Ocean temperature difference energy power generation efficiency prediction method and device

Abstract

The invention relates to an ocean temperature difference energy power generation efficiency prediction method and device, and belongs to the technical field of ocean energy development and prediction. Through the self-learning capability of the PINN, the relation between network input (energy accumulator pressure, motor rotating speed, current flowing through a load resistor and time) and network output (system efficiency) is established, and the prediction of the ocean temperature difference energy power generation efficiency is realized. According to the method, the PINN model is used for embedding physical constraints, so that the physical reasonability can still be ensured under the condition of insufficient data, transfer learning except conditions used in a small training data set and training is remarkably improved, high-precision rapid prediction can be realized, and the reliability is improved.

Description

Technical Field

[0001] The present invention relates to a method and device for predicting the power generation efficiency of ocean thermal energy conversion, belonging to the technical field of ocean energy development and prediction. Background Art

[0002] With the development of the global economy and the continuous growth of the population, the demand for energy has been continuously rising, and energy shortage has become an important bottleneck restricting the sustainable development of countries around the world. At the same time, the problem of climate change has become increasingly prominent. Phenomena such as global warming caused by greenhouse gas emissions and frequent extreme weather events have brought huge challenges to the survival and development of mankind. Reducing carbon emissions and promoting the transformation of the energy structure towards clean and low-carbon has become the general consensus and urgent need of the international community.

[0003] As the largest resource treasure house on the earth, the ocean contains rich renewable energy. As a clean energy with huge reserves and sustainable utilization, the development of ocean thermal energy has important significance and great development potential. In the ocean thermal energy conversion power generation system, predicting the power generation efficiency can guide material selection, heat exchanger structure design and optimization of operating parameters (such as temperature difference, flow rate), maximize the energy conversion rate, and at the same time can adjust the operation strategy in real time (such as working fluid flow rate, pump speed), avoid sudden drops in efficiency or equipment damage caused by temperature difference changes, and timely detect system damage based on abnormal efficiency, and perform timely maintenance to reduce downtime losses. Chinese patent document CN111210080A proposes an energy hub and joint control method for hybrid ocean energy multi-energy complementation, which realizes the coordinated production and efficient scheduling of electricity, heat and cold energy by integrating multiple ocean energies (tidal energy, offshore wind energy, ocean thermal energy, etc.) and energy storage systems. This solution uses traditional OTEC for ocean thermal energy to directly exchange heat relying on the seawater temperature difference, and the efficiency is limited by the low temperature difference. Moreover, it relies on a static coupling matrix and preset constraint conditions, and cannot adapt to changes in the ocean environment in real time (such as temperature fluctuations, efficiency decay caused by biological attachment). Its scheduling strategy is based on historical data and lacks dynamic prediction ability.

[0004] Existing prediction technologies rely on a large amount of experimental data or high-precision grid calculations, which are costly, have poor real-time performance, and are difficult to handle problems such as phase change non-linearity and multi-physical field coupling. As a result, the reliability of the prediction results is insufficient under complex working conditions, and some real experimental data are difficult to collect, which further increases the prediction difficulty. Therefore, it is of great significance to study a new prediction method. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the present invention provides a method and device for predicting the power generation efficiency of ocean thermal energy conversion. By the self-learning ability of the physics-informed neural network (PINN), the relationship between the network input (accumulator pressure, motor speed, current flowing through the load resistor, and time) and the network output (system efficiency) is established to achieve the prediction of the power generation efficiency of ocean thermal energy conversion. The present invention uses the PINN model to ensure physical rationality even in the case of insufficient experimental data by embedding physical constraints, significantly improves transfer learning outside the small training dataset and the conditions used in training, and can achieve high-precision and rapid prediction to improve reliability.

[0006] The technical solution of the present invention is as follows: A method for predicting the power generation efficiency of ocean thermal energy conversion, comprising the following steps: Step 1, build an amesim simulation model of the ocean thermal energy conversion power generation system; Step 2, conduct an accumulator discharge experiment, and collect the pressure, motor speed, current value flowing through both ends of the load resistor, and the system output power value during the accumulator discharge process; Step 3, input the collected parameters into the PINN neural network model and preprocess them; Step 4, split the preprocessed dataset into a training set, a test set, and a validation set according to a ratio; Step 5, build a preliminary PINN model for predicting the power generation efficiency of ocean thermal energy conversion; Step 6, use the data of the split training set as the input of the model for training, and obtain a PINN prediction model for the power generation efficiency of ocean thermal energy conversion through multiple training updates; Step 7, build a test device platform for the ocean thermal energy conversion power generation system; Step 8, collect the pressure, motor speed, current value flowing through the load resistor, and the power generation efficiency value during part of the accumulator discharge process through the test bench experiment, input the collected actual experimental parameters into the trained PINN prediction model for fine-tuning to obtain the final prediction model; Step 9, input the newly collected test parameters into the PINN prediction model based on the power generation efficiency of ocean thermal energy conversion, and finally obtain the corresponding system prediction efficiency.

[0007] Preferably, in step 1, an Amesim simulation platform for ocean thermal energy conversion power generation system is built, including a pressure source, an accumulator, a directional control valve, a hydraulic motor, a rotating load, an electric motor, a rectification module, a load resistor, a sensor acquisition module, a power acquisition module, and an efficiency acquisition module. The pressure source is used to provide power for the system, the directional control valve is used to control the working state of the system, the hydraulic motor is used to convert hydraulic energy into mechanical energy, the hydraulic motor is connected to a rotating load which can be used to receive the torque fed back by the electric motor, the electric motor is used to convert mechanical energy into electrical energy, the rectification module is used to convert three-phase alternating current into direct current, and the load resistor is used to consume the electric energy output by the electric motor; The sensor acquisition module includes a rotational speed sensor, a current sensor, and a pressure sensor, which are used to collect key parameters in real time. During the entire discharge experiment, the energy is transmitted from the accumulator to the hydraulic motor, the electric motor, and the load resistor, experiencing the conversion from hydraulic energy to mechanical energy to electrical energy; The power acquisition module is used to collect the input power and output power of the system; The efficiency acquisition module is used to collect the power generation efficiency y of the system.

[0008] Preferably, step 2 is to conduct an accumulator discharge experiment using the simulation platform built in step 1. Without considering the energy storage process of the electric motor, connect the output end of the electric motor to the load resistor, directly consume the electric energy generated by the electric motor, and connect a current sensor next to the load resistor to collect the current value flowing through the load resistor.

[0009] The method for obtaining the input data set of the training data is as follows: Determine that the volume of the accumulator is a certain value, conduct an accumulator discharge experiment, and respectively collect the pressure P of the accumulator, the rotational speed n of the electric motor, the current I flowing through the load resistor, and the power generation efficiency y of the system through sensors.

[0010] Preferably, step 3 is to synthesize the data sets collected in step 2 into a complete data set. The data set contains five columns of data: time, accumulator pressure, electric motor rotational speed, current flowing through the load resistor, and power generation efficiency. Perform filtering preprocessing on the four columns of data except time, and at the same time perform maximum-minimum normalization processing on the data to map all the data to the interval [-1, 1], improving the training efficiency and stability of the model. For the filtering process, moving average filtering is used to remove high-frequency noise in the data. The formula is as follows: (1) In the formula, x[n] represents the data sequence to be processed, y[n] represents the sequence after moving average filtering, N represents the window size of the moving average, and n represents the index of the data point.

[0011] The formula for normalizing the data is as follows: (2) In the formula: x max represents the maximum value in the data set to be processed, xmin Represents the minimum value in the dataset to be processed. x m Represents the data after normalization.

[0012] Preferably, in step 4, overfitting may occur during training. The essence of overfitting is that the model over-learns the noise and specific patterns of the training data and fails to effectively capture the true patterns behind the data. To solve this problem, the preprocessed dataset is split into a training set, a test set, and a validation set in a ratio of 6:2:2.

[0013] Preferably, in step 5, a preliminary PINN model for predicting the ocean thermal energy conversion power generation efficiency is built. This model uses a fully connected multi-layer neural network as a feature code to embed the input parameter features, and at the same time adds multiple physical constraints to improve the reliability of the PINN model output.

[0014] Among them, the PINN neural network model is built based on the Python language.

[0015] The PINN neural network includes an input layer, a hidden layer, and an output layer. The activation function of the output layer is sigmoid, which compresses the output into the interval [0,1].

[0016] The hidden layer of the neural network has two layers, and each hidden layer has 64 neurons. The hidden layer uses a fully connected layer, and its activation function is tanh (hyperbolic tangent), which compresses the output into the interval [-1,1].

[0017] Preferably, in step 6, the model is trained. The training process is as follows: First, the data of the divided training set is brought into the built PINN neural network model; then the network weights are initialized, the learning rate and optimizer are set; the collected time series data and physical parameters are input, and the data-driven loss and physical constraint loss are calculated respectively, and the weighted sum is used to obtain the total loss L; then automatic differentiation is used for optimization, and the optimized correction value is propagated forward to update the grid weights; finally, after multiple iterative updates, the PINN neural network model for ocean thermal energy conversion power generation is output to complete the training.

[0018] Further preferably, in step 6, two optimizers, adam and lbfgs, are provided for learning during model training; there are two optimizers in the model. During training, adam is used first, and then lbfgs is used.

[0019] Use the Adam optimizer for learning. Based on the gradient descent optimization algorithm, the learning rate is 1×10 3 , the total number of training times is 2000, and the model performance is verified on the test data every 100 rounds.

[0020] Learn using the LBFGS optimizer, adopt a second-order optimization algorithm, set the initial learning rate to 1, and then automatically determine the step size through search without manually setting the learning rate. During each iteration, find a step size along the current search direction. α such that the objective function f ( x + αd ) meets the descent condition. The total number of training times is 2000, and the model performance is verified on the test data every 100 rounds.

[0021] Further preferably, in step 6, the loss function of the PINN neural network is L, and its calculation formula is as follows: L = λ d L data + λ p L phys + λ b L bound (3) In the formula, L data represents the data loss term of the PINN neural network, L phys is the physical loss term, L bound is the efficiency boundary loss term, and λ d 、 λ p 、 λ b are the weights of different terms respectively.

[0022] λ d 、 λ p 、 λ b The weight values are given by the parameter training network. Define the initial weights as 、 、 . Each training will save and update the weight values. The updated weight is . Substitute λ 1 into the PINN neural network to continue participating in the model training.

[0023] Perform multiple loop iterations of training until the PINN neural network model for predicting the ocean thermal energy conversion efficiency converges.

[0024] Further preferably, the data loss term of the PINN neural network L data Adopt the mean square error, and the calculation formula is as follows: (4) In the formula, N is the number of initial samples, t i represents the i sample point corresponding to the y pre th time point, y data represents the predicted value obtained after training by the PINN neural network model, y pre ( t i ) represents the predicted value obtained after training the data of the i th initial sample time point at the initial moment through the PINN neural network model, y data ( t i ) represents the output sample data of the i th initial sample time point at the initial moment.

[0025] Further preferably, the physical loss term of the PINN neural network L phys is the residual of the physical equation, and the calculation formula is as follows: (5) (6) (7) In the formula, M is the number of physical constraint sampling points, t j represents the j sample point corresponding to the f phys th time point, f phys ( t j ) represents the physical equation corresponding to the j th sampling time point, y pre ( t j ) represents the predicted value of the system efficiency corresponding to the j th sampling time point, R represents the resistance value of the load resistor, P represents the accumulator pressure, represents the angular velocity of the motor.

[0026] According to the system energy conservation during the accumulator discharging process and Newton's second law, the following formula is obtained: (8) In the formula, V V₀, P p₀ respectively represent the initial volume and initial pressure of the nitrogen gas in the accumulator, Q ( t j ) represents the hydraulic motor inlet flow rate value corresponding to the j th sampling time point, k represents the adiabatic index of the gas (here referring to nitrogen), E represents the effective volume elastic modulus of the accumulator, t represents the j th sampling time point corresponding to the T, J J represents the moment of inertia, c b represents the damping constant, D q represents the displacement of the hydraulic motor, k e Kₑ represents the back electromotive force constant, I ( t j ) represents the current value flowing through the load resistor corresponding to the j th sampling time point, i represents the j th sampling time point corresponding to the I.

[0027] The relationship between the motor speed n, motor angular velocity ω and flow rate Q in the system is as follows: (9) Q = nD (10) Therefore, substituting equations (7), (8), and (10) into equation (6) for calculation gives: (11) Among them V V₀, P p₀, k, E, J J, c b, D q, k e Kₑ, R are all known and determined values.

[0028] Further preferably, the boundary condition for restricting the efficiency is 0 < y pre η < y max, so the efficiency boundary loss term of the PINN neural network L bound The calculation formula is as follows: (12) In the formula, ReLU is the activation function, which is used to punish values outside the allowable range. ReLU ( y pre ( t i ) - y max ) represents that when the predicted efficiency of the PINN neural network y pre ( t i ) exceeds the efficiency upper limit y max , the value of this term is positive, otherwise it is 0. ReLU (- y pre ( t i )) represents that when the predicted value of the PINN neural network y pre ( t i ) is negative, the value of this term is positive, otherwise it is 0.

[0029] Further preferably, in step 6, when training the neural network, the initial conditions are fitted through supervised learning, while the power and efficiency constraints are incorporated into the loss function in an unsupervised form; the optimization process is optimized using automatic differentiation, that is, using the Adam and L-BFGS algorithms until the model converges, and the mean square error (MSE) is used to judge the convergence of the model. The formula is as follows: (13) In the formula, represents the number of samples, y pre represents the predicted output of the sample, y ref represents the actual reference output of the sample, and the weights are adjusted during the training process to minimize the error between the predicted value and the actual reference value.

[0030] An ocean thermal energy conversion power generation efficiency prediction device includes a device platform, and the device platform includes a pump, an accumulator, a pressure sensor, a flow sensor, a reversing valve, a hydraulic motor, an electric motor, a rectification module, and a load resistor that are connected in sequence to provide a power source; a torque sensor is arranged on the hydraulic motor, a speed sensor is arranged on the electric motor, and a current sensor is arranged on the load resistor; The pump is used to provide a power source for the system. The pressure sensor and the flow sensor respectively detect the pressure and flow rate values of the accumulator in real time. The reversing valve is used to control the working state of the system. The hydraulic motor is used to convert hydraulic energy into mechanical energy. The torque sensor is used to measure the change of the output torque of the hydraulic motor. The motor is used to convert mechanical energy into electrical energy. The speed sensor is used to measure the real-time speed value of the motor. The rectification module is used to convert three-phase alternating current into direct current. The load resistor is used to consume the electric energy output by the motor. At the same time, the current sensor is used to collect the current value flowing through the load resistor; The pressure sensor, the flow sensor, the torque sensor, the speed sensor, and the current sensor are connected to the computer, and the computer executes the steps in the above method.

[0031] This application is directed to an ocean thermal energy conversion power generation system based on phase change materials. Through the latent heat energy storage characteristics of solid-liquid phase change materials, heat can be efficiently absorbed / released during temperature difference fluctuations, improving the instantaneous power output and energy density of the system. At the same time, physical constraints are embedded in machine learning to ensure that the prediction results conform to physical laws, and the system efficiency can also be accurately predicted in the case of insufficient data, so as to timely adjust the device parameters for adaptive optimization. Emphasizing the prediction of efficiency, constructing a PINN neural network can handle complex non-linear relationships, combining physical information (such as thermodynamic equations) and data-driven methods, providing higher prediction accuracy, and at the same time being able to capture the dynamic changes of the system efficiency, suitable for the prediction of long time series data.

[0032] The beneficial effects of the present invention are as follows: 1. The present invention proposes a method and device for predicting the efficiency of ocean thermal energy conversion power generation. Through the self-learning ability of the physics-informed neural network (PINN), the relationship between the network input (accumulator pressure, motor speed, current flowing through the load resistor, and time) and the network output (system efficiency) is established, realizing rapid and accurate prediction of the efficiency of ocean thermal energy conversion power generation. Combining physical models and data-driven machine learning methods, using the physics-informed neural network (PINN) to perform time series prediction on efficiency makes the prediction results more accurate and reliable. Compared with traditional machine learning methods, PINN has higher accuracy and robustness when dealing with complex physical systems, and can also make accurate predictions with a small order of magnitude.

[0033] 2. Compared with the prior art, the present invention uses energy constraints and efficiency constraints, directly embeds physical information into the network structure. By embedding physical constraints into the PINN model, physical rationality can still be ensured in the case of insufficient data, and significantly improves transfer learning outside the small training data set and the conditions used in training, and can achieve high-precision and rapid prediction, improving reliability. The integration of physical knowledge is realized through the physical constraint term in the loss function, which can handle complex non-linear relationships and is suitable for dynamic system modeling.

[0034] 3. The present invention is driven by both physics and data, capable of using physical laws and data-driven methods for time series prediction, accurately predicting the power generation efficiency of ocean thermal energy conversion, still providing more accurate system efficiency prediction in the case of insufficient data, identifying the change trend of system efficiency in advance through prediction, and taking timely measures for adjustment. Moreover, the prediction results conform to the actual change law, enabling timely adjustment and optimization of the structural design and working parameters of the power generation equipment, and simultaneously enabling real-time adjustment of the operation strategy to maximize the energy conversion efficiency, which is of great significance for ocean thermal energy conversion and the development of clean energy, and also plays an important role in promoting the realization of the "dual carbon" goal. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is the amesim simulation model diagram of the ocean thermal energy power generation system of the present invention; Figure 2 It is the flow chart of the implementation method for predicting the power generation efficiency of ocean thermal energy of the present invention; Figure 3 It is the PINN framework diagram for predicting the power generation efficiency of ocean thermal energy of the present invention; Figure 4 It is the schematic diagram of the module connection of the test device platform of the ocean thermal energy power generation system of the present invention; In the figure, 1 - accumulator, 2 - reversing valve, 3 - input power acquisition module, 4 - hydraulic motor, 5 - rotating load, 6 - rotational speed sensor, 7 - motor, 8 - rectification module, 9 - load resistor, 10 - current sensor, 11 - output power acquisition module, 12 - efficiency acquisition module, 13 - pressure sensor, 14 - pressure source. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] To make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the present invention will be further described below through embodiments in conjunction with the drawings, but not limited thereto. For those not elaborated in detail in the present invention, they are all conventional technologies in the art.

[0037] Embodiment 1: This embodiment discloses a method for predicting the power generation efficiency of ocean thermal energy, as Figure 2 shown, including the following steps: Step 1. Build an amesim simulation model of the ocean thermal energy power generation system.

[0038] As Figure 1As shown, the AMESim simulation platform includes a pressure source 14, an accumulator 1, a directional control valve 2, a hydraulic motor 4, a rotational load 5, a motor 7, a rectifier module 8, a load resistor 9, a sensor acquisition module, an input power acquisition module 3, an output power acquisition module 11, and an efficiency acquisition module 12. The sensor acquisition module includes a rotational speed sensor 6, a current sensor 10, and a pressure sensor 13. The accumulator 1 first sets a charging pressure, that is, the initial pressure of the accumulator. P 0, and the pressure source 14 instantaneously replenishes the accumulator 1 to the maximum system pressure. At the same time, the module N2 in the AMESim simulation diagram is a general gas definition, and this module is used to define the gas properties of the entire model. Nitrogen is selected in this model.

[0039] During the overall discharge experiment, the pressure sensor 13 at the outlet of the accumulator 1 can monitor the outlet pressure in real time. The directional control valve 2 can control the working state of the system. When the directional control valve 2 is in the right position, the system does not perform the discharge process, and vice versa. The input power acquisition module 3 can collect the power released by the accumulator 1, that is, the input power of the system. The hydraulic motor 4 can convert hydraulic energy into mechanical energy. The hydraulic motor 4 is connected to a rotational load 5, which can be used to receive the torque fed back by the motor 7. A rotational speed sensor 6 is set between the motor 7 and the rotational load 5 to monitor the rotational speed change of the motor 7 in real time. The rectifier module 8 is used to convert three-phase alternating current into direct current. The load resistor 9 can consume the electricity generated by the motor. The current sensor 10 can monitor the current value flowing through the load resistor 9 in real time. The output power acquisition module 11 can collect the electric power consumed by the load resistor 9, that is, the output power of the system. The efficiency acquisition module 12 can collect the power generation efficiency value of the system. And during the entire discharge experiment process, the energy is transmitted from the accumulator 1 to the hydraulic motor 4, the motor 7, and the load resistor 9, experiencing the conversion from hydraulic energy to mechanical energy to electrical energy.

[0040] Step 2: Conduct a discharge experiment on the accumulator 1, and collect the pressure, motor rotational speed, current flowing through the load resistor 9, and the output power value of the system during the discharge process of the accumulator 1.

[0041] Without considering the energy storage process of the motor 7, connect the output end of the motor 7 to the load resistor 9, and directly consume the electricity generated by the motor 7. At this time, the load power is the output power.

[0042] The method for obtaining the input data set of the training data is: determine that the volume of the accumulator 1 is a certain value, conduct a discharge experiment on the accumulator 1, and respectively collect the pressure P of the accumulator 1, the rotational speed n of the motor 7, the current value I flowing through the load resistor 9, and the power generation efficiency y of the system through sensors.

[0043] The pressure P of the accumulator is a key indicator to measure the amount of energy stored in the accumulator, which is directly related to its ability to provide power to the system. The accumulator flow rate refers to the volume of oil flowing into or out of the accumulator per unit time. In this study, only the discharge process of the accumulator is considered. The charged gas volume is determined to be 0.25L. Using sensors to collect the pressure P data of the accumulator's time series can comprehensively reflect the working state of the entire discharge process of the accumulator and the changes in the system load.

[0044] Step 3: Input the collected parameters into the PINN neural network model and preprocess them.

[0045] To improve the training efficiency and prediction accuracy of the model, the datasets collected in Step 2 are combined into a complete dataset. Each column of data except time is preprocessed by filtering, and at the same time, the data is normalized by the maximum and minimum values, so that the data is mapped to the interval [-1,1], improving the training efficiency and stability of the model. For the filtering process, moving average filtering is used to remove the high-frequency noise in the data. The formula is as follows: (1) In the formula, x[n] represents the data sequence to be processed, y[n] represents the sequence after moving average filtering, N represents the window size of the moving average, and n represents the index of the data point.

[0046] The formula for normalizing the data is as follows: (2) In the formula: x max represents the maximum value in the dataset to be processed, x min represents the minimum value in the dataset to be processed, x m represents the data after normalization processing.

[0047] Step 4: Split the preprocessed dataset into a training set, a test set, and a validation set according to a certain proportion.

[0048] Overfitting may occur during training. The essence of overfitting is that the model over-learns the noise and specific patterns of the training data and fails to effectively capture the real laws behind the data. To solve this problem, the preprocessed dataset is split into a training set, a test set, and a validation set according to a proportion of 6:2:2.

[0049] Step 5: Build a preliminary PINN model for predicting the power generation efficiency of ocean thermal energy conversion.

[0050] This model uses a fully connected multi-layer neural network as a feature code to embed the input parameter features, and at the same time adds multiple physical constraints to improve the reliability of the PINN model output.

[0051] Among them, the PINN neural network prediction model is built based on the Python language.

[0052] The PINN neural network includes an input layer, a hidden layer, and an output layer. The activation function of the output layer is sigmoid, which compresses the output into the interval [0, 1].

[0053] The hidden layer of the neural network has two layers, and each hidden layer has 64 neurons. The hidden layer uses a fully connected layer, and its activation function is tanh (hyperbolic tangent), which compresses the output into the interval [-1, 1].

[0054] Step 6: Use the data of the split training set as the input of the model for training, and obtain the PINN prediction model for the ocean thermal energy conversion power generation efficiency through multiple training updates. The training process is as follows: First, bring the data of the divided training set into the built PINN neural network model; then initialize the network weights, set the learning rate and optimizer; input the collected time series data and physical parameters, calculate the data-driven loss and physical constraint loss respectively, and sum them up with weights to obtain the total loss L; then use automatic differentiation for optimization, propagate the optimized correction value forward, and update the grid weights; finally, after multiple iterations of update, output the PINN neural network model for ocean thermal energy conversion power generation to complete the training.

[0055] Among them, the input and output layers of the PINN neural network are both parameters of the time series. Set X_f as the input layer of the PINN neural network, X_f = [t, P, n, I], and Y_f as the output layer of the PINN neural network, Y_f = y pre , each time point t corresponds to a P and Q, and at the same time corresponds to a y pre .

[0056] Two optimizers, adam and lbfgs, are provided for learning during model training; there are two optimizers in the model. First, use adam during training, and then use lbfgs.

[0057] Use the Adam optimizer for learning. Based on the optimization algorithm of gradient descent, the learning rate is 1×10 3 , the total number of training times is 2000, and the model performance is verified on the test data every 100 rounds.

[0058] Use the lbfgs optimizer for learning. Adopt the second-order optimization algorithm, set the initial learning rate to 1, and then automatically determine the step size through search without manually setting the learning rate. In each iteration process, find a step size along the current search direction α, so that the objective function f ( x + αd ) meets the descent condition. The total number of training times is 2000, and the model performance is verified on the test data every 100 rounds.

[0059] The loss function of the PINN neural network is L, and its calculation formula is as follows: L = λ d L data + λ p L phys + λ b L bound (3) In the formula, L data represents the data loss term of the PINN neural network, L phys is the physical loss term, L bound is the efficiency boundary loss term, and λ d , λ p , λ b are the weights of different terms respectively.

[0060] λ d , λ p , λ b The weight values are given by the parameter training network. Define the initial weights as , , . Each training will save and update the weight values. The updated weight is . Substitute λ 1 into the PINN neural network to continue participating in the model training.

[0061] Train through multiple loop iterations until the PINN neural network model for predicting the ocean thermal energy conversion efficiency converges.

[0062] The data loss term L data of the PINN neural network adopts the mean square error, and the calculation formula is as follows: (4) In the formula, N is the number of initial samples, ti Indicates the sample point corresponding to the i th time point, y pre Indicates the predicted value obtained after training by the PINN neural network model, y data Indicates the output sample data at the initial moment, y pre ( t i ) Indicates the predicted value obtained by training the data of the i th initial sample time point at the initial moment through the PINN neural network model, y data ( t i ) Indicates the output sample data of the i th initial sample time point at the initial moment.

[0063] The physical loss term of the PINN neural network L phys Is the residual of the physical equation, and the calculation formula is as follows: (5) (6) (7) In the formula, M is the number of physical constraint sampling points, t j Indicates the j th sampling point corresponding to the time point, f phys Indicates the physical equation of the PINN neural network model, f phys ( t j ) Indicates the j th physical equation corresponding to the sampling time point, y pre ( t j ) Indicates the predicted value of the system efficiency corresponding to the j th sampling time point, R represents the resistance value of the load resistor, P Indicates the accumulator pressure, Indicates the angular velocity of the motor.

[0064] According to the system energy conservation and Newton's second law during the accumulator discharge process, the following formula is obtained: (8) In the formula, V 0, Prespectively represent the initial volume and initial pressure of nitrogen in the accumulator, Q ( t j ) represents the hydraulic motor inlet flow rate value corresponding to the j th sampling time point, represents the adiabatic index of the gas (here refers to nitrogen), represents the effective volume elastic modulus of the accumulator, represents the j th sampling time point corresponding , J represents the moment of inertia, c represents the damping constant, D represents the displacement of the hydraulic motor, k e represents the back electromotive force constant, I ( t j ) represents the current value flowing through the load resistor corresponding to the j th sampling time point, represents the j th sampling time point corresponding .

[0065] In the system, the motor speed n, the motor angular velocity and the flow rate Q have the following relationships respectively: (9) Q = nD (10) Therefore, substituting equations (7), (8), and (10) into equation (6) for calculation gives: (11) where V 0, P 0, , , J , c , D , k e , R are all known and determined values.

[0066] The boundary condition for restricting the efficiency is 0 < y pre < y max , so the efficiency boundary loss term L bound of the PINN neural network is calculated as follows: (12) In the formula, ReLU is the activation function, which is used to penalize values outside the allowed range,ReLU ( y pre ( t i ) - y max ) indicates that when the prediction efficiency of the PINN neural network y pre ( t i ) exceeds the efficiency upper limit y max at this time, the value of this item is positive, otherwise the value is 0. ReLU (- y pre ( t i )) indicates that when the predicted value of the PINN neural network y pre ( t i ) is negative, the value of this item is positive, otherwise the value is 0.

[0067] When training the neural network, the initial conditions are fitted through supervised learning, while the power and efficiency constraints are incorporated into the loss function in an unsupervised form. The optimization process is optimized using automatic differentiation, that is, the Adam or L-BFGS algorithm can be used until the model converges; the mean squared error (MSE) is used to judge the convergence of the model, and the formula is as follows: (13) In the formula, represents the number of samples, y pre represents the predicted output of the sample, y ref represents the actual reference output of the sample, and the weights are adjusted during the training process to minimize the error between the predicted value and the actual reference value. As Figure 3 shown, when the mean squared error MSE is less than the set value ε, the model converges and the training is completed, otherwise the training continues.

[0068] Step 7, build an experimental device platform for an ocean thermal energy conversion power generation system; As Figure 4 shown, the device platform includes a pump, an accumulator, a pressure sensor, a flow sensor, a reversing valve, a hydraulic motor, an electric motor, a rectification module, and a load resistor that are connected in sequence to provide a power source; a torque sensor is provided on the hydraulic motor, a speed sensor is provided on the electric motor, and a current sensor is provided on the load resistor. Compared with the virtual amesim simulation model, the experimental device platform is a specific device built in reality.

[0069] Step 8: Collect the pressure, motor speed, current value flowing through the load resistor, and power generation efficiency value during the discharge process of some accumulators through the test device bench. Input the collected actual experimental parameters into the trained PINN prediction model for fine-tuning to obtain the final prediction model; There may be some differences between the simulation data and the real experimental data. The PINN prediction model trained with simulation data is subsequently fine-tuned using the experimental data on a small scale to improve the prediction accuracy of the model.

[0070] Step 9: Input the newly collected test parameters into the PINN prediction model for the power generation efficiency based on ocean thermal energy conversion, and finally obtain the corresponding system prediction efficiency.

[0071] Embodiment 2 An apparatus for predicting the power generation efficiency of ocean thermal energy conversion includes a device bench, and the schematic diagram of its module connection is as Figure 4 shown.

[0072] The device bench includes a pump, an accumulator, a pressure sensor, a flow sensor, a directional control valve, a hydraulic motor, an electric motor, a rectification module, and a load resistor that are connected in sequence to provide a power source. A torque sensor is provided on the hydraulic motor, a speed sensor is provided on the electric motor, and a current sensor is provided on the load resistor.

[0073] The pump source provides a power source for the system. The pressure sensor and the flow sensor respectively detect the pressure and flow rate values of the accumulator in real time. The directional control valve can be used to control the working state of the system. The hydraulic motor converts hydraulic energy into mechanical energy. The torque sensor can measure the change in the output torque of the hydraulic motor. The electric motor can convert mechanical energy into electrical energy. The speed sensor can detect the real-time speed value of the electric motor. The rectification module can convert three-phase alternating current into direct current. The load resistor is used to consume the electric energy output by the electric motor, and at the same time, the current sensor can collect the current value flowing through the load resistor.

[0074] The pressure sensor, the flow sensor, the torque sensor, the speed sensor, and the current sensor are connected to a computer, and the computer executes the steps in the method of Embodiment 1.

[0075] The above is the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle described in the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for predicting the power generation efficiency of ocean thermal energy conversion, characterized in that, It includes the following steps: Step 1: Build an Amesim simulation model of an ocean thermal energy conversion power generation system; Step 2: Conduct an accumulator discharge experiment, and collect the pressure, motor speed, current value flowing through both ends of the load resistor, and system output power value during the accumulator discharge process; Step 3: Input the collected parameters into the PINN neural network model and preprocess them; Step 4: Split the preprocessed dataset into a training set, a test set, and a validation set according to a certain ratio; Step 5: Build a preliminary PINN model for predicting the ocean thermal energy conversion power generation efficiency; Step 6: Use the data of the split training set as the input of the model for training, and through multiple trainings and updates, obtain a PINN prediction model for the ocean thermal energy conversion power generation efficiency; Step 7: Build a test bench for the ocean thermal energy conversion power generation system; Step 8: Through the test bench experiment, collect the pressure, motor speed, current value flowing through the load resistor, and power generation efficiency value during part of the accumulator discharge process. Input the collected actual experimental parameters into the trained PINN prediction model to fine-tune it, and obtain the final prediction model; Step 9: Input newly collected test parameters into the PINN prediction model based on the ocean thermal energy conversion power generation efficiency, and finally obtain the corresponding system prediction efficiency.

2. The method for predicting the power generation efficiency of ocean thermal energy conversion according to claim 1, wherein In Step 1, building the Amesim simulation platform for the ocean thermal energy conversion power generation system includes a pressure source, an accumulator, a directional control valve, a hydraulic motor, a rotating load, a motor, a rectification module, a load resistor, a sensor acquisition module, a power acquisition module, and an efficiency acquisition module. The pressure source is used to provide power for the system. The directional control valve is used to control the working state of the system. The hydraulic motor is used to convert hydraulic energy into mechanical energy. The hydraulic motor is connected to a rotating load to receive the torque fed back by the motor. The motor is used to convert mechanical energy into electrical energy. The rectification module is used to convert three-phase alternating current into direct current. The load resistor is used to consume the electrical energy output by the motor. The sensor acquisition module includes a speed sensor, a current sensor, and a pressure sensor, which are used to collect key parameters in real time. And during the whole discharge experiment process, the energy is transmitted from the accumulator to the hydraulic motor, the motor, and the load resistor, experiencing the conversion from hydraulic energy to mechanical energy to electrical energy. The power acquisition module is used to collect the input power and output power of the system. The efficiency acquisition module is used to collect the system power generation efficiency y.

3. The method for predicting the power generation efficiency of ocean thermal energy conversion according to claim 1, wherein Step 2 is to conduct an accumulator discharge experiment using the simulation platform built in Step 1. Without considering the energy storage process of the motor, connect the output end of the motor to the load resistor, directly consume the electrical energy generated by the motor, and connect a current sensor beside the load resistor to collect the current value flowing through the load resistor; The method for obtaining the input dataset of the training data is: determine that the volume of the accumulator is a certain value, conduct an accumulator discharge experiment, and collect the pressure P of the accumulator, the motor speed n, the current I value flowing through the load resistor, and the system power generation efficiency y through sensors respectively.

4. The method for predicting the power generation efficiency of ocean thermal energy conversion according to claim 1, wherein Step 3 is to synthesize the collected data sets in Step 2 into a complete data set. The data set contains five columns of data: time, accumulator pressure, motor speed, current flowing through the load resistor, and power generation efficiency. Filter preprocessing is performed on each column of data except time, and at the same time, maximum-minimum normalization processing is performed on the data to map the data to the interval [-1, 1]. For the filtering process, moving average filtering is used to remove high-frequency noise in the data. The formula is as follows: （1） In the formula, x[n] represents the data sequence to be processed, y[n] represents the sequence after moving average filtering, N represents the window size of the moving average, and n represents the index of the data point; The formula for data normalization is as follows: （2） Wherein: x max represents the maximum value in the dataset to be processed, x min represents the minimum value in the dataset to be processed, x m represents the data after normalization processing.

5. The method for predicting the power generation efficiency of ocean thermal energy conversion according to claim 1, wherein In Step 4, the preprocessed data set is split into a training set, a test set, and a validation set according to a ratio of 6:2:

2.

6. The method for predicting the power generation efficiency of ocean thermal energy conversion according to claim 1, wherein In Step 5, a preliminary PINN model for predicting the power generation efficiency of ocean thermal energy conversion is built. This model uses a fully connected multi-layer neural network as a feature code to embed the input parameter features, and at the same time adds multiple physical constraints; Among them, the PINN neural network model is built based on the Python language; The PINN neural network includes an input layer, a hidden layer, and an output layer. The activation function of the output layer is sigmoid, which compresses the output to the interval [0, 1]; The hidden layer of the neural network has two layers, and each hidden layer has 64 neurons. The hidden layer uses a fully connected layer, and its activation function is tanh, which compresses the output to the interval [-1, 1].

7. The method for predicting the power generation efficiency of ocean thermal energy conversion according to claim 1, wherein In Step 6, the model is trained. The training process is as follows: First, bring the data of the divided training set into the built PINN neural network model; then initialize the network weights, set the learning rate and optimizer; input the collected time series data and physical parameters, calculate the data-driven loss and physical constraint loss respectively, and sum them weighted to obtain the total loss L; then use automatic differentiation for optimization, propagate the optimized correction value forward, and update the grid weights; finally, after multiple iterative updates, output the PINN neural network model for ocean thermal energy conversion power generation to complete the training.

8. The method for predicting the power generation efficiency of ocean thermal energy conversion according to claim 7, wherein In Step 6, two optimizers, adam and lbfgs, are provided for learning during model training; there are two optimizers in the model. First, adam is used during training, and then lbfgs is used; Learning is carried out using the Adam optimizer, an optimization algorithm based on gradient descent, with a learning rate of 1×10 3 , the total number of training times is 2000, and the model performance is verified on the test data every 100 rounds; Learn using the LBFGS optimizer, adopt a second-order optimization algorithm, set the initial learning rate to 1, and then automatically determine the step size through search without manually setting the learning rate. In each iteration process, find a step size along the current search direction α , such that the objective function f ( x + αd ) meets the descent condition. The total number of training times is 2000, and the model performance is verified on the test data every 100 rounds.

9. The method for predicting the power generation efficiency of ocean thermal energy conversion according to claim 7, characterized in that, In Step 6, the loss function of the PINN neural network is L, and its calculation formula is as follows: L = λ d L data + λ p L phys + λ b L bound （3） In the formula, L data represents the data loss term of the PINN neural network, L phys is the physical loss term, L bound is the efficiency boundary loss term, and λ d 、 λ p 、 λ b are the weights of different terms respectively; λ d , λ p , λ b The weight value is given by the parameter training network, and the initial weight is defined as , , . Each training will save and update the weight value, and the updated weight is . Substitute λ 1 into the PINN neural network to continue participating in the model training; Perform multiple loop iterations of training until the PINN neural network model for predicting the power generation efficiency of ocean thermal energy conversion converges; Data loss term of PINN neural network L data The mean square error is adopted, and the calculation formula is as follows: （4） Where N is the number of initial samples, t i represents the sample point corresponding to the i th time point, y pre represents the predicted value obtained after training by the PINN neural network model, y data represents the output sample data at the initial moment, y pre ( t i ) represents the predicted value obtained after training the data at the initial moment of the i th initial sample time point through the PINN neural network model, y data ( t i ) represents the output sample data of the i th initial sample time point at the initial moment; Physical loss term of PINN neural network L phys is the residual of the physical equation, and the calculation formula is as follows: （5） （6） （7） Where M is the number of physically constrained sampling points, t j denotes the sampling point corresponding to the j th time point, f phys represents the physical equation of the PINN neural network model, f phys ( t j ) denotes the physical equation corresponding to the j th sampling time point, y pre ( t j ) denotes the predicted value of the system efficiency corresponding to the j th sampling time point, R represents the resistance value of the load resistor, P represents the accumulator pressure, represents the angular velocity of the motor; According to the system energy conservation and Newton's second law during the accumulator discharge process, the following formula is obtained: （8） Wherein, V 0, P 0 respectively represent the initial volume and initial pressure of nitrogen in the accumulator, Q ( t j ) represents the hydraulic motor inlet flow rate value corresponding to the j th sampling time point, represents the adiabatic index of the gas, represents the effective volume elastic modulus of the accumulator, represents the j th sampling time point corresponding , J represents the moment of inertia, c represents the damping constant, D represents the displacement of the hydraulic motor, k e represents the back electromotive force constant, I ( t j ) represents the current value flowing through the load resistor corresponding to the j th sampling time point, represents the j th sampling time point corresponding ; The motor speed n and the motor angular velocity in the system and the flow rate Q have the following relationships respectively: （9） Q=nD(10) Therefore, substituting equations (7), (8), and (10) into equation (6) for calculation gives: （11） Among them V 0, P 0, , , J , c , D , k e , and R are all known and determined values; The boundary condition that limits efficiency is 0 < y pre < y max , so the efficiency boundary loss term of the PINN neural network L bound The calculation formula is as follows: （12） In the formula, ReLU is an activation function, which is used to penalize values outside the allowed range. ReLU ( y pre ( t i ) - y max ) represents that when the prediction efficiency y pre ( t i ) of the PINN neural network exceeds the efficiency upper limit y max , the value of this term is positive, otherwise the value is 0. ReLU (- y pre ( t i )) represents that when the predicted value y pre ( t i ) of the PINN neural network is negative, the value of this term is positive, otherwise the value is 0. When training the neural network, the initial conditions are fitted through supervised learning, while the power and efficiency constraints are incorporated into the loss function in an unsupervised form; the optimization process uses automatic differentiation for optimization, that is, the Adam and L-BFGS algorithms are used until the model converges, and the mean square error (MSE) is used to judge the convergence of the model. The formula is as follows: （13） Wherein, represents the number of samples, y pre represents the predicted output of the sample, y ref represents the actual reference output of the sample. During the training process, the weights are adjusted to minimize the error between the predicted value and the actual reference value.

10. An apparatus for predicting the power generation efficiency of ocean thermal energy conversion, characterized in that, It includes a device platform, and the device platform includes a pump for providing a power source, an accumulator, a pressure sensor, a flow sensor, a directional control valve, a hydraulic motor, an electric motor, a rectification module, and a load resistor that are connected in sequence; a torque sensor is provided on the hydraulic motor, a rotational speed sensor is provided on the electric motor, and a current sensor is provided on the load resistor; The pump is used to provide a power source for the system. The pressure sensor and the flow sensor respectively detect the pressure and flow rate values of the accumulator in real time. The directional control valve is used to control the working state of the system. The hydraulic motor is used to convert hydraulic energy into mechanical energy. The torque sensor is used to measure the change in the output torque of the hydraulic motor. The electric motor is used to convert mechanical energy into electrical energy. The rotational speed sensor is used to measure the real-time rotational speed value of the electric motor. The rectification module is used to convert three-phase alternating current into direct current. The load resistor is used to consume the electrical energy output by the electric motor. At the same time, the current sensor is used to collect the current value flowing through the load resistor; The pressure sensor, the flow sensor, the torque sensor, the rotational speed sensor, and the current sensor are connected to a computer, and the computer executes the steps in the method for predicting the power generation efficiency of ocean thermal energy conversion according to any one of claims 1-9.

Images