Tidal current generating capacity prediction method based on STL decomposition and multi-model fusion

Abstract

The invention discloses a power flow generating capacity prediction method based on STL decomposition and multi-model fusion, which adopts an STL decomposition method to perform trend term, season term and residual term decomposition on historical generating capacity data, optimizes the decomposition process through a Bisquare weighting function and an internal loop iteration mechanism, and enhances the processing capacity for abnormal values and high-frequency fluctuations. The method comprises the following steps: extracting multi-scale frequency domain features by using a TimeNet model in combination with fast Fourier transform (FFT), predicting a trend term and a seasonal term, and performing accurate modeling on a residual term by using an Itransform model in combination with a self-attention mechanism and gating residual connection. Meteorological factors related to the tidal current generating capacity are screened, prediction is carried out in combination with historical operation data, and high-precision prediction of the tidal current generating capacity is achieved. The method has high adaptability and robustness, can effectively deal with multi-source influence and high-frequency fluctuation in tidal current power generation data, and provides an accurate and stable prediction result.

Description

Technical Field

[0001] The present invention belongs to the technical field of power generation power prediction, and particularly relates to a tidal power generation prediction method based on STL decomposition and multi-model fusion. Background Art

[0002] As a clean and renewable energy, tidal energy is significantly affected by the periodic changes of tides, and has the characteristics of intermittency and volatility. After tidal energy is connected to the power system, due to its power generation process being affected by multiple factors such as tides and meteorology, it brings great challenges to the safe and economic dispatching of the power system. However, accurate prediction of tidal energy power generation can provide important reference for the dispatching plan, state estimation and real-time control of the power system, help optimize the active reserve capacity configuration on the power generation side and the demand-side dispatching plan, and improve the response ability of the power system to the access of tidal energy.

[0003] Currently, most tidal energy power generation prediction methods are based on statistical analysis techniques, that is, according to historical tide, meteorological data and power generation data, a prediction model is constructed, such as using multiple linear regression, time series analysis or grey theory prediction method. However, these methods often ignore the complex multi-period characteristics of tidal changes, the dynamic influence of meteorological factors and the differences in equipment operation states, resulting in insufficient prediction accuracy.

[0004] In the prior art, certain achievements have been made in photovoltaic power prediction methods. For example, Chinese Patent CN118657243A discloses a photovoltaic power prediction method based on multivariate time series decomposition and multi-model combination, which includes: correlation screening, seasonal trend decomposition, time-domain and frequency-domain attention, multi-model fusion prediction, and zero-output adjustment. However, these existing methods mainly target photovoltaic power generation systems and do not fully consider the special properties of tidal energy. There are several deficiencies in the prior art for predicting the power generation of tidal energy. First, when the existing STL decomposition method processes tidal energy data, it often fails to effectively handle extreme outliers and high-frequency fluctuations in the tidal energy data. For example, equipment failures or rare meteorological events (such as storm surges) may have a great impact on the power generation of tidal energy, but the prior art lacks a robust optimization mechanism for dealing with these extreme situations, which easily leads to the deviation of the trend term and the seasonal term. In addition, the data characteristics of tidal energy are relatively complex, and its periodic fluctuations include the superposition of multiple frequencies such as daily cycles and monthly cycles. The existing BiLSTM model can only capture a single periodic pattern well and is difficult to cope with the characteristics of multi-frequency superposition in tidal energy. Moreover, in the existing methods, the focus of frequency-domain feature extraction is usually optimized for short-term fluctuations (such as sudden changes in intra-day irradiation) in photovoltaic data, while the power generation characteristics of tidal energy involve more complex periodic changes such as semi-diurnal tides, diurnal tides, and seasonal tides. Therefore, the existing frequency-domain feature extraction technology fails to fully adapt to the special needs of tidal energy. Although the sunshine duration and weather changes in photovoltaic power generation are affected by seasonal changes to a certain extent, compared with tidal energy power generation, the periodicity of photovoltaic is relatively simple, and the requirements for the prediction model are also relatively single.

[0005] Therefore, there is an urgent need for a prediction method that can effectively capture the multi-period characteristics of tides to improve the accuracy and practicality of tidal energy power generation prediction. Summary of the Invention

[0006] The purpose of the present invention is to provide a tidal power generation prediction method based on STL decomposition and multi-model fusion to overcome the above-mentioned defects existing in the prior art.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] The present invention provides a tidal power generation prediction method based on STL decomposition and multi-model fusion, including the following steps:

[0009] Obtain historical power generation data, historical operation data, and historical meteorological data of a tidal energy power station;

[0010] Screen out historical meteorological factors related to tidal power generation from the historical meteorological data;

[0011] Perform STL decomposition on the historical power generation data to obtain a trend term, a seasonal term, and a residual term;

[0012] Input the trend term, the seasonal term, the historical operation data, and the selected historical meteorological factors into the pre-trained TimesNet model for fast Fourier transform and multi-scale modeling prediction to obtain predicted values of the trend term and the seasonal term;

[0013] Input the residual term, the historical operation data, and the selected historical meteorological factors into the pre-trained Itransformer model for sequence modeling and feature extraction to obtain a predicted value of the residual term;

[0014] Combine with the physical calculation formula of tidal energy, add the predicted values of the trend term, the seasonal term, and the residual term, and perform physical constraints to obtain the final predicted value of tidal current power generation.

[0015] Further, the historical power generation data is the actual hourly power generation data of the tidal current power station within a specified time period, and the specified time period is a time range of the past several days, weeks, months, etc., specifically including at least 24 hours of continuous data.

[0016] Further, the historical operation data includes: the swept area of the tidal current turbine rotor, the blade angle of the tidal current turbine, the generator speed, the generator power output, the operating status of the power generation equipment, and the power generation efficiency.

[0017] Further, the historical meteorological data includes: temperature, humidity, wind speed, tidal height, flow velocity, seawater density, atmospheric pressure, and wind direction.

[0018] Further, the historical meteorological factors selected from the historical meteorological data specifically include:

[0019] Record the historical hourly power generation data of the tidal current power station as the target variable X, and the historical hourly meteorological data as the meteorological variable Y. The historical hourly power generation data is obtained from the historical power generation data;

[0020] Sort each variable in X and Y separately from smallest to largest, and assign a rank to each observation value. If there are identical values, take their average rank. The rank is the position after sorting;

[0021] Calculate the rank difference d of each pair of observation values i , and the calculation formula is as follows:

[0022] d i =rank(X i )-rank(Y i )

[0023] where d iis the rank difference of the i-th group of observed values, and rank(X i ) is the rank of the i-th group of observed values in the target variable X, and rank(Y i ) is the rank of the i-th group of observed values in the meteorological variable Y;

[0024] Calculate the sum of squares of all rank differences, and use the Spearman correlation coefficient formula to calculate the correlation between each meteorological variable and the tidal power generation:

[0025]

[0026] where γ is the Spearman correlation coefficient and n is the number of observation samples;

[0027] Judge the correlation between each meteorological variable and the tidal power generation. If the absolute value of the Spearman correlation coefficient γ is greater than the first preset value, it is considered that the meteorological variable has a significant correlation with the tidal power generation, and it is used as the selected historical meteorological factor.

[0028] Furthermore, the STL decomposition of the historical power generation data is performed to obtain a trend term, a seasonal term, and a residual term, specifically including:

[0029] Decompose the historical hourly power generation data X t of the tidal power station, initialize the trend term T t = 0, initialize the seasonal term S t = 0, initialize the residual term R t = X t - T t - S t , where t = 1, 2, 3... n and n is the length of the time series;

[0030] Assign weights to each residual value through the Bisquare weight function, and the formula is:

[0031]

[0032] where R t represents the residual term, median(·) represents taking the median, and ω t is the weight;

[0033] Gradually optimize the trend term and the seasonal term through multiple iterations of the inner loop according to the weights until convergence, and obtain the iterated trend term, seasonal term, and residual term.

[0034] Furthermore, the step of gradually optimizing the trend term and the seasonal term through multiple iterations of the inner loop according to the weights specifically includes:

[0035] Step A1: Subtract the currently estimated trend term T from the historical hourly power generation data X t ​t , detrended data is obtained:

[0036]

[0037] wherein, is the detrended data;

[0038] Step A2: Divide the detrended data into P subsequences according to the seasonal cycle P. For example, if the data is 24-hour data per day, then P = 24;

[0039] Step A3: Use locally weighted regression to model each subsequence and estimate the preliminary seasonal term Regression is performed by minimizing the weighted sum of squared residuals:

[0040]

[0041] Step A4: Perform moving average smoothing on the preliminary seasonal term to remove the long-term trend and obtain the smoothed seasonal term, and obtain the smoothed seasonal term Adjust the smoothed seasonal term to make its mean zero, and obtain the updated seasonal term

[0042]

[0043] wherein, P is the number of divided subsequences;

[0044] Step A5: Subtract the updated seasonal term from the original data to obtain the deseasonalized data

[0045]

[0046] Step A6: Perform weighted low-pass filtering on the deseasonalized data to estimate the updated trend term

[0047]

[0048] wherein, t is the index of the time point and m is half the length of the sliding window;

[0049] Step A7: Repeat Step A1 to Step A6 until the convergence condition is met:

[0050] and

[0051] wherein, Denote the new trend term and new seasonal term after one iteration. Denote the trend term and seasonal term calculated during the previous iteration. After convergence, and are used as the trend term and seasonal term, and the residual term after iteration is obtained

[0052] Furthermore, input the trend term, seasonal term, historical operation data, and the selected historical meteorological factors into the pre-trained TimesNet model for fast Fourier transform and multi-scale modeling prediction to obtain the predicted values of the trend term and seasonal term. Specifically, it includes:

[0053] Take the trend term, seasonal term, historical operation data, and the selected historical meteorological factors as the original time series and input them into the pre-trained TimesNet model;

[0054] In the TimesNet model, use the fast Fourier transform (FFT) to perform frequency domain conversion on the input data, obtain the amplitude spectrum of each frequency component, and identify the dominant periods and frequency characteristics with significant energy contributions;

[0055] According to the identified dominant period parameters, perform period folding on the original time series, convert the input data from a one-dimensional time series to a two-dimensional tensor, where the row dimension represents the period length, the column dimension represents the number of periods, and the periodicity and period evolution characteristics in the original time series;

[0056] In the multi-scale modeling process, through the multi-level neural network structure in the TimesNet model, use convolutional kernels of different scales to extract features from this two-dimensional tensor, perform deep modeling at different levels from short-term fluctuations to long-term trends, and finally output the predicted values of the trend term and seasonal term.

[0057] Furthermore, input the residual term, historical operation data, and the selected historical meteorological factors into the pre-trained Itransformer model for sequence modeling and feature extraction to obtain the predicted value of the residual term. Specifically, it includes:

[0058] Perform time modeling on the residual term, historical operation data, and the selected historical meteorological factors to obtain a time series, where each time step contains multiple features, corresponding to the residual term, historical operation data, and historical meteorological factors respectively;

[0059] Input the time series into the pre-trained Itransformer model. The Itransformer model processes the input features through a linear projection layer to generate a query matrix Q, a key matrix K, and a value matrix V respectively. The query matrix Q is generated from the residual term data, and the key matrix K and value matrix V are generated from the historical operation data and historical meteorological factors;

[0060] Using the query matrix Q, key matrix K, and value matrix V, the Itransformer model adopts a reverse attention mechanism to calculate the similarity scores in the time step dimension;

[0061] Based on the calculated similarity scores, the Itransformer model calculates the attention weights through a softmax operation;

[0062] Multiply the attention weights by the value matrix V to generate a weighted feature representation;

[0063] The Itransformer model processes the weighted feature representation through multi-layer self-attention calculation and feature aggregation to generate a preliminary residual term prediction value; adopts a gated residual connection to perform a non-linear transformation on the preliminary residual term prediction value and the input residual term data, combines layer normalization to normalize the hidden states of different layers, and maps the processed residual features through the output layer to obtain the final residual term prediction value.

[0064] Furthermore, by combining the physical calculation formula of tidal energy, adding the trend term prediction value, seasonal term prediction value, and residual term prediction value, and performing physical constraints, the final tidal power generation prediction value is obtained, specifically including:

[0065] Add the trend term prediction value, seasonal term prediction value, and residual term prediction value to obtain the prediction value of tidal power generation:

[0066] P pred =P trend +P seasonal +P residual

[0067] where P trend represents the trend term prediction value, P seasonal represents the seasonal term prediction value, P residual the residual term prediction value, and P pred is the prediction value of tidal power generation;

[0068] Calculate the maximum value of the real-time theoretical power according to the physical calculation formula of tidal energy:

[0069]

[0070] where ρ represents the seawater density, A represents the swept area of the tidal current turbine rotor, V represents the tidal current velocity, η represents the power generation efficiency of the generator, and P max is the maximum value of the real-time theoretical power;

[0071] Perform physical compliance constraints on the superimposed tidal power generation prediction value:

[0072] If Ppred >P max then force the predicted value to be corrected to P max , that is: P pred =P max ;

[0073] If the real-time flow velocity V is less than the turbine startup threshold, then the output power is set to zero, that is: P pred =0;

[0074] Take the corrected P after physical constraint pred as the final predicted value of the tidal current power generation.

[0075] Compared with the prior art, the present invention has the following advantages:

[0076] (1) By introducing an improved STL decomposition method and using a Bisquare weight function to dynamically downweight extreme outliers, the present invention improves the robustness of the time series decomposition of tidal current energy data, effectively reduces the interference of outliers on the trend term and the seasonal term, so that the prediction model can still maintain a stable trend extraction ability in the face of sudden meteorological events (such as storm surges, equipment failures), and improves the accuracy and reliability of the prediction.

[0077] (2) By introducing an iterative optimization mechanism into the inner loop of the STL decomposition, the present invention improves the convergence stability of the trend term, realizes accurate modeling for the high-frequency fluctuations and non-stationary signals of tidal energy data, makes the decomposed trend term smoother and more in line with the periodic characteristics of tidal current energy, and improves the interpretability and generalization ability of subsequent modeling.

[0078] (3) By adopting the TimesNet model and combining with the fast Fourier transform (FFT) to extract the multi-scale frequency domain features of tidal current energy data, the present invention effectively captures the multi-frequency superposition characteristics such as semi-diurnal tides, diurnal tides, and monthly cycles in tidal energy data, solves the defect that the existing BiLSTM model is difficult to adapt to the complex periodicity of tidal current energy, and significantly improves the adaptability of the prediction model to periodic signals of different scales.

[0079] (4) By introducing an improved self-attention network iTransformer into the residual term modeling and using a dynamic weight allocation mechanism, the present invention realizes the accurate extraction of non-periodic fluctuation patterns in the residual term, solves the problem of local receptive field limitation caused by the causal convolution structure of the TCN (temporal convolutional network), enables the model to better capture the random fluctuations of tidal current power generation affected by the external environment, and improves the prediction accuracy.

[0080] (5) By adopting the feature enhancement technology targeting the characteristics of tidal energy, the present invention optimizes the frequency-domain feature extraction method, realizes more accurate periodic feature modeling compared with the existing wavelet transform method, enables the prediction model to still efficiently extract key frequency information in the face of the superposition of multiple tidal patterns, and improves the generalization ability and prediction stability of the model.

[0081] (6) By adopting the fast Fourier transform (FFT), the present invention explicitly models the coupling effect of short-term tidal fluctuations and long-term meteorological trends in tidal energy power generation, realizes multi-scale feature extraction of tidal energy data, enables the prediction model to simultaneously consider the influence of short-period tidal fluctuations (such as semi-diurnal tides) and long-period meteorological factors (such as seasonal changes), and thus improves the adaptability to complex tidal energy power generation scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 is a schematic flowchart of the tidal power generation prediction method provided by an embodiment of the present invention;

[0083] Figure 2 is a comparison chart of predicted values and true values provided by an embodiment of the present invention;

[0084] Figure 3 is a schematic structural diagram of the tidal power generation prediction method provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0085] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0086] Embodiment 1:

[0087] This embodiment provides a tidal power generation prediction method based on STL decomposition and multi-model fusion, as Figure 1 shown, including the following steps:

[0088] Obtain the historical power generation data, historical operation data, and historical meteorological data of the tidal energy power station;

[0089] Screen out the historical meteorological factors related to tidal power generation from the historical meteorological data;

[0090] Perform STL decomposition on the historical power generation data to obtain a trend term, a seasonal term, and a residual term;

[0091] Input the trend term, seasonal term, historical operation data, and the selected historical meteorological factors into the pre-trained TimesNet model for fast Fourier transform and multi-scale modeling prediction to obtain the predicted values of the trend term and the seasonal term;

[0092] Input the residual term, historical operation data, and the selected historical meteorological factors into the pre-trained Itransformer model for sequence modeling and feature extraction to obtain the predicted value of the residual term;

[0093] Combine with the physical calculation formula of tidal energy, add the predicted values of the trend term, seasonal term, and residual term, and perform physical constraints to obtain the final predicted value of the tidal power generation.

[0094] Further, the historical power generation data is the actual hourly power generation data of the tidal power station within a specified time period, and the specified time period is a time range of the past several days, weeks, months, etc., specifically including at least 24 hours of continuous data.

[0095] Using continuous historical data helps to avoid the problem of data fragmentation. In tidal energy power generation prediction, if discrete time points are used for training, it may cause the model to fail to correctly understand the change trend of tides. By using at least 24 hours of continuous data, it can ensure the integrity of the data when the model learns the time series, making it easier to capture the timing characteristics of tidal energy power generation and improving the prediction effect.

[0096] Further, the historical operation data includes: the swept area of the tidal current turbine rotor, the blade angle of the tidal current turbine, the generator speed, the generator power output, the operation status of the power generation equipment, and the power generation efficiency.

[0097] In tidal energy power generation prediction, in addition to environmental factors (such as tidal current speed, tidal level change), the operation status and performance parameters of the power generation equipment are also important factors determining the power generation. In the selection of historical operation data in the present invention, it covers multiple key variables such as the swept area of the tidal current turbine rotor, blade angle, generator speed, generator power output, operation status of the power generation equipment, and power generation efficiency. The reasonable selection and utilization of these variables help to improve the prediction accuracy and enhance the adaptability of the model. The prediction model can not only consider the influence of the tidal current environment, but also comprehensively analyze the operation characteristics of the equipment itself, so as to achieve more refined tidal energy power generation prediction and improve the prediction accuracy and applicability.

[0098] Further, the historical meteorological data includes: temperature, humidity, wind speed, tidal height, flow velocity, seawater density, atmospheric pressure, and wind direction.

[0099] Further, the historical meteorological factors selected from the historical meteorological data specifically include:

[0100] Record the historical hourly power generation data of the tidal current power station as the target variable X, and the historical hourly meteorological data as the meteorological variable Y. The historical hourly power generation data is obtained from the historical power generation data.

[0101] Sort each variable in X and Y separately from smallest to largest, and assign a rank to each observation. If there are identical values, take the average rank. The rank is the position after sorting.

[0102] Calculate the rank difference d of each pair of observations i , and the calculation formula is as follows:

[0103] d i = rank(X i ) - rank(Y i )

[0104] where d i is the rank difference of the i-th group of observations, rank(X i ) is the rank of the i-th group of observations in the target variable X, and rank(Y i ) is the rank of the i-th group of observations in the meteorological variable Y;

[0105] Calculate the sum of the squares of all rank differences, and use the Spearman correlation coefficient formula to calculate the correlation between each meteorological variable and the tidal current power generation:

[0106]

[0107] where γ is the Spearman correlation coefficient and n is the number of observation samples;

[0108] Judge the correlation between each meteorological variable and the tidal current power generation. If the absolute value of the Spearman correlation coefficient γ is greater than the first preset value, it is considered that the meteorological variable has a significant correlation with the tidal current power generation, and it is used as the selected historical meteorological factor.

[0109] In the prediction of tidal current power generation, the influence of meteorological factors on power generation has a complex non-linear relationship. Therefore, not all meteorological variables contribute significantly to the prediction. If all historical meteorological data is directly used for modeling, redundant information or noise may be introduced, reducing the generalization ability and computational efficiency of the model. Therefore, the present invention screens meteorological factors highly correlated with tidal current power generation through the Spearman rank correlation coefficient method to improve the effectiveness of data and enhance the accuracy and robustness of the prediction model. The technical effect of improving the accuracy and stability of tidal current power generation prediction is achieved. By only retaining meteorological variables highly correlated with tidal current power generation, the interference of irrelevant features on model training is reduced, the computational efficiency of the model is improved, and its adaptability to complex tidal current characteristics is enhanced. This method can not only more accurately describe the influence of meteorological conditions on tidal current power generation, but also improve the interpretability of the prediction system, making it more valuable in practical applications.

[0110] Further, the STL decomposition of the historical power generation data to obtain a trend term, a seasonal term, and a residual term specifically includes:

[0111] Decompose the historical hourly power generation data X t of the tidal current power station, and initialize the trend term T t = 0, initialize the seasonal term S t = 0, initialize the residual term R t = X t - T t - S t , where t = 1, 2, 3... n, and n is the length of the time series;

[0112] Assign weights to each residual value through the Bisquare weight function, and the formula is:

[0113]

[0114] where, R t represents the residual term, median(·) represents taking the median, and ω t is the weight;

[0115] According to the weights, gradually optimize the trend term and the seasonal term through multiple iterations of the inner loop until convergence, and obtain the iterated trend term, seasonal term, and residual term.

[0116] Further, the step of gradually optimizing the trend term and the seasonal term according to the weights through multiple iterations of the inner loop specifically includes:

[0117] Step A1: Subtract the currently estimated trend term T t from the historical hourly power generation data X t to obtain detrended data:

[0118]

[0119] Among them, is the detrended data;

[0120] Step A2: Divide the detrended data into P subsequences according to the seasonal period P. For example, if the data is 24-hour data per day, then P = 24;

[0121] Step A3: Use locally weighted regression to model each subsequence and estimate the preliminary seasonal term Perform regression by minimizing the weighted sum of squared residuals:

[0122]

[0123] Step A4: Perform moving average smoothing on the preliminary seasonal term to remove the long-term trend and obtain the smoothed seasonal term, obtaining the smoothed seasonal term Adjust the smoothed seasonal term to make its mean zero and obtain the updated seasonal term

[0124]

[0125] where P is the number of divided subsequences;

[0126] Step A5: Subtract the updated seasonal term from the original data to obtain the deseasonalized data

[0127]

[0128] Step A6: Perform weighted low-pass filtering on the deseasonalized data to estimate the updated trend term

[0129]

[0130] where t is the index of the time point and m is half the length of the sliding window;

[0131] Step A7: Repeat Step A1 to Step A6 until the convergence condition is met:

[0132] and

[0133] Among them, represents the new trend term and new seasonal term after one iteration, Denote the trend term and seasonal term calculated in the previous iteration, and after convergence and as the trend term and seasonal term, and obtain the residual term after iteration

[0134] In the prediction of tidal current power generation, tidal current data usually has obvious seasonal and trend characteristics, and there may be high-frequency fluctuations and outliers in the data. If the original data is directly used for modeling, the model may be affected by these noises, resulting in a decrease in prediction accuracy. Therefore, the present invention adopts the STL (Seasonal-Trend Decomposition Method) method to decompose the historical power generation data, and separately extracts the trend term, seasonal term and residual term, so as to better capture the long-term trend, seasonal fluctuations and non-periodic changes in the residual part of tidal current power generation. This method can more clearly understand the time-varying characteristics of tidal current power generation, and provide more accurate and clean input features for subsequent prediction modeling.

[0135] The Bisquare weight function is used to downweight extreme outliers (such as sudden equipment failures, strong storm surges, etc.) to avoid these outliers from contaminating the decomposition process of the trend term and seasonal term, and ensure that the trend and seasonal components can more accurately reflect the long-term and periodic change characteristics of tidal current. This step can effectively improve the robustness of the decomposition process and reduce the sensitivity to outliers.

[0136] The role of the inner loop process in the present invention is mainly to optimize the trend term and seasonal term through multiple iterations in the STL decomposition, so that the decomposition result is more accurate and can better reflect the law in the tidal current power generation data. This process includes operations such as detrending, seasonalizing and deseasonalizing, and is adjusted according to the weighted residual in each iteration to ensure that the matching degree of the trend term and seasonal term with the actual data reaches the optimum. The inner loop process can effectively eliminate short-term and long-term fluctuations in the data, improve the accuracy of the trend term and seasonal term, and avoid the interference of external abnormal data on the model. Secondly, using the Bisquare weight function to weight the residual can significantly reduce the impact of extreme outliers on the result, thereby enhancing the robustness of the model. Finally, this optimization process can converge to the optimal solution through continuous iteration, improving the accuracy of the prediction of tidal current power generation. Especially when dealing with tidal current data with strong periodicity and volatility, it can provide more stable and reliable prediction results.

[0137] Further, inputting the trend term, seasonal term, historical operation data and the selected historical meteorological factors into the pre-trained TimesNet model, performing fast Fourier transform and multi-scale modeling prediction, and obtaining the trend term prediction value and seasonal term prediction value specifically includes:

[0138] Take the trend term, seasonal term, historical operation data, and the selected historical meteorological factors as the original time series and input them into the pre-trained TimesNet model;

[0139] In the TimesNet model, use the fast Fourier transform (FFT) to perform frequency domain conversion on the input data, obtain the amplitude spectra of each frequency component, and identify the dominant periods and frequency characteristics with significant energy contributions;

[0140] According to the identified dominant period parameters, perform period folding on the original time series, convert the input data from a one-dimensional time series to a two-dimensional tensor, where the row dimension represents the period length and the column dimension represents the number of periods, and the periodicity and period evolution characteristics in the original time series;

[0141] In the multi-scale modeling process, through the multi-level neural network structure in the TimesNet model, use convolutional kernels of different scales to extract features from this two-dimensional tensor, perform deep modeling at different levels from short-term fluctuations to long-term trends, and finally output the predicted values of the trend term and the seasonal term.

[0142] In the present invention, by inputting the trend term, seasonal term, historical operation data, and meteorological factors into the pre-trained TimesNet model, combining the fast Fourier transform (FFT) and multi-scale modeling, accurate prediction of the tidal current power generation is achieved. The FFT is used to extract frequency domain features and identify key periodic fluctuations. Period folding converts the time series into a two-dimensional tensor, facilitating the model to capture multi-period changes. Through the multi-level neural network structure, the model extracts data features at different scales, thereby accurately predicting the trend term and the seasonal term. This method can effectively improve the prediction accuracy of tidal current power generation, especially having significant advantages when dealing with complex periodic changes.

[0143] Furthermore, inputting the residual term, historical operation data, and the selected historical meteorological factors into the pre-trained Itransformer model for sequence modeling and feature extraction to obtain the predicted value of the residual term specifically includes:

[0144] Perform time modeling on the residual term, historical operation data, and the selected historical meteorological factors to obtain a time series, where each time step contains multiple features, corresponding to the residual term, historical operation data, and historical meteorological factors respectively;

[0145] Input the time series into the pre-trained Itransformer model. The Itransformer model processes the input features through a linear projection layer to generate a query matrix Q, a key matrix K, and a value matrix V respectively. The query matrix Q is generated from the residual term data, and the key matrix K and the value matrix V are generated from the historical operation data and the historical meteorological factors;

[0146] Using the query matrix Q, the key matrix K, and the value matrix V, the Itransformer model adopts a reverse attention mechanism to calculate the similarity scores in the time step dimension;

[0147] Based on the calculated similarity scores, the Itransformer model calculates the attention weights through a softmax operation;

[0148] Multiply the attention weights by the value matrix V to generate a weighted feature representation;

[0149] The Itransformer model processes the weighted feature representation through multi-layer self-attention calculation and feature aggregation to generate a preliminary residual term prediction value; adopts a gated residual connection to perform a non-linear transformation on the preliminary residual term prediction value and the input residual term data, combines layer normalization to normalize the hidden states of different layers, and maps the processed residual features through an output layer to obtain the final residual term prediction value.

[0150] The present invention inputs the residual term, historical operation data, and selected meteorological factors into a pre-trained Itransformer model for time series modeling and feature extraction. The model generates a query matrix, a key matrix, and a value matrix, and uses a reverse attention mechanism to calculate the similarity between time steps to capture the key dependencies in the data. By weighting each time step, the model can effectively identify and model the complex non-linear variation patterns of the residual term.

[0151] During the prediction process, the Itransformer adopts multi-layer self-attention calculation and feature aggregation to generate a preliminary residual term prediction value, and optimizes it through gated residual connection and layer normalization techniques. These steps can enhance the stability and prediction accuracy of the model, especially when dealing with complex time series, significantly improving the accuracy of residual term prediction, thus providing reliable support for the prediction of tidal power generation.

[0152] Further, combining with the physical calculation formula of tidal energy, adding the trend term prediction value, the seasonal term prediction value, and the residual term prediction value, and performing physical constraints to obtain the final tidal power generation prediction value, specifically including:

[0153] Add the trend term prediction value, the seasonal term prediction value, and the residual term prediction value to obtain the prediction value of tidal power generation:

[0154] P pred =P trend +P seasonal +P eeSidual

[0155] Among them, P trend represents the trend term prediction value, P seasonalRepresents the predicted value of the seasonal item, P residual Residual term predicted value, P pred Is the predicted value of the tidal power generation;

[0156] Calculate the maximum value of the real-time theoretical power according to the physical calculation formula of tidal energy:

[0157]

[0158] Among them, ρ represents the seawater density, A represents the swept area of the tidal current turbine rotor, V represents the tidal current velocity, η represents the power generation efficiency of the generator, P max Is the maximum value of the real-time theoretical power;

[0159] Conduct physical compliance constraints on the superimposed predicted value of tidal power generation:

[0160] If P pred >P max , then forcibly correct the predicted value to P max , that is: P pred =P max ;

[0161] If the real-time flow velocity V is less than the turbine startup threshold, the output power is set to zero, that is: P pred =0;

[0162] Take the P pred after physical constraint correction as the final predicted value of tidal power generation.

[0163] It should be noted that the current prediction methods for tidal power generation mainly include methods based on traditional statistical analysis methods and physical models. Based on traditional statistical analysis methods, that is, according to the historical data and environmental factors of tidal power plants, power generation prediction models are constructed, such as using multiple linear regression, time series analysis or simple regression models. However, such methods have some significant disadvantages. First, traditional methods often ignore the complexity of tidal power generation, such as changes in tidal speed and direction, and the mutual influence between tidal power generators. The impact of these factors on power generation is dynamic, and traditional models are difficult to capture these changes, resulting in insufficient accuracy of prediction results. Secondly, methods based on statistical analysis usually rely on the linear relationship of historical data and cannot effectively handle nonlinear and complex time series characteristics. This significantly reduces the prediction ability of the model when facing emergencies or extreme weather conditions. In addition, traditional methods often lack flexibility when dealing with seasonal and periodic changes in tidal power generation, and cannot adapt to the changing characteristics of tidal energy resources. For example, periodic changes in tidal currents may cause prediction deviations of the model in certain time periods. Finally, traditional statistical methods usually require a large amount of historical data for training during model construction, while in the field of tidal power generation, data acquisition and quality may be limited. This affects the generalization ability of the model and leads to large prediction errors in practical applications.

[0164] The tidal energy power generation prediction method based on STL decomposition and multi-model fusion combined with physical constraints proposed in this study shows multi-dimensional breakthrough value in terms of technical path and theoretical innovation. This method decouples the original power generation sequence into three components with clear physical meanings: trend term, seasonal term and residual term through the STL (Seasonal-Trend decomposition using Loess) algorithm, and constructs a differentiated prediction model for the dynamic characteristics of each component: First, the Timesnet deep neural network is used to capture the nonlinear evolution law of the trend term and the periodic oscillation characteristics of the seasonal term. Its bidirectional cyclic structure can effectively model long-term dependencies, which is particularly suitable for processing low-frequency components affected by astronomical tidal cycles and seasonal energy density fluctuations in tidal energy generation; secondly, the iTransformer model is innovatively introduced to process high-frequency residual terms, and the feature space is reconstructed through an improved attention mechanism. Its sparse self-attention layer can accurately capture the non-steady-state random disturbances implicit in the residual sequence, solving the error accumulation problem caused by noise interference in traditional residual prediction. What is particularly critical is that this method introduces fluid mechanics constraints in the prediction result fusion stage. By constructing a physical boundary model of power output based on the tidal energy equation, the data-driven prediction values are constrained within the physically feasible domain such as the maximum capture efficiency of the turbine and the water velocity threshold, thus achieving a deep integration of data intelligence and domain knowledge.

[0165] The innovative advantages of this method are mainly reflected in the following four dimensions:

[0166] First, the heterogeneous model collaboration mechanism for component prediction breaks through the performance bottleneck of a single model. The STL decomposition enables Timenet and iTransformer to focus on modeling different frequency features respectively, improving the analytical ability for complex coupled features compared with traditional end-to-end models;

[0167] Second, the innovative application of iTransformer in residual prediction significantly improves the feature extraction efficiency in high-frequency noise environments through dynamic sparse attention weight allocation and residual connection optimization;

[0168] Third, the embedding mechanism of physical constraints creates a new fusion paradigm of data-driven and mechanism models. By constructing a differentiable physical loss function layer, the mathematical accuracy and physical rationality of the prediction results are synchronously optimized during the backpropagation process, solving the problem of deviation from physical laws that may occur in pure data-driven methods;

[0169] Fourth, the model architecture has the advantage of strong interpretability. The prediction results of each component after STL decomposition can be independently verified and correlated with ocean environmental parameters, providing a basis for operation and maintenance decisions;

[0170] Example 2:

[0171] Parts not mentioned in this example are the same as those in Example 1.

[0172] The following are application examples of the present invention in specific products or related technologies:

[0173] 1. Dataset collection

[0174] The dataset uses historical data from a tidal current power station on the east coast of China. The time is from 0:00 on January 1, 2023 to 0:00 on January 1, 2024. The data sampling frequency is 1 hour, with a total of 8760 data points, including 9 different feature data: power generation, swept area of the tidal current turbine rotor, power generation efficiency, temperature (°C), humidity (%), wind speed (m / s), tidal height (m), flow velocity (m / s), seawater density (kg / m 3 )), wind direction (°).

[0175] One year's data is taken as the data sample for model prediction. The training set and test set are divided in a ratio of 8:2, and all data is preprocessed for model training.

[0176] 2. Experiment

[0177] 2.1 Calculate the Spearman correlation coefficient:

[0178] Table 1 Results of correlation calculation

[0179] Characteristic Coefficient Temperature (°C) -0.2 Humidity (%) -0.1 Wind speed (m / s) 0.31 Tide height (m) 0.74 Flow velocity (m / s) 0.83 <![CDATA[Seawater density (kg / m 3 )]]> 0.12 Wind direction (°) 0.3

[0180] According to the results in Table 1, select the features with the absolute value of the Spearman correlation coefficient greater than 0.5, which are: tide height (m), flow velocity (m / s).

[0181] 2.2 Evaluation indicators

[0182] To verify the effectiveness of this method, different types of evaluation indicators are selected, and the mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R2) are used to evaluate the degree of agreement between the predicted value and the true value. The calculation formulas of MAE, RMSE, and R2 are as follows:

[0183]

[0184] Where: n is the prediction length, y i is the actual power value of tidal energy power generation at time i, is the power prediction value at time i, and is the average value of the actual tidal energy power, represents the average value of power prediction.

[0185] In the performance evaluation of the tidal energy power generation prediction model, the values of MAE (mean absolute error) and RMSE (root mean square error) are negatively correlated with the prediction accuracy, that is, the smaller the index value, the lower the prediction deviation of the model for power fluctuations; R 2 (coefficient of determination) is used to measure the explanatory ability of the predicted value to the change of the actual value, and its value range is [0,1]. When R 2 approaches 1, it indicates that the model can fully explain the change law of the target variable, while R 2 approaching 0 reflects that there is no linear correlation between the prediction result and the actual value. These three indicators quantify the model performance from three dimensions: the absolute error magnitude (MAE), the sensitivity of error distribution (RMSE), and the trend explanatory ability (R 2 ). MAE intuitively reflects the average deviation degree between the predicted value and the true value. RMSE amplifies the weight of large error terms through square operation to evaluate the robustness of the model. R 2 reveals the correlation strength between the input features and the target variable from a statistical perspective. The joint verification of the three can avoid the limitations of a single indicator, form a complement in error distribution analysis, model generalization ability evaluation, and physical law consistency test, and ensure the objectivity and comprehensiveness of the evaluation results.

[0186] 2.3 Result comparison

[0187] To determine the optimal model architecture for predicting the power generation of tidal current energy, this study conducted comparative experiments and comprehensive evaluations on various prediction methods.

[0188] Table 2 Comparison results

[0189]

[0190]

[0191] The comparison results in Table 2 show that the method of this patent is superior to the comparative models in all indicators. Among them, MAE (0.0592), RMSE (0.0828), MSE (0.0068) and R 2 (0.9484) all reach the optimal values, indicating its significant advantages in prediction accuracy and stability. As Figure 2 shown in the comparison diagram of predicted value and true value.

[0192] Compared with the benchmark model LSTM, the MAE of the method of this patent is reduced by 17.9%, the RMSE is reduced by 7.5%, and R 2 is increased by 2.3%, verifying the effectiveness of multivariate time series decomposition and multi-model combination.

[0193] After removing the physical constraints, the MAE of the model increases to 0.0608, the RMSE increases to 0.0843, and R 2 drops to 0.9421, verifying the important role of Betz power constraint and flat tide period correction in the physical rationality of prediction results.

[0194] When using only TimesNet or iTransformer for individual prediction, the MAE is 0.0627 and 0.0632 respectively, and the RMSE is 0.0893 and 0.0875 respectively, both lower than the method of this patent; while the combination of iTransformer + TimesNet improves the performance (MAE = 0.0615, RMSE = 0.0842), but still does not reach the level of the method of this patent, proving that the collaborative optimization of multi-model combination and physical constraints is the key.

[0195] The method of this patent significantly improves the accuracy and reliability of tidal current energy power generation prediction through the organic combination of multivariate time series decomposition, multi-model combination and physical constraints. Ablation experiments further verify the contributions of each module, indicating that STL decomposition, physical constraints and multi-model collaborative optimization are the key factors for improving model performance.

[0196] Through the above steps, the tidal current power generation prediction system can accurately and timely predict the future power generation of tidal current power plants and make corresponding adjustments according to real-time data.

[0197] In summary, this method decomposes the tidal current power generation data into trend, seasonal, and residual components through STL decomposition, combines the TimesNet and Itransformer models for prediction, and then combines physical constraints to finally obtain accurate power generation prediction results. This method can effectively identify various factors affecting tidal current power generation and make dynamic adjustments based on real-time data, thereby improving the accuracy and reliability of the prediction.

[0198] This embodiment also provides a tidal current power generation station power generation prediction system, as Figure 3 shown, including

[0199] a data acquisition module for acquiring the historical operation data and historical meteorological data of the tidal current power generation station, as well as the recent data of the tidal current power generation station

[0200] a historical meteorological factor selection module for selecting historical meteorological factors related to the power generation of the tidal current power generation station from the historical meteorological data;

[0201] a historical power generation selection module for selecting historical power generation related to the historical meteorological factors from the historical operation data;

[0202] a model training module for training the initial power generation prediction model using the historical meteorological factors and the historical power generation.

[0203] a model prediction module for predicting the future power generation of the field to obtain tidal current power generation station power generation data.

[0204] It should be noted that the tidal current power generation station power generation prediction system provided in the embodiment of the present invention is used to execute all the process steps of the tidal current power generation station power generation prediction method in the above embodiment, and the working principles and beneficial effects of the two correspond one by one, so they will not be elaborated here.

[0205] This embodiment also provides an electronic device. The electronic device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a tidal current power generation prediction program. When the processor executes the computer program, it implements the steps in the above-mentioned various embodiments of the tidal current power generation prediction method. Or, when the processor executes the computer program, it implements the functions of each module / unit in the above-mentioned system embodiments, such as the historical meteorological factor selection module.

[0206] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0207] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for predicting the power generation of tidal current based on STL decomposition and multi-model fusion, characterized in that Including the following steps: Obtain the historical power generation data, historical operation data, and historical meteorological data of the tidal current power station; Screen out the historical meteorological factors related to tidal current power generation from the historical meteorological data; Perform STL decomposition on the historical power generation data to obtain a trend term, a seasonal term, and a residual term; Input the trend term, seasonal term, historical operation data, and the screened historical meteorological factors into a pre-trained TimesNet model for fast Fourier transform and multi-scale modeling prediction to obtain predicted values of the trend term and seasonal term; Input the residual term, historical operation data, and the screened historical meteorological factors into a pre-trained Itransformer model for sequence modeling and feature extraction to obtain a predicted value of the residual term; Combine with the physical calculation formula of tidal energy, add the predicted values of the trend term, seasonal term, and residual term, and perform physical constraints to obtain the final predicted value of tidal current power generation; 2. The method for predicting tidal power generation based on STL decomposition and multi-model fusion according to claim 1, wherein The historical power generation data is the actual hourly power generation data of the tidal current power station within a specified time period; 3. A method for predicting the power generation of tidal current based on STL decomposition and multi-model fusion according to claim 1, characterized in that The historical operation data includes: the swept area of the tidal current turbine rotor, the blade angle of the tidal current turbine, the generator speed, the generator power output, the operating state of the power generation equipment, and the power generation efficiency; 4. A method for predicting the power generation of tidal current based on STL decomposition and multi-model fusion according to claim 1, characterized in that, The historical meteorological data includes: temperature, humidity, wind speed, tidal height, flow velocity, seawater density, atmospheric pressure, and wind direction; 5. A method for predicting the power generation of a tidal current based on STL decomposition and multi-model fusion according to claim 1, characterized in that The screening of the historical meteorological factors related to tidal current power generation from the historical meteorological data specifically includes: Denote the historical hourly power generation data of the tidal current power station as the target variable X, and the historical hourly meteorological data as the meteorological variable Y. The historical hourly power generation data is obtained from the historical power generation data; Sort each variable in X and Y separately from smallest to largest, and assign a rank to each observation value. If there are identical values, take their average rank. The rank is the position after sorting; Calculate the rank difference d for each pair of observations i , and the calculation formula is as follows: d i = rank(X i ) - rank(Y i ) where d i is the rank difference of the i-th group of observed values, rank(X i ) is the rank of the i-th group of observed values in the target variable X, and rank(Y i ) is the rank of the i-th group of observed values in the meteorological variable Y; Calculate the sum of the squares of all rank differences, and use the Spearman correlation coefficient formula to calculate the correlation between each meteorological variable and tidal current power generation: where γ is the Spearman correlation coefficient and n is the number of observation samples; Judge the correlation between each meteorological variable and tidal current power generation. If the absolute value of the Spearman correlation coefficient γ is greater than a first preset value, it is considered that the meteorological variable has a significant correlation with tidal current power generation, and it is used as the screened historical meteorological factor; 6. The power generation prediction method of tidal current based on STL decomposition and multi-model fusion according to claim 1, wherein The STL decomposition of the historical power generation data to obtain a trend term, a seasonal term, and a residual term specifically includes: For the historical hourly power generation data X of a tidal current power station t perform decomposition, initialize the trend term T t = 0, initialize the seasonal term S t = 0, initialize the residual term R t = X t - T t - S t , where t = 1, 2, 3... n, and n is the length of the time series; Assign weights to each residual value through a Bisquare weight function, and the formula is: where R t represents the residual term, median(·) represents taking the median, and ω t is the weight; Gradually optimize the trend term and seasonal term through multiple iterations of the inner loop according to the weights until convergence, and obtain the iterated trend term, seasonal term, and residual term; 7. A method for predicting the power generation of tidal current based on STL decomposition and multi-model fusion according to claim 6, characterized in that, The gradual optimization of the trend term and seasonal term through multiple iterations of the inner loop according to the weights specifically includes: Step A1: Subtract the currently estimated trend term T from the historical hourly power generation data X t to obtain the detrended data: t ​ Among them, is detrended data; Step A2: Detrended data is divided into P subsequences according to the seasonal cycle P. For example, if the data is 24-hour data per day, then P = 24; Step A3: Model each subsequence using locally weighted regression to estimate the preliminary seasonal terms Perform regression by minimizing the weighted sum of squared residuals: Step A4: For the preliminary seasonal term Perform moving average smoothing to remove the long-term trend, obtaining the smoothed seasonal term, obtaining the smoothed seasonal term Adjust the smoothed seasonal term To make its mean zero, obtaining the updated seasonal term where P is the number of subsequences divided; Step A5: Subtract the updated seasonal term from the original data to obtain deseasonalized data Step A6: Perform weighted low-pass filtering on the deseasonalized data to estimate the updated trend term where t is the index of the time point and m is half the length of the sliding window; Step A7: Repeat steps A1 to A6 until the convergence condition is met: and Among them, represents the new trend term and new seasonal term after one iteration, represents the trend term and seasonal term calculated during the previous iteration. After convergence, and are used as the trend term and seasonal term, and the residual term after iteration is obtained 8. A method for predicting the power generation of tidal current based on STL decomposition and multi-model fusion according to claim 1, characterized in that Input the trend term, seasonal term, historical operation data, and the selected historical meteorological factors into the pre-trained TimesNet model for fast Fourier transform and multi-scale modeling prediction to obtain the predicted values of the trend term and seasonal term, specifically including: Take the trend term, seasonal term, historical operation data, and the selected historical meteorological factors as the original time series and input them into the pre-trained TimesNet model; In the TimesNet model, use the fast Fourier transform (FFT) to perform frequency domain conversion on the input data, obtain the amplitude spectra of each frequency component, and identify the dominant periods and frequency characteristics with significant energy contributions; According to the identified dominant period parameters, perform period folding on the original time series, convert the input data from a one-dimensional time series to a two-dimensional tensor, where the row dimension represents the period length, the column dimension represents the number of periods, and the periodicity and period evolution characteristics in the original time series; In the multi-scale modeling process, through the multi-level neural network structure in the TimesNet model, use convolutional kernels of different scales to extract features from this two-dimensional tensor, perform deep modeling at different levels from short-term fluctuations to long-term trends, and finally output the predicted values of the trend term and seasonal term.

9. A method for predicting tidal power generation based on STL decomposition and multi-model fusion according to claim 1, characterized in that Input the residual term, historical operation data, and the selected historical meteorological factors into the pre-trained Itransformer model for sequence modeling and feature extraction to obtain the predicted value of the residual term, specifically including: Perform time modeling on the residual term, historical operation data, and the selected historical meteorological factors to obtain a time series, where each time step contains multiple features corresponding to the residual term, historical operation data, and historical meteorological factors respectively; Input the time series into the pre-trained Itransformer model. The Itransformer model processes the input features through a linear projection layer to generate the query matrix Q, key matrix K, and value matrix V respectively. The query matrix Q is generated from the residual term data, and the key matrix K and value matrix V are generated from the historical operation data and historical meteorological factors; Using the query matrix Q, key matrix K, and value matrix V, the Itransformer model adopts the reverse attention mechanism to calculate the similarity scores in the time step dimension; Based on the calculated similarity scores, the Itransformer model calculates the attention weights through the softmax operation; Multiply the attention weights by the value matrix V to generate a weighted feature representation; The Itransformer model processes the weighted feature representation through multi-layer self-attention calculation and feature aggregation to generate a preliminary predicted value of the residual term; adopt a gated residual connection to perform a non-linear transformation on the preliminary predicted value of the residual term and the input residual term data, combine layer normalization to normalize the hidden states of different layers, and map the processed residual features through the output layer to obtain the final predicted value of the residual term.

10. A method for predicting the power generation of tidal current based on STL decomposition and multi-model fusion according to claim 1, characterized in that Combine the physical calculation formula of tidal energy, add the predicted values of the trend term, seasonal term, and residual term, and perform physical constraints to obtain the final predicted value of tidal current power generation, specifically including: Add the predicted trend value, the predicted seasonal value, and the predicted residual value to obtain the predicted value of the tidal power generation: P pred = P trend + P seasonal + P residual Among them, P trend represents the predicted value of the trend term, P seasonal represents the predicted value of the seasonal term, P residual the predicted value of the residual term, P pred is the predicted value of the tidal power generation; Calculate the maximum value of the real-time theoretical power according to the physical calculation formula of tidal energy: where ρ represents the seawater density, A represents the swept area of the tidal current turbine rotor, V represents the tidal current velocity, η represents the power generation efficiency of the generator, and P max is the maximum value of the real-time theoretical power; Apply physical compliance constraints to the predicted value of the tidal power generation after superposition: If P pred >P max , then the predicted value is forced to be corrected to P max , that is: P pred =P max ; If the real-time flow velocity V is less than the turbine startup threshold, the output power is set to zero, i.e.: P pred = 0; The corrected P after physical constraints pred is used as the final predicted value of the tidal power generation.

Images