Energy storage type hydraulic wind turbine super short-term output power prediction method

Abstract

The application provides an energy storage type hydraulic wind turbine super-short-term output power prediction method, relates to the field of intelligent prediction of machine learning, and comprises the following steps: S1, collecting wind speed historical data; S2, obtaining a wind speed prediction value by adopting a CEEMDAN-LSTM combined model; S3, constructing an optimized WPC model; S4, obtaining a wind turbine output power prediction value; S5, constructing a hydraulic system efficiency physical model to obtain a hydraulic system transmission efficiency theoretical value; S6, constructing a hydraulic system transmission efficiency model to obtain a hydraulic system transmission efficiency prediction value; S7, obtaining a mechanical power prediction value of a hydraulic motor output shaft; S8, constructing a hydraulic energy storage prediction model; and S9, obtaining a final output power prediction value according to a PINN-LSTM. The application realizes high-precision, strong generalization and interpretable super-short-term output power prediction by introducing physical consistency constraints and a state feature sequence learning mechanism of machine learning.

Description

Technical Field

[0001] This invention relates to the field of intelligent prediction using machine learning, specifically a method for predicting the ultra-short-term output power of a hydraulically powered wind turbine generator set. Background Technology

[0002] Hydraulic wind turbine generators with energy storage are a new type of wind energy conversion system. They utilize a wind turbine to drive a hydraulic pump, generating high-pressure oil flow, which is then transmitted via hydraulic pipelines to a ground-based hydraulic motor connected in series with an energy storage unit. This drives the generator to achieve energy output or regulation. This structure eliminates the need for traditional mechanical gearboxes, offering advantages such as high transmission flexibility, flexible layout, and strong environmental adaptability. It is particularly suitable for applications with severe wind conditions or difficult maintenance, such as offshore and high-altitude environments. However, because the system's energy chain involves multi-stage transmission of wind energy—hydraulic energy—mechanical energy—electrical energy—energy storage, and each stage exhibits nonlinear characteristics, such as pitch and displacement systems and the charging and discharging behavior of the accumulator, its output power response is strongly influenced by multiple coupled factors.

[0003] In existing technologies, power prediction methods for hydraulic systems mainly fall into two categories: physical modeling methods and data-driven methods. Physical modeling methods establish mathematical models based on the energy conversion principles of pumps, motors, and pipelines, possessing good interpretability. However, these methods have significant limitations in actual operation, primarily manifested in: the models are highly sensitive to changes in system parameters, struggle to handle highly nonlinear and time-varying operating conditions, and fail to cover unmodeled influencing factors such as temperature changes and dynamic displacement adjustments. Data-driven methods utilize machine learning algorithms such as neural networks and support vector machines to learn and model from historical operating data. While machine learning has a strong ability to fit complex nonlinear relationships, it heavily relies on a large amount of labeled data and is prone to failure when samples are insufficient or the operating environment changes. Furthermore, its black-box nature limits the widespread application of these models in control systems.

[0004] Single modeling methods are no longer sufficient to meet the high accuracy, high stability, and interpretability requirements of power prediction for hydraulic wind power generation systems under varying operating conditions. Therefore, there is an urgent need for a high-precision, highly generalizable prediction method that integrates the advantages of physical knowledge and data learning, capable of wind speed prediction, hydraulic efficiency estimation, and system-level output power correction within a short timescale, supporting the operation optimization and energy dispatch of wind-storage systems. Summary of the Invention

[0005] This invention addresses the aforementioned problems by proposing an ultra-short-term output power prediction method for hydraulically powered wind turbine generators. It integrates wind speed prediction, hydraulic system efficiency modeling, and a PINN-LSTM joint correction mechanism, belonging to the interdisciplinary field of renewable energy system modeling, control, and energy storage scheduling. Specifically, this method is based on the physical processes of the wind turbine-hydraulic transmission-energy storage-power generation energy chain in the system structure. It utilizes machine learning and a collaborative modeling strategy to output the final predicted output power value, comprehensively improving prediction accuracy, robustness, and system interpretability. This invention effectively addresses the power fluctuations and modeling challenges of hydraulically powered wind turbine generator systems under complex and variable operating conditions, enhancing intelligent operation and maintenance and energy storage scheduling capabilities.

[0006] A method for predicting the ultra-short-term output power of an energy storage hydraulic wind turbine generator set, the specific steps of which are as follows:

[0007] S1: Collect historical wind speed data;

[0008] S2: The CEEMDAN-LSTM combined model is used to predict wind speed and obtain the predicted wind speed value;

[0009] S3: Construct an optimized WPC model;

[0010] S4: Using the optimized WPC model, the predicted output power of the wind turbine is obtained based on the predicted wind speed. ;

[0011] S5: Construct a physical model of hydraulic system efficiency to obtain the theoretical value of hydraulic system transmission efficiency. , used as physical constraints for S6;

[0012] S6: Construct a hydraulic system transmission efficiency model based on PINN to obtain the predicted value of the hydraulic system transmission efficiency. The joint loss function of the hydraulic system transmission efficiency model based on PINN includes a data loss term and a physical residual term.

[0013] S7: According to and Obtain the predicted mechanical power of the hydraulic motor output shaft. ;

[0014] S8: Construct an LSTM-based hydraulic energy storage prediction model to obtain the predicted energy storage power. and predicted power output value ;

[0015] S9: Obtain the final predicted output power value based on the PINN-LSTM joint neural network. ;

[0016] A PINN-based motor conversion efficiency model was constructed to obtain the corrected motor conversion efficiency. The PINN-based motor conversion efficiency model and the S8 LSTM-based hydraulic energy storage prediction model together constitute a PINN-LSTM joint neural network. A training loss function for the PINN-LSTM joint neural network is set. The final output power prediction value is obtained based on the final output power prediction expression. .

[0017] Preferably, S2 uses the CEEMDAN-LSTM combined model to predict wind speed, obtaining the predicted wind speed value, specifically:

[0018] S21: Use CEEMDAN to extract the historical wind speed sequence from S1 Decomposed into multiple intrinsic mode functions and residual terms ;

[0019] S22: For each IMF and Separate LSTM networks are used for prediction, and the prediction results of each LSTM network are combined to obtain the wind speed prediction value. .

[0020] Preferably, S3 constructs an optimized WPC model, specifically as follows:

[0021] S31: Constructing the WPC model using a logistic distribution model:

[0022] (3)

[0023] Wherein, parameter vector Let be the parameter vector of the WPC curve. Indicates wind speed. Indicates the power amplitude of the wind turbine. Indicates the shape adjustment factor. This represents the curve slope control factor, where e is the natural constant;

[0024] S32: Parameter optimization;

[0025] Use the Jaya algorithm to process the parameter vector Perform optimization, objective function To minimize predicted power and historical measured power The root mean square error between them; using the optimal parameter vector The WPC model is called the optimized WPC model.

[0026] Preferably, S5 constructs a physical model of the hydraulic system efficiency to obtain the theoretical value of the hydraulic system transmission efficiency. Used as a physical constraint for S6; specifically:

[0027] S51: Establish a physical model for the hydraulic system efficiency of the hydraulic transmission system of an energy storage hydraulic wind turbine generator unit:

[0028] (6)

[0029] in, This represents the theoretical value of the transmission efficiency of the hydraulic system. For hydraulic pump efficiency, For hydraulic motor efficiency, For hydraulic pipeline efficiency.

[0030] Preferably, in step S6, a hydraulic system transmission efficiency model based on PINN is constructed to obtain the predicted value of the hydraulic system transmission efficiency. Specifically:

[0031] S61: Construct a hydraulic system transmission efficiency model based on PINN and introduce physical constraints;

[0032] A transmission efficiency model for a hydraulic system based on PINN is constructed, and its structure is as follows:

[0033] (10)

[0034] in, The predicted output power of the wind turbine obtained by S4. This refers to the pressure difference between the high-pressure and low-pressure pipelines. This refers to the volumetric flow rate of the pipeline. Hydraulic oil temperature The speed of the hydraulic motor. The speed of the hydraulic pump. This is a predicted value for the transmission efficiency of the hydraulic system.

[0035] Construct the joint loss function that includes the physical residuals as follows: :

[0036] (11)

[0037] in, For the joint loss function, For data loss items, The weighting factor for the physical constraint term. For physical residuals;

[0038] Data loss items Used to monitor prediction accuracy:

[0039] (12)

[0040] in, This represents the predicted transmission efficiency value of the hydraulic system for the i-th sample using the PINN-based hydraulic system transmission efficiency model. The actual hydraulic system transmission efficiency for the i-th sample is calculated by measuring the actual input and output power of the hydraulic system using sensors, where N is the number of samples.

[0041] Physical residuals Used to monitor physical consistency:

[0042] (13)

[0043] in, This represents the theoretical value of the hydraulic system transmission efficiency corresponding to the i-th sample, calculated using the hydraulic system efficiency physical model.

[0044] Preferably, S62 specifically includes:

[0045] S62: Parameters updated online;

[0046] An extended Kalman filter is introduced to update key time-varying parameters in real time. The update formula is as follows:

[0047] (14)

[0048] in, Here are the parameter estimates at time t. The estimated values ​​of the prior parameters are derived from the previous time step. The Kalman gain is used to balance the weights between predicted and measured values, determining the update magnitude. This represents the actual measured output at time t. For prediction parameters Through system observation model function The obtained predicted observations, To observe residuals;

[0049] Specifically, the leakage coefficient of the hydraulic pump. Leakage coefficient of hydraulic motor Frictional resistance loss of hydraulic pump Or the frictional resistance loss of the hydraulic motor The updated parameters are used to correct the physical residuals of the PINN-based hydraulic system transmission efficiency model.

[0050] Preferably, S8 constructs an LSTM-based hydraulic energy storage prediction model to obtain the predicted energy storage power. and predicted power output value Specifically:

[0051] The input to an LSTM-based hydraulic energy storage prediction model includes at least the predicted mechanical power of the hydraulic motor output shaft. Current wind speed Hydraulic pressure Hydraulic motor speed accumulator pressure Current pressure of accumulator and energy storage system energy .

[0052] Preferably, S9 obtains the final output power prediction value based on the PINN-LSTM joint neural network model. Specifically:

[0053] S91: Construct a motor conversion efficiency model based on PINN;

[0054] A PINN-based motor conversion efficiency model is constructed, taking the hydraulic system operating state as input and outputting the corrected motor conversion efficiency. :

[0055] (20)

[0056] in, The current pressure of the energy storage device. The shaft speed is For energy storage HPM emissions, P is the total power of the system. motor This is the predicted mechanical power value of the hydraulic motor output shaft. Energy storage capacity;

[0057] S92: Design of the joint loss function for PINN-LSTM;

[0058] The PINN-based motor conversion efficiency model and the LSTM-based hydraulic energy storage prediction model together constitute the PINN-LSTM joint neural network. A training loss function is designed for the PINN-LSTM joint neural network.

[0059] S93: Obtain the final predicted output power value;

[0060] The final output power prediction expression obtained using the PINN-LSTM joint neural network is:

[0061] (25)

[0062] in, This is the predicted final output power value. This is the predicted mechanical power value of the hydraulic motor output shaft. The stored energy is the predicted power output of the LSTM in the PINN-LSTM joint neural network. The corrected motor conversion efficiency is the output of PINN in the PINN-LSTM joint neural network.

[0063] Preferably, in S92, a training loss function is designed for the PINN-LSTM joint neural network; specifically:

[0064] The training loss function consists of three parts:

[0065] ;(twenty one)

[0066] Specifically, the data fitting term:

[0067] ;(twenty two)

[0068] Power generation conservation error term:

[0069] ;(twenty three)

[0070] Energy storage power consistency error term:

[0071] ;(twenty four)

[0072] in, , This is the adjustment coefficient for the constraint loss term. This represents the actual output power of the system.

[0073] The features and beneficial effects of this invention are:

[0074] 1. Integrate wind speed prediction and wind power curve modeling to achieve system feedforward drive;

[0075] This invention employs the CEEMDAN-LSTM method to perform fine decomposition and prediction of short-term wind speed, and combines it with the Jaya algorithm to construct an optimized wind power curve (WPC), which effectively improves the prediction accuracy of input power at the wind turbine end and provides feedforward support for subsequent power evolution modeling.

[0076] 2. Improve the modeling accuracy of intermediate links by using data-model fusion for hydraulic system efficiency modeling;

[0077] This invention introduces a Physical Information Neural Network (PINN) to integrate the nonlinear energy transfer process and state parameter changes of the hydraulic pump-motor-pipeline system into the neural network training, solving the problems of parameter uncertainty and model rigidity in traditional physical modeling, and achieving high-precision prediction of the mechanical power of the motor output shaft.

[0078] 3. An LSTM-based hydraulic energy storage prediction model is introduced, which incorporates hydraulic energy storage branch modeling to reflect the bidirectional power regulation characteristics of the system;

[0079] This invention incorporates the hydraulic energy storage subsystem (including variable pumps / motors and accumulators) into energy path modeling, and uses an LSTM network to predict the instantaneous power absorbed or released by the energy storage branch, thereby improving the ability to predict the system's regulation behavior under complex fluctuation scenarios.

[0080] 4. Construct a PINN-LSTM joint neural network, consisting of a PINN-based motor conversion efficiency model and an LSTM-based hydraulic energy storage prediction model, to strengthen system-level physical consistency constraints.

[0081] This invention is based on the PINN-LSTM joint neural network, which integrates the prediction capability of LSTM with the physical conservation constraints of PINN to construct a unified prediction-correction closed loop. By designing a joint loss function that includes data error terms, power balance terms and energy storage conservation terms, the stability and interpretability of the system output power prediction are effectively improved.

[0082] 5. Possesses good practicality and expandability;

[0083] The method proposed in this invention is applicable to various types of hydraulic wind power generation systems, especially suitable for off-grid / microgrid application scenarios with energy storage functions; the method has a clear structure and can be integrated with simulation platforms such as AMESim and Simulink or actual measurement and control systems to realize system operation status perception and real-time predictive scheduling. Attached Figure Description

[0084] Figure 1 This is a schematic diagram of the energy storage hydraulic wind turbine generator system of the present invention;

[0085] Figure 2 This is a flowchart of the ultra-short-term output power prediction method for energy storage hydraulic wind turbine generator sets according to the present invention;

[0086] Figure 3 This is the result of CEEMDAN decomposition of wind speed in the CEEMDAN-LSTM combined model of Example 1 of this invention;

[0087] Figure 4 This is the result of LSTM predicting wind speed in the CEEMDAN-LSTM combined model of Example 1 of this invention;

[0088] Figure 5 This is the wind power curve (WPC) optimized using the Java algorithm in Example 1 of this invention.

[0089] Figure 6These are the predicted results of wind turbine power, variable motor output power, energy storage branch power, and final power generation in Example 1 of this invention.

[0090] Key reference numerals:

[0091] 1. Wind turbine; 2. Pitch module; 3. Fixed displacement pump; 4. High-pressure pipeline; 5. Low-pressure pipeline; 6. Variable displacement motor; 7. Energy storage system; 71. Accumulator; 72. Pump motor; 73. Energy storage tank; 8. Generator; 9. Power management system; 10. Operation control system. Detailed Implementation

[0092] To more clearly illustrate the technical solution of the present invention, Example 1 and the appendix are now described in conjunction with the present invention. Figure 1 To be continued Figure 6 The present invention will be described in detail below.

[0093] like Figure 1 As shown, the hydraulic wind turbine of this invention mainly includes a wind turbine 1, a pitch module 2, a fixed displacement pump 3, a high-pressure pipeline 4, a low-pressure pipeline 5, a variable displacement motor 6, an energy storage system 7, a generator 8, a power management system 9, and an operation control system 10. The energy storage system 7 includes an accumulator 71, a pump motor 72, and an energy storage tank 73. The wind turbine 1 captures wind energy and converts mechanical energy into hydraulic energy through the fixed displacement pump 3. The hydraulic energy is then transmitted to the variable displacement motor 6 through pipelines, and after speed regulation, it drives the generator 8 to generate electricity. To improve the stability and responsiveness of the output power, the system is designed with an energy storage branch 7 and supplemented by the operation control system 10 for predictive scheduling.

[0094] like Figure 2 As shown, this invention proposes a method for predicting the ultra-short-term output power of hydraulic wind turbines based on digital-analog fusion, comprising the following steps:

[0095] S1: Collect historical wind speed data.

[0096] By using wind speed sensors to collect wind speed data in front of the wind turbine in real time, a continuous historical wind speed sequence with high time resolution is formed. .

[0097] S2: The CEEMDAN-LSTM combined model is used to predict wind speed.

[0098] Considering the non-stationary nature of short-term wind speeds, a CEEMDAN-LSTM combined model is adopted. First, CEEMDAN (Complete Ensemble Empirical Mode Decomposition with Adaptive Noise) is used to analyze the historical wind speed series. The system decomposes the wind speed components to extract features at different scales. These components are then fed into a trained LSTM (Long Short-Term Memory) model for prediction, and finally, the predicted wind speed value is reconstructed. Specifically:

[0099] S21: Use CEEMDAN to extract the historical wind speed sequence from S1 Decomposed into multiple intrinsic mode functions (IMFs) and residual terms. :

[0100] (1)

[0101] Where K represents the number of IMFs derived from the decomposition. This represents the i-th IMF.

[0102] S22: For each IMF and Separate LSTM networks are used for prediction, and the prediction results of each LSTM network are combined to obtain the wind speed prediction value. :

[0103] (2)

[0104] in, For the change over time to be Wind speed forecast at that time Let the change of the i-th IMF component over time be... The predicted value at that time The change of the residual term over time is The predicted value at that time This refers to the time-varying quantities. By sorting the wind speed predictions of multiple time-varying quantities in chronological order, we obtain the total wind speed prediction. .

[0105] S3: Construct an optimized WPC (wind power curve) model.

[0106] To address the nonlinear relationship between wind speed and wind turbine output power, the parameters of the WPC model are adjusted based on sample data using the Jaya optimization algorithm, resulting in better generalization performance.

[0107] S31: Constructing the WPC model using a logistic distribution model:

[0108] (3)

[0109] Wherein, parameter vector Let be the parameter vector of the WPC curve. Indicates wind speed. Indicates the power amplitude of the wind turbine, Indicates shape adjustment factor, This represents the curve slope control factor, where e is the natural constant.

[0110] S32: Parameter optimization.

[0111] Use the Jaya algorithm to process the parameter vector Perform optimization, objective function To minimize predicted power and historical measured power The root mean square error (RMSE) between them:

[0112] (4)

[0113] Where N is the number of elements in the predicted power sequence obtained from the WPC model, which can be set as needed. The optimal parameter vector is obtained after optimization using the Jaya algorithm. Using the optimal parameter vector The WPC model is called the optimized WPC model.

[0114] S4: Using the optimized WPC model, the predicted output power of the wind turbine is obtained based on the predicted wind speed. .

[0115] The wind speed prediction value obtained in S2 Substituting into the optimized WPC model, i.e., in formula (3) The model becomes The model yields predicted output power values ​​for the wind turbine. :

[0116] (5)

[0117] in, The change over time is represented as Predicted output power of wind turbine at that time The change over time is represented as Hourly wind speed forecast This represents the optimal parameter vector. By sorting the predicted wind turbine output power values ​​for multiple time-varying parameters in chronological order, we obtain the predicted wind turbine output power values. .

[0118] S5: Construct a physical model of hydraulic system efficiency.

[0119] Based on the hydraulic system structure and energy transfer path, a hydraulic efficiency physical model is established to clarify energy loss and transmission characteristics, providing physical constraints for subsequent network training. The hydraulic transmission system of the energy storage hydraulic wind turbine generator unit consists of a hydraulic pump, hydraulic pipelines, and a hydraulic motor. In this embodiment, the hydraulic pump is a fixed displacement pump 3, the hydraulic pipelines consist of high-pressure pipelines 4 and low-pressure pipelines 5, and the hydraulic motor is a variable displacement motor 6.

[0120] Establish a physical model for the hydraulic system efficiency of the hydraulic transmission system of the energy storage hydraulic wind turbine generator unit:

[0121] (6)

[0122] in, This represents the theoretical value of the transmission efficiency of the hydraulic system. For hydraulic pump efficiency, For hydraulic motor efficiency, For hydraulic pipeline efficiency.

[0123] Hydraulic pump efficiency :

[0124] (7)

[0125] Among them, C s，p The leakage coefficient of the hydraulic pump. This refers to the pressure difference between the inlet and outlet of the hydraulic pump. This refers to the displacement of the hydraulic pump. This is the frictional resistance loss of the hydraulic pump. Typically related to the speed of the hydraulic pump and hydraulic oil temperature Related.

[0126] Hydraulic motor efficiency :

[0127] (8)

[0128] Among them, C s，m The leakage coefficient of the hydraulic motor. This refers to the pressure difference between the inlet and outlet of the hydraulic motor. The rotational speed of the hydraulic motor. Hydraulic oil temperature This refers to the displacement of the hydraulic motor. This refers to the frictional resistance loss of the hydraulic motor. Typically related to the speed of the hydraulic motor and hydraulic oil temperature Related.

[0129] Hydraulic pipeline efficiency :

[0130] (9)

[0131] in, For the coefficient of friction, For pipe length, For pipeline volumetric flow rate, For the cross-sectional area of ​​the pipe, For pipe diameter, For gravitational acceleration, This refers to the pressure difference between high-pressure and low-pressure pipelines.

[0132] In this embodiment, the hydraulic pipeline consists of a high-pressure pipeline 4 and a low-pressure pipeline 5. Therefore, the efficiency of the high-pressure pipeline and the efficiency of the low-pressure pipeline are obtained by using formula (9) for the high-pressure pipeline 4 and the low-pressure pipeline 5 respectively. Then, the average value of the two is taken as the hydraulic pipeline efficiency in this embodiment. .

[0133] These will later form the residual templates used for physical supervision in the PINN-based hydraulic system transmission efficiency model, serving as physical constraints that the PINN-based hydraulic system transmission efficiency model must satisfy as much as possible.

[0134] S6: Construct a hydraulic system transmission efficiency model based on PINN.

[0135] The hydraulic system efficiency physical model built in S5 is embedded into the PINN (Physics-Informed Neural Network) framework. Its core lies in the deep integration of data-driven methods with physical mechanisms, and the introduction of an online parameter update strategy to adapt to the time-varying characteristics of the system, thereby achieving stable and reliable modeling under different operating conditions. This method aims to improve the accuracy and physical consistency of the nonlinear transmission efficiency model of the hydraulic system, and to train the neural network using collected data, thereby accurately predicting the actual output power at the variable displacement motor end under the constraints of physical laws.

[0136] S61: Construct a hydraulic system transmission efficiency model based on PINN and introduce physical constraints.

[0137] The hydraulic system transmission efficiency model is obtained using a nonlinear modeling approach. To improve modeling accuracy and ensure physical consistency, a Physical Information Neural Network (PINN) is constructed as the hydraulic system transmission efficiency model. Taking the hydraulic system's operating state as input, it outputs the hydraulic system transmission efficiency. Its structure is as follows:

[0138] (10)

[0139] in, The predicted output power of the wind turbine obtained by S4. This refers to the pressure difference between the high-pressure and low-pressure pipelines. This refers to the volumetric flow rate of the pipeline. Hydraulic oil temperature The speed of the hydraulic motor. The speed of the hydraulic pump. This is a predicted value for the transmission efficiency of the hydraulic system.

[0140] Each ( , , , , , A sample is referred to as the input to the PINN-based hydraulic system transmission efficiency model.

[0141] Construct the joint loss function that includes the physical residuals as follows: :

[0142] (11)

[0143] in, For the joint loss function, For data loss items, The weighting factor for the physical constraint term. This is the physical residual term.

[0144] Data loss items Used to monitor prediction accuracy:

[0145] (12)

[0146] in, This represents the predicted transmission efficiency value of the hydraulic system for the i-th sample using the PINN-based hydraulic system transmission efficiency model. The actual hydraulic system transmission efficiency for the i-th sample is calculated by measuring the actual input and output power of the hydraulic system using sensors, where N is the number of samples.

[0147] Physical residuals Used to monitor physical consistency:

[0148] (13)

[0149] in, The theoretical value of the hydraulic system transmission efficiency corresponding to the i-th sample is calculated by the physical model of hydraulic system efficiency and is obtained according to formula (6).

[0150] The predicted transmission efficiency of the hydraulic system is obtained using the PINN-based hydraulic system transmission efficiency model as described above. It not only fits historical real efficiency data, but also tries to get as close as possible to the total system efficiency obtained from the physical model of hydraulic system efficiency. The PINN-based hydraulic system transmission efficiency model maintains high modeling accuracy and physical consistency under different ambient temperatures, load disturbances, and system nonlinearities.

[0151] S62: Parameters updated online.

[0152] This step is optional and is used to enhance the system's adaptability to parameter drift or operating condition changes. An extended Kalman filter (EKF) is introduced to update key time-varying parameters in real time. The update formula is as follows:

[0153] (14)

[0154] in, Here are the parameter estimates at time t. The estimated values ​​of the prior parameters are derived from the previous time step. The Kalman gain is used to balance the weights between predicted and measured values, determining the update magnitude. This represents the actual measured output at time t. For prediction parameters

[0155] Through system observation model function The obtained predicted observations, To observe the residuals.

[0156] In this embodiment, it can be specifically the leakage coefficient of the hydraulic pump. Leakage coefficient of hydraulic motor Frictional resistance loss of hydraulic pump Or the frictional resistance loss of the hydraulic motor .

[0157] The updated parameters are used to correct the physical residuals of the PINN-based hydraulic system transmission efficiency model, resulting in a dynamically updated version of the hydraulic system transmission efficiency prediction model.

[0158] (15)

[0159] The modules used to implement this step can be enabled or disabled as needed to adapt to the balance between accuracy and computational complexity requirements in different deployment environments.

[0160] S7: Obtain the predicted mechanical power of the hydraulic motor output shaft. .

[0161] The predicted output power of the wind turbine obtained from S4 The predicted transmission efficiency of the hydraulic system obtained from S6 Multiplying these values ​​yields the predicted mechanical power P of the hydraulic motor output shaft. motor :

[0162] (16)

[0163] Predicted mechanical power P of the hydraulic motor output shaft motor This will be used for the final effective power prediction and system-level energy balance correction.

[0164] S8: Construct a hydraulic energy storage prediction model based on LSTM.

[0165] LSTM is suitable for modeling nonlinear time-varying dynamic processes and can learn the evolution trend of input variables in time series. Key operating parameters extracted from energy storage system 7 are used as inputs to the LSTM-based hydraulic energy storage prediction model.

[0166] The input to an LSTM-based hydraulic energy storage prediction model includes at least the predicted mechanical power of the hydraulic motor output shaft. Current wind speed hydraulic pressure Hydraulic motor speed accumulator pressure Current pressure of the energy device Energy storage system energy .

[0167] In this embodiment, the input feature sequence of the hydraulic energy storage prediction model is as follows:

[0168] (17)

[0169] in, This is the predicted mechanical power value of the hydraulic motor output shaft. Given the current wind speed, ambient temperature, and hydraulic pressure, The speed of the hydraulic motor. For accumulator pressure, The propeller pitch angle of the wind turbine. The swing angle of the energy storage HPM (control variable). The current pressure of the energy storage device. Energy for energy storage systems.

[0170] The LSTM-based hydraulic energy storage prediction model uses a sliding window time series. As input, the output consists of two predicted quantities: one is the predicted energy storage power. One is the predicted power output value. .

[0171] S9: The final output power prediction value is obtained based on the PINN-LSTM joint neural network.

[0172] This paper integrates the prediction results of a data-driven LSTM-based hydraulic energy storage prediction model with strict physical laws. By introducing a PINN-based motor conversion efficiency model and power conservation constraints, a physically consistent final output power prediction model is constructed. The core of this approach is to address the potential physical inconsistencies that may arise from purely data-driven models, ensuring that the prediction results conform to historical data patterns while strictly adhering to the fundamental physical law of energy conservation. This significantly improves the reliability and generalization ability of the prediction results.

[0173] S91: Construct a motor conversion efficiency model based on PINN.

[0174] To ensure that the predictions of the LSTM-based hydraulic energy storage prediction model conform to the energy conservation law of the system, a PINN-based motor conversion efficiency model is introduced for correction.

[0175] In existing technologies, the total power of the system satisfy:

[0176] (18)

[0177] In the formula, P motor This is the predicted mechanical power value of the hydraulic motor output shaft. For energy storage power, This is the motor conversion efficiency, and this value is usually obtained based on empirical values ​​or fitting functions.

[0178] Among them, energy storage power The prediction model satisfies:

[0179] (19)

[0180] In the formula, For energy storage system flow, For energy storage HPM emissions, The current pressure of the energy storage device. The shaft speed is It means releasing energy to generate electricity. This indicates that the liquid absorbs energy.

[0181] Based on this, a PINN-based motor conversion efficiency model is constructed, taking the hydraulic system operating state as input and outputting the corrected motor conversion efficiency:

[0182] (20)

[0183] S92: Design of the joint loss function for PINN-LSTM.

[0184] The PINN-based motor conversion efficiency model and the LSTM-based hydraulic energy storage prediction model together constitute the PINN-LSTM joint neural network. A training loss function is designed for the PINN-LSTM joint neural network.

[0185] The training loss function consists of three parts:

[0186] .(twenty one)

[0187] Specifically, the data fitting term:

[0188] .(twenty two)

[0189] Power generation conservation error term:

[0190] .(twenty three)

[0191] Energy storage power consistency error term:

[0192] ;(twenty four)

[0193] in, , This is the adjustment coefficient for the constraint loss term. This represents the actual output power of the system.

[0194] S93: Obtain the final predicted output power value.

[0195] The LSTM output in the PINN-LSTM combined neural network As for formula (18) PINN output As for formula (18) The final output power prediction expression obtained using the PINN-LSTM joint neural network is:

[0196] (25)

[0197] in, This is the predicted final output power value. The predicted mechanical power P of the hydraulic motor output shaft motor , The predicted energy storage power of the LSTM output in the PINN-LSTM joint neural network. , This represents the corrected motor conversion efficiency of the PINN output in the PINN-LSTM joint neural network. This is because the input to the PINN-LSTM joint neural network contains... Therefore, formula (25) is actually obtained by iterative method to obtain the final output power prediction value.

[0198] Example 1

[0199] In Example 1, a set of wind speed data under turbulent conditions with an average wind speed of 12 m / s for 600 s was selected as historical wind speed data. A 5 kW hydraulic fan was selected as the prediction target for subsequent predictions.

[0200] Subsequently, after executing S2 on the selected historical wind speed data, the following results were obtained: Figure 3 The image shows the decomposition results of a typical wind speed signal after processing with CEEMDAN. Figure 4 The output of the LSTM network for predicting wind speed in future time periods is shown, from top to bottom: high-frequency wind speed prediction, mid-frequency wind speed prediction, low-frequency wind speed prediction, and a combined prediction of actual wind speed and a portion of the test set. It can be seen that the wind speed prediction performance is good, with the mean square error of the low-frequency component prediction being less than 10. -6 The mean square error of the mid-frequency and low-frequency component predictions is 0.0008, and the mean square error of the mid-frequency and low-frequency component predictions is 0.00096.

[0201] Based on the nonlinear relationship between wind speed and wind turbine output power, the Jaya algorithm is used to model the wind power curve and the parameters are optimized according to the objective function. This invention uses a logistic distribution model to construct the wind power curve. That is, formula (3), therefore there are four parameter vectors in total. In this embodiment, the optimal parameter vector is finally obtained. With these parameters, the normalized root mean square error (NRMSE) of the samples is 0.0263, indicating a good fit. Figure 5 The fitting results are shown.

[0202] The predicted wind speed is input into the WPC model with optimal parameters to obtain the predicted output power of the wind turbine. Then, combining the characteristics of the system's pumps, motors, and pipelines, a physical model of the hydraulic system is established. This model, along with a PINN-based hydraulic system transmission efficiency model, yields the predicted mechanical power of the hydraulic motor output shaft. Finally, an LSTM-based hydraulic energy storage prediction model is used to obtain the short-term power output prediction of the energy storage branch. Finally, the corrected motor conversion efficiency is incorporated. The final predicted output power value is obtained.

[0203] like Figure 6 As shown, from top to bottom, the ultra-short-term prediction results of wind turbine output power, variable motor output power, energy storage branch power, and final power generation are presented. The predicted values ​​and actual values ​​are generally highly consistent, and the prediction error is stably controlled within ±2.5%, demonstrating significant engineering practical value.

[0204] The final output power prediction value obtained by this invention can be used for off-grid system load scheduling, wind-storage power balance control, real-time fault early warning and status diagnosis, or dynamic power regulation and energy storage scheduling of offshore wind farms.

[0205] This invention, based on the concept of data-model fusion, effectively combines physical modeling with data-driven algorithms, significantly improving the accuracy and stability of power prediction for hydraulic wind turbines. The proposed method exhibits good scalability and deployment flexibility, making it suitable for intelligent scheduling and optimized control in scenarios with drastic wind speed fluctuations.

[0206] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for ultra-short-term output power prediction of a pumped hydro wind turbine generator system, characterized in that, It includes the following steps: S1: Collect historical wind speed data; S2: The CEEMDAN-LSTM combined model is used to predict wind speed and obtain the predicted wind speed value; S3: Construct an optimized power component curve model; S31: Constructing the power component curve model using a logical distributed model: ； Wherein, parameter vector Let be the parameter vector of the WPC curve. Indicates wind speed. Indicates the power amplitude of the wind turbine. Indicates the shape adjustment factor. This represents the curve slope control factor, where e is the natural constant; S32: Parameter optimization; Use the Jaya algorithm to process the parameter vector Perform optimization, objective function To minimize predicted power and historical measured power The root mean square error between them; using the optimal parameter vector The power component curve model is called the optimized power component curve model; S4: Using the optimized power component curve model, the predicted output power of the wind turbine is obtained based on the predicted wind speed. ; S5: Construct a physical model of hydraulic system efficiency to obtain the theoretical value of hydraulic system transmission efficiency. , used as physical constraints for S6; S6: Construct a hydraulic system transmission efficiency model based on PINN to obtain the predicted value of the hydraulic system transmission efficiency. The joint loss function of the hydraulic system transmission efficiency model based on PINN includes a data loss term and a physical residual term. S7: According to and Obtain the predicted mechanical power of the hydraulic motor output shaft. ; S8: Construct an LSTM-based hydraulic energy storage prediction model to obtain the predicted energy storage power. and predicted power output value ; S9: Obtain the final predicted output power value based on the PINN-LSTM joint neural network. ; A PINN-based motor conversion efficiency model was constructed to obtain the corrected motor conversion efficiency. The PINN-based motor conversion efficiency model and the S8 LSTM-based hydraulic energy storage prediction model together constitute a PINN-LSTM joint neural network. A training loss function for the PINN-LSTM joint neural network is set. The final output power prediction value is obtained based on the final output power prediction expression. .

2. The method for predicting the ultra-short-term output power of a storage-type hydraulic wind turbine generator set according to claim 1, characterized in that, The S2 uses the CEEMDAN-LSTM combined model to predict wind speed, obtaining the predicted wind speed value as follows: S21: Use CEEMDAN to extract the historical wind speed sequence from S1 Decomposed into multiple intrinsic mode functions and residual terms ; S22: For each IMF and Separate LSTM networks are used for prediction, and the prediction results of each LSTM network are combined to obtain the wind speed prediction value. .

3. The method for predicting the ultra-short-term output power of a storage-type hydraulic wind turbine generator set according to claim 1, characterized in that, In step S5, a physical model of the hydraulic system efficiency is constructed to obtain the theoretical value of the hydraulic system transmission efficiency. Used as a physical constraint for S6; specifically: S51: Establish a physical model for the hydraulic system efficiency of the hydraulic transmission system of an energy storage hydraulic wind turbine generator unit: ； in, This represents the theoretical value of the transmission efficiency of the hydraulic system. For hydraulic pump efficiency, For hydraulic motor efficiency, For hydraulic pipeline efficiency.

4. The method for predicting the ultra-short-term output power of a hydraulically powered wind turbine generator set according to claim 1, characterized in that, In step S6, a hydraulic system transmission efficiency model based on PINN is constructed to obtain the predicted value of the hydraulic system transmission efficiency. Specifically: S61: Construct a hydraulic system transmission efficiency model based on PINN and introduce physical constraints; A transmission efficiency model for a hydraulic system based on PINN is constructed, and its structure is as follows: ； in, The predicted output power of the wind turbine obtained by S4. This refers to the pressure difference between the high-pressure and low-pressure pipelines. This refers to the volumetric flow rate of the pipeline. Hydraulic oil temperature The speed of the hydraulic motor. The speed of the hydraulic pump. This is a predicted value for the transmission efficiency of the hydraulic system. Construct the joint loss function that includes the physical residuals as follows: : ； in, For the joint loss function, For data loss items, The weighting factor for the physical constraint term. For physical residuals; Data loss items Used to monitor prediction accuracy: ； in, This represents the predicted transmission efficiency value of the hydraulic system for the i-th sample using the PINN-based hydraulic system transmission efficiency model. The actual hydraulic system transmission efficiency for the i-th sample is calculated by measuring the actual input and output power of the hydraulic system using sensors, where N is the number of samples. Physical residuals Used to monitor physical consistency: ； in, This represents the theoretical value of the hydraulic system transmission efficiency corresponding to the i-th sample, calculated using the hydraulic system efficiency physical model.

5. The method for predicting the ultra-short-term output power of an energy storage hydraulic wind turbine generator set according to claim 4, characterized in that, It also includes S62, specifically: S62: Parameters updated online; An extended Kalman filter is introduced to update key time-varying parameters in real time. The update formula is as follows: ； in, Here are the parameter estimates at time t. The estimated values ​​of the prior parameters are derived from the previous time step. The Kalman gain is used to balance the weights between predicted and measured values, determining the update magnitude. This represents the actual measured output at time t. For prediction parameters Through system observation model function The obtained predicted observations, To observe residuals; Specifically, the leakage coefficient of the hydraulic pump. Leakage coefficient of hydraulic motor Frictional resistance loss of hydraulic pump Or the frictional resistance loss of the hydraulic motor The updated parameters are used to correct the physical residuals of the PINN-based hydraulic system transmission efficiency model.

6. The method for predicting the ultra-short-term output power of a storage-type hydraulic wind turbine generator set according to claim 1, characterized in that, In step S8, an LSTM-based hydraulic energy storage prediction model is constructed to obtain the predicted energy storage power. and predicted power output value ; Specifically: The input to an LSTM-based hydraulic energy storage prediction model includes at least the predicted mechanical power of the hydraulic motor output shaft. Current wind speed Hydraulic pressure Hydraulic motor speed accumulator pressure Current pressure of accumulator and energy storage system energy .

7. The method for predicting the ultra-short-term output power of a hydraulically powered wind turbine generator set according to claim 1, characterized in that, In step S9, the final output power prediction value is obtained based on the PINN-LSTM joint neural network. ; Specifically: S91: Construct a motor conversion efficiency model based on PINN; A PINN-based motor conversion efficiency model is constructed, taking the hydraulic system operating state as input and outputting the corrected motor conversion efficiency. : ； in, The current pressure of the energy storage device. The shaft speed is For energy storage HPM emissions, P is the total power of the system. motor This is the predicted mechanical power value of the hydraulic motor output shaft. Energy storage capacity; S92: Design of the joint loss function for PINN-LSTM; The PINN-based motor conversion efficiency model and the LSTM-based hydraulic energy storage prediction model together constitute the PINN-LSTM joint neural network. A training loss function is designed for the PINN-LSTM joint neural network. S93: Obtain the final predicted output power value; The final output power prediction expression obtained using the PINN-LSTM joint neural network is: ； in, This is the predicted final output power value. This is the predicted mechanical power value of the hydraulic motor output shaft. The stored energy is the predicted power output of the LSTM in the PINN-LSTM joint neural network. The corrected motor conversion efficiency is the output of PINN in the PINN-LSTM joint neural network.

8. The method for predicting the ultra-short-term output power of a storage-type hydraulic wind turbine generator set according to claim 7, characterized in that, In S92, a training loss function is designed for the PINN-LSTM joint neural network; specifically: The training loss function consists of three parts: ； Specifically, the data fitting term: ； Power generation conservation error term: ； Energy storage power consistency error term: ； in, , This is the adjustment coefficient for the constraint loss term. This represents the actual output power of the system.

Images