Self-adaptive optimization control method for electric propulsion inertia of hydrogen fuel cell ship

Abstract

The invention discloses a hydrogen fuel cell ship electric propulsion inertia self-adaptive optimization control method, which relates to the technical field of hydrogen fuel cell ship electric propulsion control, and comprises the following steps of: acquiring ship operation state data and ship internal power network fluctuation data in real time, preprocessing, extracting features, calculating a virtual inertia coefficient and a virtual damping coefficient, and calculating an inertia coefficient and a damping coefficient; generating a virtual torque instruction; obtaining a total power demand of the ship by using a torque rotating speed calculation method, and dividing the total power demand into high-frequency, intermediate-frequency and low-frequency ship power distribution instructions; through a virtual synchronous machine model, the actual power of the ship is output, the voltage of a direct-current bus and the power of a hydrogen fuel cell are monitored in real time, a dynamic compensation power fluctuation method is utilized, the operation state of the ship is regulated and controlled, and a voltage stability index is obtained; the ship propulsion inertia is adaptively optimized through a power regulation algorithm, so that the ship can better adapt to different working conditions, and the stability and safety of the ship are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of hydrogen fuel cell ship electric propulsion control, and in particular to an inertial adaptive optimization control method for hydrogen fuel cell ship electric propulsion. Background Art

[0002] With the continuous improvement of the requirements for ship power performance, energy efficiency, and navigation safety, in the early stage, ships mainly relied on traditional diesel engines for direct propulsion. This method has a simple structure, but low energy utilization efficiency and pollution. With the development of power electronics technology and motor control technology, electric propulsion has begun to be applied to ships. Hydrogen fuel cell ships convert the electrical energy of the ship's hydrogen fuel cells into mechanical energy required for ship propulsion, with higher energy utilization efficiency, better maneuverability, and lower emissions. When quickly changing the power demand according to different navigation states of hydrogen fuel cell ships, it is impossible to accurately allocate power and dynamically respond. The control accuracy and response speed are insufficient under high-frequency fluctuations, making it difficult to maintain stability. Moreover, traditional methods cannot calculate the virtual inertia and damping coefficients in a timely and accurate manner, affecting the propulsion dynamic performance and power voltage stability. Summary of the Invention

[0003] In view of the above existing problems, the present invention is proposed.

[0004] Therefore, the present invention provides an inertial adaptive optimization control method for hydrogen fuel cell ship electric propulsion to solve the problem of inaccurate power distribution and dynamic response under complex working conditions of hydrogen fuel cell ships.

[0005] To solve the above technical problems, the present invention provides the following technical solutions: In the first aspect, the present invention provides an inertial adaptive optimization control method for hydrogen fuel cell ship electric propulsion, which includes: collecting ship operation state data and ship internal power network fluctuation data in real time, performing preprocessing and extracting features, calculating virtual inertia coefficients and virtual damping coefficients, and generating virtual torque commands; using a torque-speed calculation method to obtain the total ship power demand and divide it into high-frequency, medium-frequency, and low-frequency ship power distribution commands; through a virtual synchronous machine model, outputting the actual ship power, real-time monitoring the DC bus voltage and hydrogen fuel cell power, and using a dynamic compensation power fluctuation method to regulate the ship operation state to obtain a voltage stability index; based on the actual ship power and voltage stability index, calculating the current optimal virtual inertia coefficient through a reinforcement learning model, and using a power adjustment algorithm to adaptively optimize the ship propulsion inertia.

[0006] As a preferred embodiment of the inertial adaptive optimization control method for the power propulsion of a hydrogen fuel cell ship according to the present invention, wherein: the ship operation state data and the ship internal power network fluctuation data are collected in real time, preprocessed, and features are extracted. The specific steps are as follows. The ship operation state data is obtained by collecting the ship acceleration, angular velocity change, propeller thrust change, wind speed, and wave height data in real time, and the ship internal power network fluctuation data is obtained by collecting the ship DC bus voltage and the hydrogen fuel cell output power in real time. The linear interpolation method is used to fill in the missing values, and the outlier detection method is used to remove the outliers. The Z-score method is used to standardize to a unified range, and the frequency domain analysis method is used to extract features, obtaining the ship operation state features and the ship internal power network fluctuation features.

[0007] As a preferred embodiment of the inertial adaptive optimization control method for the power propulsion of a hydrogen fuel cell ship according to the present invention, wherein: the virtual inertia coefficient and the virtual damping coefficient are calculated to generate a virtual torque command. The specific steps are as follows. Through the ship propeller thrust change and the ship internal power network fluctuation data, the frequency domain analysis method is used for analysis to obtain the electromagnetic torque dynamic characteristics. Based on the ship operation state features and the ship internal power network fluctuation features, the virtual inertia coefficient and the virtual damping coefficient are calculated. Combining the ship angular velocity change and the electromagnetic torque dynamic characteristics, a virtual torque command is generated through the virtual torque calculation method.

[0008] As a preferred embodiment of the inertial adaptive optimization control method for the power propulsion of a hydrogen fuel cell ship according to the present invention, wherein: the torque-speed calculation method is used to obtain the total ship power demand and divide it into high-frequency, medium-frequency, and low-frequency ship power distribution commands. The specific steps are as follows. Based on the virtual torque command and the ship angular velocity change, combining the ship operation state features and the ship internal power network fluctuation features, the torque-speed calculation method is used to obtain the total ship power demand. Through the frequency domain analysis method, the total ship power demand is decomposed in the frequency domain to divide the ship power distribution commands for high-frequency, medium-frequency, and low-frequency powers.

[0009] As a preferred embodiment of the inertial adaptive optimization control method for the power propulsion of a hydrogen fuel cell ship according to the present invention, wherein: the ship actual power is output through the virtual synchronous machine model. The specific steps are as follows. Based on the historical ship power distribution commands, decomposition is performed through the fast Fourier transform to extract the ship power dynamic response features. The sliding window is used for statistical analysis to extract statistical features, and the parameters of the virtual synchronous machine model are initialized. Based on the dynamic response characteristics and statistical characteristics of the ship's power, the virtual synchronous machine model is trained using the gradient descent optimization training method, and the actual ship power is output.

[0010] As a preferred solution of the hydrogen fuel cell ship electric propulsion inertia adaptive optimization control method described in the present invention, wherein: the DC bus voltage and the hydrogen fuel cell power are monitored in real time, and the dynamic compensation power fluctuation method is used to regulate the ship's operating state to obtain the voltage stability index. The specific steps are as follows. Based on the actual ship power, the changes in the DC bus voltage, the hydrogen fuel cell power, and the ship power demand are monitored in real time, and the power fluctuation characteristics are extracted. Using the dynamic compensation power fluctuation method, the influence of the power fluctuation characteristics on the ship's operation is compensated in real time, the ship's operating state is adjusted, and the voltage stability index is obtained.

[0011] As a preferred solution of the hydrogen fuel cell ship electric propulsion inertia adaptive optimization control method described in the present invention, wherein: based on the actual ship power and the voltage stability index, the current optimal virtual inertia coefficient is calculated through the reinforcement learning model. The specific steps are as follows. Based on the historical actual ship power, the state space and the action space are defined, the reward function is defined using the historical voltage stability index, and the reinforcement learning model framework is constructed. The reinforcement learning model framework is trained using the historical actual ship power and the historical voltage stability index to obtain the reinforcement learning model. The actual ship power and the voltage stability index are input into the reinforcement learning model to calculate the current optimal virtual inertia coefficient.

[0012] As a preferred solution of the hydrogen fuel cell ship electric propulsion inertia adaptive optimization control method described in the present invention, wherein: the power adjustment algorithm is used to adaptively optimize the ship's propulsion inertia. The specific steps are as follows. Based on the optimal virtual inertia coefficient, combined with the ship power distribution commands of high frequency, medium frequency, and low frequency of the ship, the power adjustment algorithm is used to regulate the ship's operating state in real time, and the inertia response characteristics and dynamic response characteristics of the ship are optimized.

[0013] In a second aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the hydrogen fuel cell ship electric propulsion inertia adaptive optimization control method described in the first aspect of the present invention is implemented.

[0014] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the hydrogen fuel cell ship electric propulsion inertia adaptive optimization control method described in the first aspect of the present invention is implemented.

[0015] The beneficial effects of the present invention are as follows: By using the torque-speed calculation method to calculate the total power demand of the ship, the total power demand is divided into high-frequency, medium-frequency, and low-frequency power distribution commands. After frequency-domain decomposition, the dynamic output of the ship at different time scales can be controlled. It helps the ship to reasonably distribute power according to requirements in various navigation states, improve energy utilization efficiency, and make the operation of the ship more stable and efficient. By calculating the current optimal virtual inertia coefficient through the reinforcement learning model, the control strategy of the ship can be dynamically adjusted according to the actual operating conditions, with strong adaptability. Using the power regulation algorithm to adaptively optimize the ship's propulsion inertia can enable the ship to better adapt to different working conditions and enhance the stability and safety of the ship. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 It is a flowchart of the hydrogen fuel cell ship electric propulsion inertia adaptive optimization control method.

[0018] Figure 2 It is a flowchart of the operation of the virtual synchronous machine model.

[0019] Figure 3 It is a flowchart of power distribution and regulation.

[0020] Figure 4 It is a flowchart of the training and application of the reinforcement learning model. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings of the specification.

[0022] Many specific details are set forth in the following description in order to provide a thorough understanding of the present invention. However, the present invention may be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0023] Second, the so-called "one embodiment" or "embodiment" herein refers to specific features, structures, or characteristics that may be included in at least one implementation manner of the present invention. The phrase "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it an individual or alternative embodiment that is mutually exclusive with other embodiments.

[0024] Referring to Figures 1 to 4 , this embodiment provides a method for inertial adaptive optimization control of a hydrogen fuel cell ship's electric propulsion, including the following steps: S1. Collect the ship's operating state data and the fluctuation data of the ship's internal power network in real time, perform preprocessing, and extract features.

[0025] S1.1. Collect the ship's acceleration, angular velocity change, propeller thrust change, wind speed, and wave height data in real time to obtain the ship's operating state data, and collect the DC bus voltage of the ship and the output power of the hydrogen fuel cell in real time to obtain the fluctuation data of the ship's internal power network.

[0026] It should be noted that the acceleration change values of the ship in three-dimensional space are obtained at a fixed frequency through an acceleration sensor, and the acceleration values at different times are recorded; the angular velocity change data of the ship is collected, and the gyroscope sensor is used to measure the rotational angular velocity of the ship around each axis to obtain the dynamic change values of the angular velocity; the propeller thrust change data is collected, and the thrust sensor is used to detect the magnitude of the thrust generated by the propeller, and the change values of the thrust are continuously recorded; the wind speed data is collected, and the anemometer installed on the ship's superstructure is used to measure the wind speed around the ship to obtain the wind speed values at different time points; the wave height data is collected, and the wave height sensor at the bottom of the ship is used to detect the wave height, and the dynamic change values of the wave height are recorded. The collected acceleration, angular velocity change, propeller thrust change, wind speed, and wave height values are arranged in the order of a unified timestamp to obtain the ship's operating state data; The DC bus voltage data is collected in real time through a voltage sensor, the voltage value of the DC bus is collected, and the voltage fluctuation condition is recorded; the power sensor deployed at the output end of the hydrogen fuel cell collects the output power data of the hydrogen fuel cell in real time, and the change values of the output power of the hydrogen fuel cell are continuously recorded to obtain the fluctuation data of the ship's internal power network. The collected DC bus voltage of the ship and the output power of the hydrogen fuel cell are arranged in the order of a unified timestamp to obtain the fluctuation data of the ship's internal power network.

[0027] S1.2. Use the linear interpolation method to fill in the missing values and eliminate the outliers through the threshold detection method.

[0028] It should be noted that the expression for filling in the missing values using the linear interpolation method is: ; Where Is the compensated data, Is the value of the known data point before the missing value, Is the value of the known data point after the missing value, Is the time of the previous known data point, Is the time of the next known data point, Is the time point for which interpolation is required; It should be noted that for the ship operation status data and the ship internal power network fluctuation data, the variables of the ship operation status data and the ship internal power network fluctuation data are traversed and checked one by one. When a missing value is found, the previous normal data point and the next normal data point at the position of the missing value are determined. Calculate the difference between the values of the previous normal data point and the next normal data point, and then calculate the ratio of this difference to the time interval between the previous normal data point and the next normal data point, so as to obtain the change amount per unit time. According to the time interval between the missing value and the previous normal data point, multiply by the change amount per unit time to obtain the increment of the missing value relative to the previous normal data point, add the value of the previous normal data point to this increment to obtain the value used to fill the missing value, and fill this value into the position where the missing value is located; After completing the filling of the missing values, set the anomaly detection threshold according to the range of the ship operation status data and the ship internal power network fluctuation data, and judge the ship operation status data and the ship internal power network fluctuation data one by one. When the data point of a certain variable exceeds the anomaly detection threshold range (such as working condition related safety limits such as voltage fluctuation ±10%, rotational speed deviation ±5%, etc.), it is marked as an outlier. Remove the data marked as an outlier from the data set of the corresponding variable to ensure that only the ship operation status data and the ship internal power network fluctuation data within the normal range are retained in each data set, and complete the filling of the missing values and the removal of the outliers in the ship operation status data and the ship internal power network fluctuation data.

[0029] S1.3. Standardize to a unified range using the Z-score method and perform feature extraction through the frequency domain analysis method to obtain the ship operation status features and the ship internal power network fluctuation features.

[0030] It should be noted that the Z-score standardization process is performed on the overall ship operation status data and the ship internal power network fluctuation data. Calculate the sum of all values of the ship operation status data and the ship internal power network fluctuation data, and divide it by the total number of the ship operation status data and the ship internal power network fluctuation data to obtain the mean value. Then calculate the sum of the squares of the differences between the ship operation status data and the ship internal power network fluctuation data and the mean value, divide it by the total number of the ship operation status data and the ship internal power network fluctuation data, and take the square root to obtain the standard deviation. Subtract the mean value from the ship operation status data and the ship internal power network fluctuation data values and then divide by the standard deviation to complete the standardization process; Arrange the standardized ship operation status data and the ship internal power network fluctuation data in order to form a data sequence, and equally divide it into even segments. For the ship operation status data and the ship internal power network fluctuation data, use the butterfly operation. First, divide the ship operation status data and the ship internal power network fluctuation data into odd and even groups, and perform operations of multiplying by the rotation factor, adding and subtracting respectively to obtain intermediate results. Then repeat the above grouping operations on the intermediate results until the frequency-domain ship operation status data and the ship internal power network fluctuation data are obtained. After that, analyze the spectrum to determine the main frequency components and amplitudes, and extract the characteristics related to the ship operation status and the ship internal power network fluctuation, including the propeller and structural vibration frequencies, acceleration and angular velocity amplitudes in the ship operation status, and the power supply frequency, harmonic frequency components, voltage and power amplitudes in the ship internal power network fluctuation, etc., to obtain the ship operation status characteristics and the ship internal power network fluctuation characteristics.

[0031] S2. Calculate the virtual inertia coefficient and the virtual damping coefficient to generate a virtual torque command.

[0032] S2.1. Analyze the electromagnetic torque dynamic characteristics by using the frequency-domain analysis method through the ship propeller thrust change and the ship internal power network fluctuation data.

[0033] It should be noted that the fast Fourier transform is respectively performed on the ship propeller thrust change data and the ship internal power network fluctuation data to convert the ship propeller thrust change data and the ship internal power network fluctuation data in the time domain into the frequency-domain representation, and obtain the spectra of the ship propeller thrust change data and the ship internal power network fluctuation data. Search for the characteristic frequencies related to the electromagnetic torque in the spectra. The amplitude and phase information corresponding to the characteristic frequencies related to the electromagnetic torque can reflect the characteristics of the electromagnetic torque at different frequencies. Organize and analyze the characteristic frequency amplitudes and phase data in the propeller thrust spectrum and the power grid fluctuation spectrum. By cross-comparing the frequency-domain peaks, phase differences and harmonic distributions of the two, identify the resonance frequency, harmonic coupling strength and dynamic response delay of the electromagnetic torque, and combine the typical ship working conditions (such as the correlation between the thrust mutation during rapid acceleration and the grid voltage flicker) to determine the frequency-variable characteristics of the electromagnetic torque; S2.2. Calculate the virtual inertia coefficient and the virtual damping coefficient based on the characteristics of the ship's operating state and the fluctuations of the internal ship power network. Combine the angular velocity change of the ship and the dynamic characteristics of the electromagnetic torque, and generate a virtual torque command through the virtual torque calculation method.

[0034] It should be noted that a large amount of data on the angular velocity change of the ship, the dynamic characteristics of the electromagnetic torque, and the corresponding actual operating effects under different states are collected. Using the least squares fitting algorithm, a function expression that can describe the relationship between the angular velocity change of the ship, the dynamic characteristics of the electromagnetic torque, and the corresponding actual operating effects is found. Thus, the reference values of the initial virtual inertia coefficient and the virtual damping coefficient are calculated based on the angular velocity change and the dynamic characteristics of the electromagnetic torque in the current state mode. Obtain the reference values of the initial virtual inertia coefficient and the virtual damping coefficient. Combine the characteristics of the fluctuations of the internal ship power network, calculate the difference in the power network fluctuations in adjacent time intervals, and obtain a sequence of power change amounts. Perform a fast Fourier transform on the sequence of power change amounts to convert the time-domain data into frequency-domain data. In the frequency-domain data, identify the frequency components with cumulative contributions exceeding 80% of the total energy or significant local peaks (such as amplitudes higher than 3 dB of adjacent frequencies). Combine the typical resonance frequency bands of ship power (such as the 50 Hz fundamental wave, 100 Hz second harmonic, etc.) to determine the main frequency components with concentrated energy, and thus determine the fluctuation frequency of the power output in the ship power network. Calculate the standard deviation of the power change amount within a certain time period as the change rate of the net power fluctuation in the ship power network, and perform weighted calculation and adjustment. Different weights are assigned according to the importance of the fluctuation frequency of the power output in the ship power network and the change rate of the net power fluctuation in the ship power network. The virtual inertia coefficient and the virtual damping coefficient under the current actual situation are obtained through weighted summation. It should be noted that the expression for calculating the virtual inertia coefficient is: ; Where is the virtual inertia coefficient, is the fluctuation frequency of the power output in the ship power network, is the frequency weight coefficient, is the amplitude of the power fluctuation in the ship power network, is the amplitude weight coefficient; It should be noted that the expression for calculating the virtual damping coefficient is: ; Where is the virtual damping coefficient, is the virtual damping weight coefficient, is the change rate of the net power fluctuation in the ship power network; Substitute the obtained virtual inertia coefficient, virtual damping coefficient, and data related to the change in ship angular velocity and the dynamic characteristics of electromagnetic torque into the formula of the virtual torque calculation method. Determine the change rate of electromagnetic torque according to the dynamic characteristics of electromagnetic torque. Multiply the virtual inertia coefficient by the differential value of the angular velocity change, and multiply the virtual damping coefficient by the change rate of electromagnetic torque. Then add these two products to calculate the differential value of the ship angular velocity change. Calculate sequentially according to the operation sequence specified in the virtual torque calculation method formula, and finally generate a virtual torque command.

[0035] S3. Use the torque-speed calculation method to obtain the total ship power demand and divide it into ship power distribution commands for high frequency, medium frequency, and low frequency.

[0036] S3.1. Based on the virtual torque command and the change in ship angular velocity, combined with the characteristics of the ship operation state and the fluctuation characteristics of the ship internal power network, use the torque-speed calculation method to obtain the total ship power demand.

[0037] It should be noted that the virtual torque command and the ship angular velocity change data are arranged in one-to-one correspondence according to the time sequence. Combining the characteristics of the ship operation state, such as the load, propulsion requirements, etc. in different states such as steady navigation, acceleration, deceleration, and turning, and the fluctuation characteristics of the ship internal power network, such as voltage stability, power fluctuation range, etc., perform weighted processing on the virtual torque command and the ship angular velocity change data. When the ship is in steady navigation and the power network is stable, give a larger weight to the virtual torque command and a smaller weight to the ship angular velocity change. After weighted processing, obtain the preliminary torque-speed correlation value, and then correct the correlation value according to the characteristics of the ship operation state and the fluctuation characteristics of the ship internal power network. For example, in the acceleration mode, appropriately adjust the torque-speed correlation value according to the acceleration amplitude and the power network carrying capacity. Substitute the corrected torque-speed correlation value into the torque-speed calculation method, first calculate the power component related to torque and the power component related to speed respectively, and then add these two power components to obtain the total ship power demand.

[0038] S3.2. Through the frequency domain analysis method, perform frequency domain decomposition on the total ship power demand to divide it into ship power distribution commands for high frequency, medium frequency, and low frequency power.

[0039] It should be noted that the total power demand data of the ship is subjected to fast Fourier transform to convert the total power demand from the time domain to the frequency domain, obtaining the spectrum of the total power demand. Based on the frequency range on the obtained spectrum, the boundaries of high frequency, medium frequency, and low frequency are divided. According to the set boundaries of high frequency, medium frequency, and low frequency, the frequency components belonging to the high frequency range and the corresponding amplitude information are extracted from the spectrum. The frequency components and the corresponding amplitude information in the high frequency range constitute the high frequency power part, and thus the ship power distribution instruction for the high frequency power is generated. The frequency components in the medium frequency range and the corresponding amplitude information are extracted from the spectrum to form the medium frequency power part, and then the ship power distribution instruction for the medium frequency power is generated. The frequency components falling in the low frequency range and the corresponding amplitude information are extracted from the spectrum to form the low frequency power part, thereby generating the ship power distribution instruction for the low frequency power.

[0040] S4. Output the actual power of the ship through the virtual synchronous machine model.

[0041] S4.1. Based on the historical ship power distribution instructions, decompose them through fast Fourier transform, extract the dynamic response characteristics of the ship power, use a sliding window for statistical analysis, extract statistical features, and initialize the parameters of the virtual synchronous machine model.

[0042] It should be noted that the historical ship power distribution instructions are subjected to fast Fourier transform to convert the historical ship power distribution instructions from the time domain to the frequency domain, obtaining the corresponding spectrum. In the spectrum, the boundaries of high frequency, medium frequency, and low frequency are determined based on the frequency range. According to this boundary, the frequency components and the corresponding amplitude information of the high frequency, medium frequency, and low frequency parts are respectively extracted. The frequency components and the corresponding amplitude information of the high frequency, medium frequency, and low frequency parts constitute the dynamic response characteristics of the ship power. For the extracted dynamic response characteristics of the ship power, a sliding window is set. Let the sliding window start from the starting position of the dynamic response characteristic data of the ship power, and move a fixed step length each time. Statistical analysis is performed on the dynamic response characteristic data within the window, and statistics such as mean, variance, and standard deviation are calculated to extract statistical features. According to the mapping relationship between the extracted statistical features (such as the variance in the high frequency band reflecting the power mutation intensity and the mean in the low frequency band characterizing the steady-state offset) and the stability requirements of the ship power system (such as voltage volatility ≤ 5% and frequency deviation ± 0.2 Hz), the initial parameters of the virtual synchronous machine are quantitatively derived, including parameters such as moment of inertia, damping coefficient, and synchronous reactance, to complete the initialization of the parameters of the virtual synchronous machine model.

[0043] S4.2. Based on the dynamic response characteristics and statistical features of the ship power, use the gradient descent optimization training method to train the virtual synchronous machine model and output the actual power of the ship.

[0044] It should be noted that the dynamic response characteristics and statistical characteristics of the ship power are used as input data and provided to the virtual synchronous machine model. The initial weight parameters and bias parameters of the virtual synchronous machine model are set, and the virtual synchronous machine model processes the input dynamic response characteristics and statistical characteristics of the ship power to obtain an output result, which is the ship power predicted by the virtual synchronous machine model. Calculate the error between the currently predicted ship power and the actual ship power. The error calculation uses the mean square error function; It should be explained that the expression of the mean square error function for training the virtual synchronous machine model is: ; where, is the mean square error function, is the predicted value of the ship power of the th virtual synchronous machine model for the th sample, is the measured value of the actual ship power of the th sample, is the number of training samples, is the index variable of the number of training samples; Based on the obtained mean square error function, the gradient descent optimization training method is used to update the weight parameters and bias parameters of the virtual synchronous machine model. Specifically, in the direction of the gradient of the loss function with respect to the weight parameters and bias parameters, the weight parameters and bias parameters are adjusted with the learning rate. After adjusting the parameters, the dynamic response characteristics and statistical characteristics of the ship power are input into the updated virtual synchronous machine model again. Repeat the previous steps of calculating the output result, calculating the error, and updating the parameters, and continuously iterate this process until the error passes through the number of iterations and the virtual synchronous machine model outputs the actual ship power.

[0045] S5. Real-time monitor the DC bus voltage and the power of the hydrogen fuel cell, and use the dynamic compensation power fluctuation method to regulate the operating state of the ship to obtain the voltage stability index.

[0046] S5.1. Based on the actual ship power, real-time monitor the changes in the DC bus voltage, the power of the hydrogen fuel cell, and the ship power demand, and extract the power fluctuation characteristics.

[0047] It should be noted that the correlation between the fluctuation amplitude and frequency characteristics of the DC bus voltage, the power fluctuation trend and severity of the hydrogen fuel cell, and the fluctuation range and cycle of the ship's power demand is analyzed. Compare the fluctuation conditions of the hydrogen fuel cell power and the ship's power demand when the DC bus voltage fluctuation amplitude is large in different time periods. Observe the change responses of the DC bus voltage and the ship's power demand when the hydrogen fuel cell power rises or falls rapidly. When the ship's power demand shows periodic changes, check whether there are corresponding laws for the DC bus voltage and the hydrogen fuel cell power. Compare from the starting time and duration of monitoring the fluctuations of the DC bus voltage, the hydrogen fuel cell power, and the ship's power demand, and find out the synchronization or sequence relationship between them in the time series. Based on the correlations, responses, laws, and time series relationships in multiple aspects, sort out the power fluctuation characteristics.

[0048] S5.2. Use the method of dynamically compensating power fluctuations to compensate the impact of power fluctuation characteristics on the ship's operation in real time, adjust the ship's operation state, and obtain the voltage stability index.

[0049] It should be noted that according to the sorted out power fluctuation characteristics, determine the compensation parameters required for the method of dynamically compensating power fluctuations. For the DC bus voltage fluctuation characteristics, calculate the compensation power value required to stabilize the DC bus voltage based on the fluctuation amplitude and frequency. For the hydrogen fuel cell power fluctuation characteristics, determine the corresponding compensation power adjustment amount according to the power change trend and severity. Calculate the reference value of the compensation power that meets the requirements of the ship's stable operation according to the fluctuation range and change cycle of the ship's power demand. Integrate the calculated compensation power values for the DC bus voltage, the hydrogen fuel cell power, and the ship's power demand to form a comprehensive dynamic compensation power scheme. According to the comprehensive compensation power scheme, adjust the working state of the compensation device in real time to compensate for the power fluctuations. During the compensation process, continuously monitor the changes in the DC bus voltage, the hydrogen fuel cell power, and the ship's power demand to observe whether they develop in a stable direction. Adjust the parameters of the compensation device in real time to ensure that the compensation effect meets the expectations. By monitoring the stability of the DC bus voltage, calculate whether the voltage fluctuation range shrinks, and count whether the number of voltage fluctuations within a specified time decreases. Based on the monitoring results, obtain the voltage stability index.

[0050] S6. Calculate the current optimal virtual inertia coefficient through the reinforcement learning model based on the actual power of the ship and the voltage stability index.

[0051] S6.1. Based on the historical actual power of the ship, define the state space and action space, use the historical voltage stability index to define the reward function, and construct the framework of the reinforcement learning model.

[0052] It should be noted that historical actual power data of ships is collected. By statistically analyzing the maximum value, minimum value and distribution density of the historical actual power data of ships, after removing outliers, the 5% - 95% quantiles of the power data are used as the value range. According to the power value range, different power state intervals are divided, and each interval is used as a state to define the state space. The operation action data corresponding to the historical actual power of the ship is collected, and the operation actions cover various operations such as adjusting the power output of ship equipment and changing the navigation strategy. According to the characteristics and possible values of the operation actions, the operation actions are classified, and each category of operation action is used as an action to define the action space. Through time series alignment and correlation analysis (such as Pearson coefficient, Granger causality test), the dynamic influence of the change of the actual power of the ship and the operation actions on the voltage stability index (such as voltage deviation rate, fluctuation amplitude) is quantified, and the mapping relationship of power - action - voltage stability is obtained. According to the mapping relationship of power - action - voltage stability, a reward rule is set. When an action is taken, if the voltage stability index is improved to the expected range, a higher reward is given; if the voltage stability index deteriorates, a lower reward or even a negative reward is given. Based on the defined state space, action space and the set reward rule, the proximal policy optimization algorithm is selected to build the framework of the reinforcement learning model.

[0053] S6.2. Use the historical actual power of the ship and the historical voltage stability index to train the framework of the reinforcement learning model to obtain the reinforcement learning model.

[0054] It should be noted that historical actual power data of ships and historical voltage stability index data are collected. The state space of the historical actual power data of ships is classified so that the historical actual power data of ships and the historical voltage stability index data correspond to the corresponding states. The historical voltage stability index data is associated with the corresponding operation actions and states, and samples are randomly selected from them according to a certain proportion. The selected samples are divided into multiple batches. For each batch of samples, the state space data is input into the policy network of the reinforcement learning model framework. The policy network outputs the probability distribution of selecting each action according to the input state, and an operation instruction (such as adjusting the propulsion power) is obtained through action sampling according to this probability distribution. The operation instruction (such as adjusting the propulsion power) obtained by sampling and the corresponding state are input into the value network obtained by initializing the neural network structure and using the state (such as ship power interval), execution action (such as propulsion power adjustment instruction), and immediate reward (feedback value calculated according to the voltage stability index) of the historical actual power of the ship and the historical voltage stability index. By minimizing the temporal difference error (TD error) or Monte Carlo return deviation, the weight parameters of the value network are iteratively optimized, and the value function estimation of this action in this state is calculated. According to the reward function, combined with the current state, the actions taken, and the historical voltage stability index data corresponding to the next state, the reward value is calculated. The state, action, reward value, and the next state are input into the policy network and the value network to calculate the policy loss function and the value loss function. Based on the calculated loss function values, the parameters of the policy network and the value network are updated and adjusted using an optimization algorithm. The processes of sampling, calculating the output, calculating the loss, and updating the parameters are continuously repeated until the set number of training rounds is reached or other stopping conditions are met, obtaining the reinforcement learning model.

[0055] S6.3. Input the actual power of the ship and the voltage stability index into the reinforcement learning model to calculate the current optimal virtual inertia coefficient.

[0056] It should be noted that the expression for calculating the current optimal virtual inertia coefficient is: ; where is the optimal virtual inertia coefficient, is the set of all possible actions, is the index of the action, is the estimate of the immediate value function, is the th action taken by the reinforcement learning model, is the operating state of the ship, is the cost function for executing the action, is the change in the estimate of the immediate value function, is the weight coefficient of the immediate value function, is the weight coefficient of the value change, is the weight coefficient of the cost function; It should be noted that when inputting the actual power data of the ship into the reinforcement learning model, based on the input actual power data of the ship and combined with the state space, the state corresponding to the current actual power of the ship is determined. From the probability distribution of selecting each action in the current state output by the reinforcement learning model, based on the correspondence between the action and the virtual inertia coefficient, the actions that may affect the adjustment of the virtual inertia coefficient are initially screened out. The ship voltage stability index data, the information related to the initially screened actions, and the current state information are input into the value network together. The value network calculates the estimate of the value function corresponding to the action in the current state and evaluates the influence degree of each action on the voltage stability index. According to the value function estimate results of each action output by the value network, the action with the highest value function estimate value is selected, and the virtual inertia coefficient corresponding to this action is the current optimal virtual inertia coefficient.

[0057] S7. Use the power adjustment algorithm to adaptively optimize the propulsion inertia of the ship.

[0058] S7.1. Based on the optimal virtual inertia coefficient, combined with the high-frequency, medium-frequency, and low-frequency ship power distribution commands of the ship, use the power regulation algorithm to real-time control the operating state of the ship, and optimize the inertial response characteristics and dynamic response characteristics of the ship.

[0059] It should be noted that the optimal virtual inertia coefficient and the high-frequency, medium-frequency, and low-frequency power distribution commands of the ship are input into the power regulation algorithm together. The power regulation algorithm analyzes the power adjustment requirements corresponding to the high-frequency power distribution command based on the input optimal virtual inertia coefficient and the high-frequency, medium-frequency, and low-frequency power distribution commands of the ship, and combines the optimal virtual inertia coefficient to preliminarily control the high-frequency operating state of the ship, adjusts the relevant parameters during high-frequency operation of the ship to optimize the high-frequency response characteristics. For the medium-frequency power distribution command, the power regulation algorithm implements control over the medium-frequency operating state of the ship according to its power adjustment requirements, in cooperation with the optimal virtual inertia coefficient, changes various parameters during medium-frequency operation of the ship, and improves the medium-frequency response characteristics. For the low-frequency power distribution command, the power regulation algorithm adjusts the low-frequency operating state of the ship in accordance with the power adjustment information in the command, in coordination with the optimal virtual inertia coefficient, optimizes the parameter settings during low-frequency operation of the ship, and improves the low-frequency response characteristics. During the control process, continuously monitor the changes in the inertial response characteristics and dynamic response characteristics of the ship, and fine-tune the control parameters of the power regulation algorithm based on the monitoring results to ensure that the inertial response characteristics and dynamic response characteristics of the ship are continuously optimized.

[0060] This embodiment also provides a computer device, applicable to the case of the inertial adaptive optimization control method for the electric propulsion of a hydrogen fuel cell ship, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the inertial adaptive optimization control method for the electric propulsion of a hydrogen fuel cell ship as proposed in the above embodiment.

[0061] The computer device may be a terminal, which includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner, and the wireless manner can be achieved through WIFI, carrier network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0062] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for inertial adaptive optimization control of the power propulsion of a hydrogen fuel cell ship as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic memory, flash memory, magnetic disk or optical disc.

[0063] In summary, the present invention: calculates the total power demand of the ship by using a torque-speed calculation method, divides the total power demand into high-frequency, medium-frequency, and low-frequency power distribution instructions. After frequency-domain decomposition, the dynamic output of the ship at different time scales can be controlled. It helps the ship to reasonably allocate power according to requirements in various navigation states, improve energy utilization efficiency, and make the operation of the ship more stable and efficient. By calculating the current optimal virtual inertia coefficient through a reinforcement learning model, the control strategy of the ship can be dynamically adjusted according to the actual operating conditions, with strong adaptability. Using a power regulation algorithm to adaptively optimize the propulsion inertia of the ship can enable the ship to better adapt to different working conditions and enhance the stability and safety of the ship.

[0064] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A method for inertial adaptive optimization control of a hydrogen fuel cell ship's electric propulsion, characterized in that: including Real-time collect the ship operation status data and the ship internal power network fluctuation data, perform preprocessing and extract features, calculate the virtual inertia coefficient and the virtual damping coefficient, and generate a virtual torque command; Use the torque-speed calculation method to obtain the total ship power demand and divide it into high-frequency, medium-frequency, and low-frequency ship power distribution commands; Through the virtual synchronous machine model, output the actual ship power, real-time monitor the DC bus voltage and the hydrogen fuel cell power, and use the dynamic compensation power fluctuation method to regulate the ship operation status and obtain the voltage stability index; Based on the actual ship power and the voltage stability index, calculate the current optimal virtual inertia coefficient through the reinforcement learning model, and use the power regulation algorithm to adaptively optimize the ship propulsion inertia.

2. The hydrogen fuel cell ship electric propulsion inertia adaptive optimization control method according to claim 1, characterized in that: The real-time collection of the ship operation status data and the ship internal power network fluctuation data, perform preprocessing and extract features, the specific steps are as follows Real-time collect the ship acceleration, angular velocity change, propeller thrust change, wind speed and wave height data to obtain the ship operation status data, and real-time collect the ship DC bus voltage and the hydrogen fuel cell output power to obtain the ship internal power network fluctuation data; Use the linear interpolation method to fill in the missing values, and use the threshold detection method to eliminate the outliers; Standardize to a unified range by the Z-score method, and perform feature extraction through the frequency domain analysis method to obtain the ship operation status features and the ship internal power network fluctuation features.

3. The hydrogen fuel cell ship electric propulsion inertia adaptive optimization control method according to claim 2, characterized in that: The calculation of the virtual inertia coefficient and the virtual damping coefficient, and the generation of the virtual torque command, the specific steps are as follows Through the ship propeller thrust change and the ship internal power network fluctuation data, use the frequency domain analysis method to analyze and obtain the electromagnetic torque dynamic characteristics; Based on the ship operation status features and the ship internal power network fluctuation features, calculate the virtual inertia coefficient and the virtual damping coefficient, combine the ship angular velocity change and the electromagnetic torque dynamic characteristics, and generate a virtual torque command through the virtual torque calculation method.

4. The hydrogen fuel cell ship electric propulsion inertia adaptive optimization control method according to claim 3, characterized in that: The use of the torque-speed calculation method to obtain the total ship power demand and divide it into high-frequency, medium-frequency, and low-frequency ship power distribution commands, the specific steps are as follows Based on the virtual torque command and the ship angular velocity change, combine the ship operation status features and the ship internal power network fluctuation features, and use the torque-speed calculation method to obtain the total ship power demand; Through the frequency domain analysis method, perform frequency domain decomposition on the total ship power demand, and divide it into high-frequency, medium-frequency, and low-frequency ship power distribution commands.

5. The hydrogen fuel cell ship electric propulsion inertia adaptive optimization control method according to claim 4, characterized in that: The output of the actual ship power through the virtual synchronous machine model, the specific steps are as follows Based on the historical ship power distribution commands, perform decomposition through the fast Fourier transform, extract the ship power dynamic response features, use the sliding window for statistical analysis, extract the statistical features, and initialize the parameters of the virtual synchronous machine model; Based on the ship power dynamic response features and the statistical features, use the gradient descent optimization training method to train the virtual synchronous machine model and output the actual ship power.

6. The hydrogen fuel cell ship electric propulsion inertia adaptive optimization control method according to claim 5, characterized in that: The DC bus voltage and the power of the hydrogen fuel cell are monitored in real time, and the dynamic compensation power fluctuation method is used to regulate the operation state of the ship to obtain the voltage stability index. The specific steps are as follows: Based on the actual power of the ship, the changes in the DC bus voltage, the power of the hydrogen fuel cell, and the ship's power demand are monitored in real time, and the power fluctuation characteristics are extracted. Using the dynamic compensation power fluctuation method, the influence of the power fluctuation characteristics on the ship's operation is compensated in real time, the operation state of the ship is adjusted, and the voltage stability index is obtained.

7. The inertial adaptive optimization control method for the electric propulsion of a hydrogen fuel cell ship according to claim 6, characterized in that: Based on the actual power of the ship and the voltage stability index, the current optimal virtual inertia coefficient is calculated through the reinforcement learning model. The specific steps are as follows: Based on the historical actual power of the ship, the state space and the action space are defined, and the reward function is defined using the historical voltage stability index to construct the reinforcement learning model framework. The reinforcement learning model framework is trained using the historical actual power of the ship and the historical voltage stability index to obtain the reinforcement learning model. The actual power of the ship and the voltage stability index are input into the reinforcement learning model to calculate the current optimal virtual inertia coefficient.

8. The hydrogen fuel cell ship electric propulsion inertia adaptive optimization control method according to claim 7, characterized in that: Using the power regulation algorithm to adaptively optimize the ship's propulsion inertia. The specific steps are as follows: Based on the optimal virtual inertia coefficient, combined with the ship's high-frequency, medium-frequency, and low-frequency ship power distribution commands, the power regulation algorithm is used to regulate the operation state of the ship in real time, and the inertia response characteristics and dynamic response characteristics of the ship are optimized.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, the steps of the hydrogen fuel cell ship electric propulsion inertia adaptive optimization control method described in any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the steps of the hydrogen fuel cell ship electric propulsion inertia adaptive optimization control method described in any one of claims 1 to 8 are implemented.

Images