to 6.

Images