Marine high-power container type charging station

Abstract

The invention provides a marine high-power container type charging station, which belongs to the technical field of marine charging stations, and realizes a nine-step charging control process by integrating an electric energy stability regulation module through a control chip: firstly, receiving power grid monitoring data, calculating an evaluation index, and constructing a prediction model to generate an early warning signal; adjusting the power output of the electric pile assembly according to the early warning signal; receiving data of a ship battery management system and calculating an optimal charging curve; adjusting charging parameters based on the temperature data and the thermal diffusion equation; the harmonic content is calculated in real time, and harmonic waves are suppressed through a fluctuation compensation model; monitoring a power factor and executing reactive compensation; calculating multi-path charging power distribution based on the battery optimization model; and the charging power change rate is adjusted when power grid frequency fluctuation is detected. According to the system, a multilayer convolutional neural network and a deep reinforcement learning technology are fused, and stable and efficient charging in a complex power grid environment is realized through adaptive weight adjustment.

Description

Technical Field

[0001] The present invention belongs to the technical field of marine charging stations, and more particularly, relates to a high-power containerized marine charging station. Background Art

[0002] With the development of the electrification trend of maritime navigation, marine power systems have put forward higher requirements for shore-based charging facilities. Traditional ship charging technologies mainly adopt fixed-parameter control methods, and the charging process control is achieved through preset charging curves and simple electrical parameter detection. This method can meet the basic charging requirements in a stable grid environment and is widely used in small-power ship charging scenarios. However, traditional ship charging technologies have obvious deficiencies when facing high-power charging demands. First, the quality of port power grids generally fluctuates greatly, and voltage and frequency instabilities will be directly transmitted to the charging system, resulting in a decrease in charging efficiency and an extension of charging time. Second, fixed-parameter control cannot adapt to the charging characteristic differences of different types of ship batteries, easily causing overcharging or undercharging of the batteries. In addition, the heat generated by high-power charging is difficult to effectively dissipate, increasing the potential safety hazards of the equipment. Especially in harsh seaport environments, the interaction between grid fluctuations and ship load changes forms complex power system disturbances, and traditional technologies are difficult to achieve precise control and rapid response in such complex electrical environments, and cannot solve the technical problem of low charging efficiency caused by grid fluctuations during the high-power charging process of ships. Summary of the Invention

[0003] In view of this, the present invention provides a high-power containerized marine charging station, which can solve the technical problem of low charging efficiency caused by grid fluctuations during the high-power charging process of ships in the prior art.

[0004] The present invention is implemented as follows: The control chip in the high-power containerized marine charging station of the present invention is provided with an electric energy stability adjustment module, which is used to receive grid monitoring data, calculate grid quality evaluation indicators, construct a grid prediction model to predict grid fluctuation trends, and adjust the charging power according to grid status warning signals to achieve stable charging control under grid fluctuation conditions; the control chip is also provided with a grid fluctuation compensation model and a battery charging optimization model, and realizes grid fluctuation compensation and charging strategy optimization through a multi-layer temporal convolutional neural network and a deep reinforcement learning network structure.

[0005] Among them, the power stability adjustment module is used to perform the following steps: receive the shore power system voltage data and frequency data collected by the power grid quality monitoring device, calculate the voltage deviation rate and frequency deviation rate, and establish a power grid quality evaluation index; build a power grid prediction model based on the power grid quality evaluation index, predict the power grid fluctuation trend within the next 10 seconds, and generate a power grid status warning signal; adjust the power output parameters of the high-power stack assembly according to the power grid status warning signal to achieve the adaptive matching of the charging power and the power grid status.

[0006] Among them, the power stability adjustment module is also used to perform the following steps: receive the data of the shipborne box-type power supply battery management system transmitted by the communication interface device, and calculate the optimal charging current curve; adjust the charging current and the operating parameters of the cooling system based on the system temperature data collected by the temperature monitoring device and the battery thermal diffusion equation.

[0007] Among them, the power stability adjustment module is also used to perform the following steps: calculate the system harmonic content in real time, and adjust the control strategy of the high-power stack assembly through the power grid fluctuation compensation model to suppress harmonics; monitor the power factor during the charging process, and adjust the reactive power compensation parameters of the high-power stack assembly through the elastic reactive power compensation equation.

[0008] Among them, the power stability adjustment module is also used to perform the following steps: calculate the power distribution scheme for multi-channel charging based on the shipborne box-type power supply capacity and charging demand, and in combination with the battery charging optimization model; when it is detected that the power grid frequency fluctuation exceeds the threshold, adjust the change rate of the charging power through the power grid frequency response equation.

[0009] Among them, the voltage deviation rate refers to the ratio of the difference between the actual voltage value and the rated voltage value to the rated voltage value; the frequency deviation rate refers to the ratio of the difference between the actual frequency value and the rated frequency value to the rated frequency value; the power grid quality evaluation index includes voltage deviation rate, frequency deviation rate, voltage volatility and frequency stability.

[0010] Among them, the power grid prediction model adopts the time series analysis method, and predicts the change trend of the power grid within a certain period of time in the future through the analysis of historical power grid data; the power grid status warning signal includes three levels: voltage fluctuation warning, frequency fluctuation warning and comprehensive stability warning.

[0011] Among them, when the power grid status warning signal indicates that the power grid voltage is low, the power stability adjustment module reduces the power output parameters of the high-power stack assembly; when the power grid status warning signal indicates that the power grid voltage is high, the power stability adjustment module increases the power output parameters of the high-power stack assembly; the power output parameters include output current limit value, voltage adjustment coefficient and power change rate.

[0012] Among them, the input data weights of the battery charging optimization model and the grid fluctuation compensation model are adjusted using an adaptive weight adjustment function. When the change rate of the data collected by various sensors exceeds the adjustment point, the weight adjustment is triggered; the adjustment point is obtained by calculating with a data distribution dynamic threshold function, and this function determines a reasonable threshold range based on the historical statistical characteristics of the data and the current system operating state.

[0013] Among them, the system harmonic content refers to the sum of the non-fundamental components in the current and voltage waveforms at the input and output ends of the system; when the system harmonic content exceeds the preset threshold, the power stability adjustment module outputs harmonic suppression control parameters through the grid fluctuation compensation model; the harmonic suppression control parameters are used to adjust the switching control signals of the high-power stack components.

[0014] Among them, the battery thermal diffusion equation is used to describe the physical mechanism of heat generation and transfer inside the battery during the charging process. The inputs include the internal resistance parameters of the battery pack, the optimal charging current curve, ambient temperature data, the battery heat capacity coefficient, and the battery thermal conductivity coefficient, and the outputs are the predicted values of the battery temperature field distribution and the heat accumulation rate. The elastic reactive power compensation equation is used to calculate the magnitude and phase angle of the reactive power that the system needs to compensate. The inputs include the current power factor, the load active power, the system voltage, the system frequency, and the target power factor, and the outputs are the reactive power compensation value and the phase adjustment parameter; when the current power factor is lower than 0.95, the power stability adjustment module calculates the reactive power compensation value and the phase adjustment parameter through the elastic reactive power compensation equation.

[0015] Among them, the power distribution scheme distributes the output power of each charging circuit according to the capacity of the ship container power supply connected to each charging circuit, the state of charge of the battery pack, and the charging time requirement, so as to achieve load balancing of multiple charging circuits; the load balancing means that the ratio of the output power of each charging circuit to its rated capacity is similar, avoiding overload of individual circuits.

[0016] Among them, the specific structure of the grid fluctuation compensation model is a multi-layer time series convolutional neural network structure, which includes four convolutional layers, two pooling layers, three fully connected layers, and one output layer. Among them, the size of the convolutional kernel is associated with the grid frequency period, the sampling interval of the pooling layer matches the sampling frequency of the grid quality monitoring device, and the number of neurons in the fully connected layer is adapted to the power adjustment accuracy of the high-power stack components. The specific structure of the battery charging optimization model is a deep reinforcement learning network, which consists of a policy network and a value network. The policy network adopts a three-layer long short-term memory network structure and two fully connected layers, and the value network adopts a four-layer fully connected neural network structure. The input dimension of the network matches the number of monitoring parameters, and the number of neurons in the hidden layer is associated with the number of charging circuits and the adjustment accuracy.

[0017] Through the power grid fluctuation compensation model and the battery charging optimization model, the present invention realizes the intelligent prediction and adaptive compensation of the power grid quality fluctuation. The system uses an electric energy stable regulation module to control the process of ship charging, and can dynamically adjust the charging power according to the power grid status warning signal, ensuring the safe and stable operation of the charging system under various power grid conditions. It realizes the real-time monitoring and predictive compensation of the power grid voltage deviation and frequency fluctuation, ensures the charging quality through harmonic suppression and reactive power compensation, and at the same time adjusts the charging current and cooling parameters by using the battery thermal diffusion equation, effectively solving the thermal management problem in the high-power charging process. In addition, based on the ship battery characteristics and charging requirements, the system combines the battery charging optimization model to calculate the optimal charging strategy, significantly improving the charging efficiency and the battery service life. Through the application of the multi-layer temporal convolutional neural network and the deep reinforcement learning network, the present invention successfully solves the problem of low charging efficiency caused by power grid fluctuations during the high-power charging process of ships, realizing the adaptive matching of the charging system with the power grid status and the optimized charging management of ship batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flowchart of the method of the present invention.

[0019] Figure 2 It is a schematic circuit diagram of the charging station in Embodiment 2. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] To make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention.

[0021] As Figure 1 shown, it is a flowchart of a high-power containerized charging station for ships provided by the present invention. The method includes the following steps:

[0022] A high-power containerized charging station system for ships, comprising a control chip, a transformer assembly, an AC circuit breaker assembly, a high-power stack assembly, a DC circuit breaker assembly, a high-power liquid-cooled charging socket assembly, a temperature monitoring device, a voltage monitoring device, a current monitoring device, a grid quality monitoring device, a communication interface device, a human-machine interaction device, an auxiliary transformer, a fire protection system, a lighting system, and a ventilation system. The control chip is electrically connected to the transformer assembly, the AC circuit breaker assembly, the high-power stack assembly, the DC circuit breaker assembly, the high-power liquid-cooled charging socket assembly, the temperature monitoring device, the voltage monitoring device, the current monitoring device, the grid quality monitoring device, the communication interface device, the human-machine interaction device, the auxiliary transformer, the fire protection system, the lighting system, and the ventilation system respectively. An electric energy stability adjustment module is provided in the control chip. The transformer assembly is used to convert the input AC voltage into a voltage level suitable for charging. The AC circuit breaker assembly is used to control the on-off of the AC circuit. The high-power stack assembly is used to convert AC into DC. The DC circuit breaker assembly is used to control the on-off of the DC circuit. The high-power liquid-cooled charging socket assembly is used to connect to the ship container power supply and provide a liquid-cooled heat dissipation function. The temperature monitoring device is used to monitor the working temperature of each component of the system. The voltage monitoring device is used to monitor the voltage parameters of each node of the system. The current monitoring device is used to monitor the current parameters of each loop of the system. The grid quality monitoring device is used to monitor the voltage fluctuation and frequency change of the shore power system. The communication interface device is used to exchange data with the battery management system of the ship container power supply. The human-machine interaction device is used to set charging parameters and display the system operation status. The auxiliary transformer is used to provide a working power supply for the control system and auxiliary equipment. The fire protection system is used to monitor and handle potential fire hazards inside the charging station. The lighting system is used to provide internal lighting for the charging station. The ventilation system is used to adjust the temperature and humidity inside the charging station. The sampling frequency of the temperature monitoring device is 1 time per second, the sampling frequency of the voltage monitoring device is 10 times per second, the sampling frequency of the current monitoring device is 10 times per second, and the sampling frequency of the grid quality monitoring device is 100 times per second. A grid fluctuation compensation model and a battery charging optimization model are also provided in the control chip.

[0023] The electric energy stability adjustment module is used to perform the following steps:

[0024] S01. Receive the shore power system voltage data and frequency data collected by the grid quality monitoring device, calculate the voltage deviation rate and frequency deviation rate, and establish a grid quality evaluation index;

[0025] S02. Based on the grid quality evaluation index, construct a grid prediction model, predict the grid fluctuation trend within the next 10 seconds, and generate a grid status warning signal;

[0026] S03. Adjust the power output parameters of the high-power stack assembly according to the power grid status warning signal to achieve the adaptive matching of the charging power and the power grid status;

[0027] S04. Receive the data of the ship container power supply battery management system transmitted by the communication interface device and calculate the optimal charging current curve;

[0028] S05. Based on the system temperature data collected by the temperature monitoring device and the battery thermal diffusion equation, adjust the charging current and the operating parameters of the cooling system;

[0029] S06. Calculate the harmonic content of the system in real time and adjust the control strategy of the high-power stack assembly through the power grid fluctuation compensation model to suppress harmonics;

[0030] S07. Monitor the power factor during the charging process and adjust the reactive power compensation parameters of the high-power stack assembly through the elastic reactive power compensation equation;

[0031] S08. Based on the ship container power supply capacity and the charging demand, calculate the power distribution scheme for multi-channel charging in combination with the battery charging optimization model;

[0032] S09. When it is detected that the power grid frequency fluctuation exceeds the threshold, adjust the change rate of the charging power through the power grid frequency response equation.

[0033] The voltage deviation rate refers to the ratio of the difference between the actual voltage value and the rated voltage value to the rated voltage value; the frequency deviation rate refers to the ratio of the difference between the actual frequency value and the rated frequency value to the rated frequency value; the power grid quality assessment indicators include voltage deviation rate, frequency deviation rate, voltage fluctuation rate and frequency stability.

[0034] The power grid prediction model adopts the time series analysis method and predicts the change trend of the power grid in the next period of time through the analysis of historical power grid data; the power grid status warning signal includes three levels: voltage fluctuation warning, frequency fluctuation warning and comprehensive stability warning.

[0035] When the power grid status warning signal indicates that the power grid voltage is low, the power energy stability adjustment module reduces the power output parameters of the high-power stack assembly; when the power grid status warning signal indicates that the power grid voltage is high, the power energy stability adjustment module increases the power output parameters of the high-power stack assembly; the power output parameters include output current limit value, voltage adjustment coefficient and power change rate.

[0036] The data of the ship container power battery management system includes the state of charge of the battery pack, the temperature distribution of the battery pack, and the internal resistance parameters of the battery pack; the optimal charging current curve is a curve of the change of current with time during charging calculated based on the state of charge of the battery pack, the temperature distribution of the battery pack, and the internal resistance parameters of the battery pack.

[0037] The battery thermal diffusion equation is used to describe the physical mechanism of heat generation and transfer inside the battery during charging. The inputs include the internal resistance parameters of the battery pack, the optimal charging current curve, ambient temperature data, the battery heat capacity coefficient, and the battery thermal conductivity coefficient. The outputs are the predicted values of the battery temperature field distribution and the heat accumulation rate; the ambient temperature data is sourced from the temperature monitoring device; the battery heat capacity coefficient and the battery thermal conductivity coefficient are the inherent property parameters of the battery material; the operating parameters of the cooling system include the coolant flow rate, the coolant temperature, and the rotational speed of the cooling fan.

[0038] The system harmonic content refers to the sum of the non-fundamental components in the current and voltage waveforms at the input and output ends of the system; when the system harmonic content exceeds the preset threshold, the power stability adjustment module outputs harmonic suppression control parameters through the power grid fluctuation compensation model; the harmonic suppression control parameters are used to adjust the switching control signals of the high-power stack components.

[0039] The elastic reactive power compensation equation is used to calculate the magnitude and phase angle of the reactive power that needs to be compensated by the system. The inputs include the current power factor, the load active power, the system voltage, the system frequency, and the target power factor. The outputs are the reactive power compensation value and the phase adjustment parameter; when the current power factor is lower than 0.95, the power stability adjustment module calculates the reactive power compensation value and the phase adjustment parameter through the elastic reactive power compensation equation, and sends these parameters to the high-power stack components to perform the compensation operation.

[0040] The power distribution scheme distributes the output power of each charging circuit according to the capacity of the ship container power supply connected to each charging circuit, the state of charge of the battery pack, and the charging time requirement, so as to achieve load balancing of multiple charging circuits; the load balancing means that the ratio of the output power of each charging circuit to its rated capacity is similar, avoiding overload of individual circuits.

[0041] The power grid frequency response equation is used to describe the relationship between the change of the power grid frequency and the adjustment of the load power. The inputs include the power grid frequency deviation value, the frequency change rate, the system inertia constant, the load damping coefficient, and the original charging power value. The outputs are the adjusted charging power and the power change rate limit value; the power grid frequency deviation value and the frequency change rate are sourced from the power grid quality monitoring device; the system inertia constant and the load damping coefficient are the inherent parameters of the power grid system; the original charging power value is the charging power setting value before adjustment.

[0042] The specific structure of the power grid fluctuation compensation model is a multi-layer temporal convolutional neural network structure, which includes four convolutional layers, two pooling layers, three fully connected layers and one output layer. Among them, the size of the convolutional kernel is associated with the power grid frequency period, the sampling interval of the pooling layer matches the sampling frequency of the power grid quality monitoring device, and the number of neurons in the fully connected layer is adapted to the power regulation accuracy of the high-power stack assembly; The steps for establishing the training data set of the power grid fluctuation compensation model specifically include collecting records of power grid voltage fluctuations and frequency changes in different ports over the past five years, collecting battery charging response characteristic data under different charging loads, simulating the response of the charging system under various abnormal power grid states, integrating various types of data to form a time-tagged power grid quality sequence and the corresponding optimal compensation control strategy pair, and forming a training set and a validation set after data cleaning and normalization processing; The steps for training the power grid fluctuation compensation model specifically include using the stochastic gradient descent algorithm for parameter optimization, using the cross-entropy loss function to evaluate the difference between the model output and the target compensation strategy, controlling the parameter update speed through the learning rate adaptive adjustment algorithm, introducing an early stopping mechanism to prevent overfitting, and using the model ensemble technology to improve the generalization ability of the model.

[0043] The specific structure of the battery charging optimization model is a deep reinforcement learning network, which consists of a policy network and a value network. The policy network adopts a three-layer long short-term memory network structure and two fully connected layers, and the value network adopts a four-layer fully connected neural network structure. The input dimension of the network matches the number of monitoring parameters, and the number of neurons in the hidden layer is associated with the number of charging circuits and the regulation accuracy; The steps for establishing the training data set of the battery charging optimization model specifically include collecting charge and discharge characteristic data of different types of ship container power supplies, recording the relationship between battery life and charging efficiency under various charging strategies, collecting the change law of battery internal resistance under different temperature conditions, sorting out the best configuration scheme for load balancing of each circuit in the case of multi-channel charging, and constructing a battery charging state transition probability matrix as the reinforcement learning environment model; The steps for training the battery charging optimization model specifically include using the proximal policy optimization algorithm to train the policy network, using the temporal difference learning algorithm to optimize the value network, improving the sample utilization efficiency through the experience replay mechanism, introducing an entropy regularization term to enhance the exploration ability, using the advantage function to reduce the variance of the policy gradient estimation, and setting a decaying learning rate to improve the training stability.

[0044] The input data weights of the battery charging optimization model and the power grid fluctuation compensation model are adjusted using an adaptive weight adjustment function. When the data change rate collected by various sensors exceeds the adjustment point, the weight adjustment is triggered; the adjustment point is obtained by calculating through a data distribution dynamic threshold function, which determines a reasonable threshold range based on the historical statistical characteristics of the data and the current system operating state. The inputs include the historical data average value, standard deviation, system load level, and power grid stability index, and the output is the adjustment threshold for each sensor's data; the historical data average value and the standard deviation are calculated based on the historically collected data; the system load level refers to the ratio of the current output power of the charging system to the rated power; the power grid stability index is comprehensively calculated based on the power grid quality assessment index; the adaptive weight adjustment function dynamically calculates the weight coefficient according to the data fluctuation amplitude and frequency. When the data fluctuation amplitude is larger and the fluctuation frequency is higher, the corresponding weight coefficient decreases, and vice versa, ensuring that the model assigns higher importance to stable and reliable data, improving the system's adaptability and decision-making reliability in the power grid fluctuation environment.

[0045] The following details the specific implementation of the above steps.

[0046] The specific implementation of step S01 is to receive the onshore power system voltage data and frequency data collected by the power grid quality monitoring device at a sampling frequency of 100 times per second, calculate the voltage deviation rate through the voltage deviation rate calculation module, calculate the frequency deviation rate through the frequency deviation rate calculation module, and then establish a power grid quality assessment index. In specific implementation, the voltage deviation rate calculation uses the formula of the ratio of the difference between the actual voltage value and the rated voltage value to the rated voltage value, which can effectively represent the relative deviation degree of the voltage; the frequency deviation rate calculation uses the formula of the ratio of the difference between the actual frequency value and the rated frequency value to the rated frequency value, which can effectively represent the relative deviation degree of the frequency. During the calculation process, the rated voltage value is usually set to 220V or 380V, and the rated frequency value is usually set to 50Hz or 60Hz, specifically depending on the local power grid standard. The voltage volatility is calculated by the ratio of the standard deviation to the mean of the voltage values at consecutive sampling points, reflecting the severity of the voltage fluctuation; the frequency stability is calculated by the ratio of the maximum deviation to the average value of the frequency values at consecutive sampling points, reflecting the stability of the frequency. The voltage deviation rate threshold is set to ±7%, the frequency deviation rate threshold is set to ±0.5%, the voltage volatility threshold is set to 3%, and the frequency stability threshold is set to 0.2%. When these indicators exceed the threshold, the system will trigger the corresponding warning mechanism. The role of this step is to monitor the power grid quality in real time and provide basic data support for subsequent power grid fluctuation prediction and power adjustment.

[0047] The specific implementation of step S02 is to construct a power grid prediction model based on the power grid quality assessment indicators established in step S01, use the time series analysis method to predict the power grid fluctuation trend within the next 10 seconds, and generate a power grid status warning signal. Specifically, the power grid prediction model adopts a hybrid prediction method that combines the autoregressive integrated moving average model with wavelet transform. This method can effectively capture the periodic characteristics and mutation characteristics of power grid fluctuations. The power grid quality assessment indicator data is first decomposed into different frequency components through wavelet transform, and then autoregressive integrated moving average models are established for each frequency component for prediction. Finally, the complete prediction result is obtained through wavelet reconstruction. The autoregressive term order of the model is set to 6, the moving average term order is set to 4, the integration term order is set to 1, and the wavelet decomposition layer number is set to 4. During the prediction process, the sliding window technology is adopted, the window length is set to the data of the previous 5 minutes, and the latest 1 second data is used for each prediction update. The power grid status warning signal includes three levels: voltage fluctuation warning, frequency fluctuation warning, and comprehensive stability warning. Among them, the triggering condition for the voltage fluctuation warning is that the predicted voltage deviation rate exceeds ±5% or the voltage volatility exceeds 2.5%; the triggering condition for the frequency fluctuation warning is that the predicted frequency deviation rate exceeds ±0.4% or the frequency stability exceeds 0.15%; the triggering condition for the comprehensive stability warning is that both the voltage and the frequency are close to 80% of the warning threshold at the same time. The role of this step is to predict the power grid fluctuation trend in advance, provide a decision-making basis for the early adjustment of the charging power, and reduce the adverse impact of power grid fluctuations on the charging process.

[0048] The specific implementation of step S03 is to adjust the power output parameters of the high-power stack assembly through an adaptive power adjustment algorithm according to the power grid status warning signal generated in step S02, so as to achieve the adaptive matching of the charging power and the power grid status. Specifically, when the power grid status warning signal indicates that the grid voltage is low, the power stable regulation module reduces the power output parameters of the high-power stack assembly to relieve the burden on the power grid; when the power grid status warning signal indicates that the grid voltage is high, the power stable regulation module increases the power output parameters of the high-power stack assembly to consume the excess electric energy. The adjustment of the power output parameters adopts a fuzzy control strategy to establish the mapping relationship between the power grid status and the power adjustment amount. The input fuzzy sets are the voltage deviation rate, the frequency deviation rate, and the prediction trend, and the output fuzzy sets are the adjustment ratio of the output current limit value, the voltage adjustment coefficient, and the power change rate. The fuzzy rule base contains 27 rules, covering various power grid status combinations. The defuzzification adopts the centroid method to ensure the smooth transition of the output adjustment amount. The response time of the power adjustment does not exceed 100 milliseconds to ensure that the system can quickly respond to the power grid fluctuations. The adjustment range of the output current limit value is 60% to 120% of the rated value, the adjustment range of the voltage adjustment coefficient is 0.9 to 1.1, and the adjustment range of the power change rate is 1% to 10% of the rated power per second. The function of this step is to match the charging power with the power grid status, which not only ensures the charging efficiency, but also reduces the impact on the power grid and improves the stability and reliability of the entire charging system.

[0049] The specific implementation of step S04 is to receive the data of the ship's box-type power battery management system transmitted by the communication interface device and calculate the optimal charging current curve through the battery characteristic analysis algorithm. Specifically, when implemented, the data of the ship's box-type power battery management system includes the state of charge of the battery pack, the temperature distribution of the battery pack, and the internal resistance parameters of the battery pack. These data are transmitted to the communication interface device in real time through the CAN bus or the Ethernet interface, and the sampling period is 1 second. The calculation of the optimal charging current curve adopts a multi-stage variable current rate charging strategy and is optimized in combination with the battery health state evaluation model. First, the charging stage is determined according to the state of charge of the battery pack. The state of charge of 0% - 20% is the fast charging stage, 20% - 80% is the constant current charging stage, and 80% - 100% is the trickle charging stage. Then, the upper limit of the charging current in each stage is adjusted according to the temperature distribution data of the battery pack. When the highest temperature exceeds 45°C, the upper limit of the charging current is reduced to 80% of the normal value; when the highest temperature exceeds 55°C, the upper limit of the charging current is reduced to 50% of the normal value. Finally, the battery health state is evaluated according to the internal resistance parameters of the battery pack. When the internal resistance increases by more than 30% of the initial value, the upper limit of the charging current is reduced to 90% of the normal value; when the internal resistance increases by more than 50% of the initial value, the upper limit of the charging current is reduced to 70% of the normal value. Through the comprehensive consideration of these three aspects, an optimal curve of the current changing with time during the charging process is generated, which not only meets the fast charging requirements but also protects the battery and prolongs the battery life. The function of this step is to formulate a personalized charging strategy according to the actual state of the battery, avoid overcharging or undercharging, and improve the charging efficiency and battery life.

[0050] The specific implementation of step S05 is based on the system temperature data collected by the temperature monitoring device and the battery thermal diffusion equation, and adjusts the charging current and the operating parameters of the cooling system through a thermal management optimization algorithm. In specific implementation, the temperature monitoring device collects the temperature data of each key part of the charging system at a sampling frequency of once per second, including the battery surface temperature, the charging socket temperature, the power device temperature, and the ambient temperature. The battery thermal diffusion equation is based on the heat conduction theory and describes the physical mechanism of heat generation and transfer inside the battery during the charging process. Its inputs include the internal resistance parameters of the battery pack, the optimal charging current curve, the ambient temperature data, the battery heat capacity coefficient, and the battery thermal conductivity coefficient, and the outputs are the predicted values of the battery temperature field distribution and the heat accumulation rate. The heat conduction model uses the three-dimensional finite element analysis method to divide the battery pack into multiple thermal units and calculate the heat exchange and accumulation between the units. The heat capacity coefficient is determined according to the type of battery material. Generally, for lithium-ion batteries, it is 750 - 1000 J / kg·K, and for lead-acid batteries, it is 500 - 700 J / kg·K; the thermal conductivity coefficient is determined according to the battery structure. Generally, for lithium-ion batteries, it is 1.0 - 2.5 W / m·K, and for lead-acid batteries, it is 0.6 - 1.2 W / m·K. The adjustment of the operating parameters of the cooling system uses the proportional-integral-derivative control algorithm to dynamically adjust the coolant flow rate, the coolant temperature, and the rotation speed of the cooling fan according to the deviation and change trend between the current temperature and the target temperature. The temperature control target is set to 30°C - 35°C, and the maximum allowable temperature is 60°C. When the predicted temperature exceeds 55°C, the system automatically reduces the charging current and increases the cooling intensity. The function of this step is to avoid battery overheating, ensure charging safety, and extend the service life of the device through accurate temperature prediction and control.

[0051] The specific implementation of step S06 is to calculate the harmonic content of the system in real time and adjust the control strategy of the high-power stack components through the grid fluctuation compensation model to suppress harmonics. In specific implementation, the fast Fourier transform algorithm is used to calculate the harmonic content of the system, and the current and voltage waveforms at the input and output ends of the system are spectroscopically analyzed to extract the amplitudes and phases of each harmonic component, and the total harmonic distortion rate is calculated. The sampling frequency is set to 10 kHz, the analysis window length is 20 ms, and the spectral resolution is 50 Hz. The total harmonic distortion rate threshold of the system is set to 5%. When this threshold is exceeded, harmonic suppression control is triggered. The grid fluctuation compensation model adopts a multi-layer temporal convolutional neural network structure, and according to the current grid state and harmonic distribution characteristics, it outputs harmonic suppression control parameters, including pulse width modulation waveform adjustment parameters, switching frequency adjustment values, and filter parameter adjustment values. Harmonic suppression adopts selective harmonic elimination technology, and through the combination of specific switching angles, it focuses on suppressing the 3rd, 5th, 7th, and 11th harmonics, which are the most common and influential harmonic components in the power system. At the same time, through active filtering technology, a compensation current with a phase opposite to that of the system harmonics is injected to further reduce the total harmonic content of the system. The update frequency of the control strategy is 10 times per second to ensure that the system can respond in a timely manner to grid fluctuations and load changes. The role of this step is to reduce the harmonic pollution generated by the system, improve the power quality, reduce the adverse effects on the grid and equipment, and improve the electromagnetic compatibility performance of the system.

[0052] The specific implementation of step S07 is to monitor the power factor during the charging process and adjust the reactive power compensation parameters of the high-power stack components through the elastic reactive power compensation equation. In specific implementation, the voltage-current phase difference detection method is used to monitor the power factor, and the phase angle is calculated through the time difference between the voltage zero point and the current zero point, and then the power factor is obtained. The sampling frequency is set to 10 kHz to ensure the accuracy of phase detection. The elastic reactive power compensation equation is based on the least squares method and the Lagrange multiplier method to solve the optimal value of the reactive power to be compensated under the condition of meeting the target power factor. The inputs of the equation include the current power factor, load active power, system voltage, system frequency, and target power factor, and the outputs are the reactive power compensation value and the phase adjustment parameter. The target power factor is set to be above 0.98. When the current power factor is lower than 0.95, the system will trigger reactive power compensation operations. Reactive power compensation adopts a combination of a static var compensator and an active power filter. The static var compensator is mainly used for basic reactive power compensation, and the active power filter is used for dynamic reactive power regulation and harmonic suppression. The compensation capacity range is 0 - 30% of the system rated capacity, and the adjustment accuracy is 1%. The update frequency of the compensation parameters is 5 times per second to ensure that the system can adapt to load changes. The role of this step is to improve the system power factor, reduce reactive power transmission, reduce line losses, improve the power utilization efficiency, and at the same time improve the power quality and reduce the adverse effects on the grid.

[0053] The specific implementation of step S08 is to calculate the power distribution scheme for multi-channel charging based on the capacity of the ship's container power supply and the charging demand, in combination with the battery charging optimization model. When specifically implemented, first obtain the basic information of the ship's container power supply connected to each charging circuit, including battery type, rated capacity, current state of charge, and expected charging time. Then analyze the charging characteristics and priorities of each battery through the battery charging optimization model. The priority assessment considers three factors: state of charge, remaining charging time, and battery health status. The power distribution uses the weighted proportional distribution algorithm and combines the dynamic programming method to solve the optimal distribution scheme. The weighting factors include priority weight, capacity weight, and efficiency weight, where the priority weight accounts for 50%, the capacity weight accounts for 30%, and the efficiency weight accounts for 20%. The constraint conditions of the distribution scheme include total power limit, single-channel power limit, and charging time requirement. The total power does not exceed the system rated capacity, the single-channel power does not exceed 95% of the rated power of that circuit, and at the same time, the charging time requirements of each ship's container power supply are met. The load balancing control threshold is set so that the maximum difference between the output power of each circuit and the ratio of the rated capacity does not exceed 15%. The update frequency of the distribution scheme is once per minute, or it is updated immediately when a new charging demand is connected. The role of this step is to reasonably distribute the charging power, avoid overloading of individual circuits, improve the overall charging efficiency, meet the needs of multi-user simultaneous charging, and enhance the service capacity and user experience of the charging station.

[0054] The specific implementation of step S09 is to adjust the change rate of the charging power through the grid frequency response equation when it is detected that the grid frequency fluctuation exceeds the threshold. Specifically, when implementing, first set the grid frequency fluctuation threshold to ±0.2 Hz. When the grid quality monitoring device detects that the frequency fluctuation exceeds this threshold, the frequency response regulation is triggered. The grid frequency response equation is based on the primary frequency regulation principle of the power system and describes the relationship between the grid frequency change and the load power adjustment. Its inputs include the grid frequency deviation value, the frequency change rate, the system inertia constant, the load damping coefficient, and the original charging power value, and the outputs are the adjusted charging power and the power change rate limit value. The grid frequency deviation value and the frequency change rate are obtained from the real-time measurement of the grid quality monitoring device. The system inertia constant is usually set to 5 - 8 seconds, and the load damping coefficient is usually set to 1% - 3%. The charging power adjustment adopts a proportional regulation strategy. When the grid frequency drops, the charging power is reduced; when the grid frequency rises, the charging power is increased. The adjustment coefficient is set to adjust the charging power by 5% - 10% for every 0.1 Hz deviation, and the maximum adjustment amplitude is ±30% of the original charging power. The power change rate limit value is set to not exceed 5% of the system rated power per second to avoid overly drastic power adjustment causing system instability. The response time does not exceed 200 milliseconds to ensure that the system can respond to the grid frequency fluctuation in a timely manner. The function of this step is to improve the grid stability through the active adjustment of the charging power, cooperate with the grid frequency regulation, reduce the adverse effects of frequency fluctuation on the grid and equipment, and reflect the supporting role of the charging station as an adjustable load for the grid.

[0055] The specific implementation of the hardware part is as follows: The transformer assembly adopts a dry-type transformer design with a rated capacity of 1500 kVA, a voltage transformation ratio of 10 kV / 400 V, an insulation class of F, a temperature rise limit of 100 K, an impedance voltage of 6%, and a noise level not exceeding 65 dB. The AC circuit breaker assembly uses a vacuum circuit breaker with a rated current of 2500 A and a breaking capacity of 50 kA. It adopts an electronic trip unit and has overcurrent, short-circuit, undervoltage, and overvoltage protection functions. The high-power stack assembly uses a fully controlled IGBT power module with a rated current of 1000 A, a withstand voltage of 1200 V, a switching frequency of 5 kHz to 20 kHz, and uses a water-cooled heat dissipation method with a thermal resistance of 0.01 °C / W. The DC circuit breaker assembly uses a hybrid DC circuit breaker, combining the advantages of mechanical switches and electronic switches, with a breaking time less than 2 milliseconds, a rated current of 1500 A, and a breaking capacity of 80 kA. The high-power liquid-cooled charging socket assembly uses a sealed structure with an IP67 protection level, a rated current of 1000 A, an operating voltage range of 200 V to 1000 V, a coolant flow rate of 20 L / minute, and a heat transfer efficiency greater than 90%. The temperature monitoring device uses a PT100 platinum resistance temperature sensor with a measurement range of -50 °C to 200 °C and an accuracy of ±0.5 °C. The voltage monitoring device uses a Hall voltage sensor with a measurement range of 0 to 1500 V, an accuracy of ±0.5%, and a bandwidth of 10 kHz. The current monitoring device uses a Hall current sensor with a measurement range of 0 to 2000 A, an accuracy of ±0.5%, and a bandwidth of 100 kHz. The power grid quality monitoring device uses a power quality analyzer chip with a sampling rate of 25.6 kHz and a resolution of 16 bits, and can simultaneously monitor voltage, current, frequency, harmonics, flicker, and unbalance. The communication interface device supports multiple interfaces such as CAN bus, Ethernet, RS-485, and wireless communication, with a communication rate of 1 Mbps to 100 Mbps. The human-machine interaction device uses a 10.4-inch color touch screen with a resolution of 1024×768, has anti-glare and anti-fouling treatments, and can work normally in an environment of -20 °C to 60 °C.

[0056] The specific structure of the power grid fluctuation compensation model is a multi-layer temporal convolutional neural network structure, including four convolutional layers, two pooling layers, three fully connected layers, and one output layer. The first convolutional layer has a convolutional kernel size of 5×1, 32 convolutional kernels, a stride of 1, and a ReLU activation function; the second convolutional layer has a convolutional kernel size of 3×1, 64 convolutional kernels, a stride of 1, and a ReLU activation function; the third convolutional layer has a convolutional kernel size of 3×1, 128 convolutional kernels, a stride of 1, and a ReLU activation function; the fourth convolutional layer has a convolutional kernel size of 3×1, 256 convolutional kernels, a stride of 1, and a ReLU activation function. The first pooling layer uses max pooling, with a pooling kernel size of 2×1 and a stride of 2; the second pooling layer uses max pooling, with a pooling kernel size of 2×1 and a stride of 2. The first fully connected layer has 512 neurons, a ReLU activation function, uses batch normalization and Dropout regularization, with a Dropout rate of 0.3; the second fully connected layer has 256 neurons, a ReLU activation function, uses batch normalization and Dropout regularization, with a Dropout rate of 0.3; the third fully connected layer has 128 neurons, a ReLU activation function, uses batch normalization and Dropout regularization, with a Dropout rate of 0.3. The number of neurons in the output layer matches the dimension of the control parameters, usually 10 - 20, and the activation function is Sigmoid or Tanh, selected according to the parameter range. The specific implementation of establishing the training dataset for the power grid fluctuation compensation model includes three stages: data collection, data preprocessing, and dataset division. In the data collection stage, records of power grid voltage fluctuations and frequency changes within five years are obtained from the power grid monitoring systems of different ports, with a sampling rate of 100 times per second and a total data volume exceeding 15TB; at the same time, data on the battery charging response characteristics under different charging loads are collected in a laboratory environment, with the load levels covering 10% - 120% of the rated load at a step of 10%; the response of the charging system under various abnormal power grid states is simulated through a power grid simulator, including typical scenarios such as voltage dips, voltage swells, frequency fluctuations, three-phase unbalances, and harmonic interference. In the data preprocessing stage, data cleaning is first performed to remove outliers and missing values, with the outlier determination criterion being beyond the mean ± 3 times the standard deviation; then data normalization is carried out using the Min - Max normalization method to map all features to the 0 - 1 interval; finally, feature engineering is performed to extract time-domain features and frequency-domain features. The time-domain features include mean, variance, peak value, valley value, and change rate, and the frequency-domain features include the harmonic content of each order and power spectral density. In the dataset division stage, the training set, validation set, and test set are divided in a ratio of 8:1:1 to ensure the distribution consistency of the three datasets.

[0057] The specific structure of the battery charging optimization model is a deep reinforcement learning network, which consists of a policy network and a value network. The policy network adopts a three-layer long short-term memory network structure and two-layer fully connected layers. The number of units in the first long short-term memory layer is 128, the number of units in the second long short-term memory layer is 256, and the number of units in the third long short-term memory layer is 128. Residual connections are used between the long short-term memory layers to improve the efficiency of gradient propagation. The number of neurons in the first fully connected layer is 64, the activation function is ReLU, the number of neurons in the second fully connected layer matches the dimension of the action space, usually 10 - 30, and the Softmax function is used for output to obtain the probability distribution of each action. The value network adopts a four-layer fully connected neural network structure. The number of neurons in the first layer is 256, the activation function is ReLU, and batch normalization is used. The number of neurons in the second layer is 128, the activation function is ReLU, and batch normalization is used. The number of neurons in the third layer is 64, the activation function is ReLU, and batch normalization is used. The number of neurons in the fourth layer is 1, no activation function is used, and the state value estimate is directly output. The two networks share a state encoding layer to improve the efficiency of feature representation. The specific implementation method for establishing the training data set of the battery charging optimization model includes three stages: environment modeling, reward function design, and data collection. In the environment modeling stage, first, the charge and discharge characteristic data of different types of ship box power supplies are collected, including voltage curves, internal resistance changes, and temperature change rules. Then, the relationship between battery life and charging efficiency under various charging strategies is recorded, and a battery health state assessment model is established. Next, the internal resistance change rules of the battery under different temperature conditions are collected, the temperature range is -20°C to 60°C, and the step size is 5°C. Finally, the optimal configuration scheme for load balancing of each circuit in the case of multi-channel charging is sorted out, and a battery charging state transition probability matrix is constructed as the reinforcement learning environment model. The reward function design is based on the concept of multi-objective optimization, comprehensively considering three aspects: charging efficiency, battery health maintenance, and load balancing, and the weights of each part are 0.4, 0.4, and 0.2 respectively. The charging efficiency reward is calculated based on the charging speed and power conversion efficiency, the battery health maintenance reward is calculated based on the temperature control effect and voltage stress control effect, and the load balancing reward is calculated based on the power distribution balance degree of each circuit. Data collection adopts an interactive learning method, generates training data through the interaction between the policy network and the environment model, and at the same time introduces expert demonstration data to improve the learning efficiency. The expert demonstration data comes from the highly efficient charging strategies designed manually.

[0058] Specifically, the principle of the present invention is: The core technical principle of the present invention lies in constructing a complete set of intelligent management and optimization systems for power quality, and realizing precise regulation of the ship charging process through multiple closed-loop controls. First, the system uses high-frequency sampling technology to monitor grid parameters in real time, constructs a grid quality assessment system through indicators such as voltage deviation rate and frequency deviation rate, establishes a mapping relationship between the grid state and the charging system response, and provides a data basis for subsequent intelligent decision-making.

[0059] The power grid fluctuation compensation model adopts a multi-layer time series convolutional neural network structure. Its convolutional kernel design is associated with the power grid frequency cycle, and it can effectively capture the time series characteristics and frequency domain information of power grid fluctuations. By analyzing historical power grid data and current fluctuation trends, the model predicts future power grid state changes, generates warning signals, and calculates the optimal compensation strategy. During the compensation process, the system dynamically adjusts the power output parameters of the high-power stack components according to the power grid fluctuation characteristics, including the output current limit, voltage adjustment coefficient, and power change rate, to achieve the adaptive matching of the charging power and the power grid state.

[0060] The battery charging optimization model is based on a deep reinforcement learning network. Through the collaborative action of the policy network and the value network, it transforms the complex charging decision-making problem into a learning process of state-action-reward. The model comprehensively considers the state of charge of the battery pack, temperature distribution, and internal resistance parameters, and combines the battery thermal diffusion equation to calculate the optimal charging current curve, maximizing the charging efficiency while ensuring the safety and life of the battery. At the same time, the system introduces an adaptive weight adjustment function to dynamically adjust the input weights of the model according to the data fluctuation amplitude and frequency, improving the adaptability and decision-making reliability of the system in the power grid fluctuation environment.

[0061] In addition, the present invention also establishes a power factor correction and frequency response support mechanism through the elastic reactive power compensation equation and the power grid frequency response equation, which not only improves the power quality of the charging system itself but also can provide auxiliary services for the port power grid, enhancing the stability of the overall system. The power distribution scheme for multi-channel charging ensures the efficient utilization of system resources and load balancing, avoiding overload of individual charging circuits. The organic combination of these technical principles fundamentally solves the problems of charging efficiency and safety caused by power grid fluctuations during the high-power charging of ships.

[0062] A specific embodiment 1 of the present invention is provided below. The specific implementation manners of each step in this embodiment 1 are described in detail as follows.

[0063] The specific implementation manner of step S01 is to receive the shore power system voltage data and frequency data collected by the power grid quality monitoring device at a sampling frequency of 100 times per second, calculate the voltage deviation rate through the voltage deviation rate calculation module, calculate the frequency deviation rate through the frequency deviation rate calculation module, and then establish a power grid quality evaluation index. The voltage deviation rate calculation uses the ratio formula of the difference between the actual voltage value and the rated voltage value to the rated voltage value, which is specifically expressed as follows:

[0064]

[0065] In the formula, δ V is the voltage deviation rate; V real is the actual voltage value, with the unit of volt (V); V ratedis the rated voltage value, usually set to 220V or 380V, depending on the local grid standard.

[0066] The frequency deviation rate is calculated using the ratio formula of the difference between the actual frequency value and the rated frequency value to the rated frequency value, specifically expressed as follows:

[0067]

[0068] In the formula, δ f is the frequency deviation rate; f real is the actual frequency value, with the unit of Hertz (Hz); f rated is the rated frequency value, usually set to 50Hz or 60Hz, depending on the local grid standard.

[0069] The voltage volatility is calculated by the ratio of the standard deviation to the mean of the voltage values at consecutive sampling points, specifically expressed as follows:

[0070]

[0071] In the formula, σ V is the voltage volatility; V i is the voltage value at the i-th sampling point, with the unit of Volt (V); is the arithmetic mean of the voltage values within the sampling period, with the unit of Volt (V); n is the number of sampling points.

[0072] The frequency stability is calculated by the ratio of the maximum deviation to the mean of the frequency values at consecutive sampling points, specifically expressed as follows:

[0073]

[0074] In the formula, σ f is the frequency stability; f i is the frequency value at the i-th sampling point, with the unit of Hertz (Hz); is the arithmetic mean of the frequency values within the sampling period, with the unit of Hertz (Hz).

[0075] The power grid quality assessment indicators include voltage deviation rate, frequency deviation rate, voltage volatility, and frequency stability. The voltage deviation rate threshold is set to ±7%, the frequency deviation rate threshold is set to ±0.5%, the voltage volatility threshold is set to 3%, and the frequency stability threshold is set to 0.2%. When these indicators exceed the thresholds, the system will trigger the corresponding warning mechanism. The role of this step is to monitor the power grid quality in real time and provide basic data support for subsequent power grid fluctuation prediction and power adjustment.

[0076] The specific implementation of step S02 is to construct a power grid prediction model based on the power grid quality assessment indicators established in step S01, use the time series analysis method to predict the power grid fluctuation trend within the next 10 seconds, and generate a power grid status warning signal. The power grid prediction model adopts a hybrid prediction method combining the autoregressive integrated moving average (ARIMA) model with wavelet transform. First, the power grid quality assessment indicator data is decomposed into different frequency components through wavelet transform, which is specifically expressed as follows:

[0077]

[0078] In the formula, X(t) is the time series to be decomposed, including indicators such as voltage deviation rate and frequency deviation rate; ψ j，k (t) is the wavelet basis function; φ J，k (t) is the scaling function; d j，k is the wavelet coefficient; a J，k is the scaling coefficient; j is the decomposition level, j = 1, 2,..., J, and J = 4 in this embodiment; k is the time shift parameter.

[0079] Then, an autoregressive integrated moving average model is established for each frequency component for prediction. The autoregressive integrated moving average model is specifically expressed as follows:

[0080] Φ(B)(1 - B) d X t = Θ(B)ε t ;

[0081] In the formula, X t is the sequence value at time t; B is the lag operator, that is, BX t = X t-1 ; Φ(B) = 1 - φ1B - φ2B 2 -…- φ p B p is the autoregressive polynomial, with order p = 6; Θ(B) = 1 + θ1B + θ2B 2 +…+ θ q B q is the moving average polynomial, with order q = 4; (1 - B) d is the difference term, with order d = 1; ε t is the white noise process.

[0082] Finally, the complete prediction result is obtained through wavelet reconstruction, which is specifically expressed as follows:

[0083]

[0084] In the formula, is the h-step prediction value; and They are the predicted values of wavelet coefficients and scale coefficients predicted by the ARIMA model for each frequency component; h is the prediction step length, with the unit of second. In this embodiment, h = 1, 2, …, 10.

[0085] During the prediction process, the sliding window technology is adopted. The window length is set to the data of the previous 5 minutes, and the latest 1 second data is used for each prediction update. The power grid state warning signals include three levels: voltage fluctuation warning, frequency fluctuation warning, and comprehensive stability warning. Among them, the triggering condition for the voltage fluctuation warning is that the predicted voltage deviation rate exceeds ±5% or the voltage volatility exceeds 2.5%; the triggering condition for the frequency fluctuation warning is that the predicted frequency deviation rate exceeds ±0.4% or the frequency stability exceeds 0.15%; the triggering condition for the comprehensive stability warning is that the voltage and frequency are simultaneously close to 80% of the warning threshold. The function of this step is to predict the power grid fluctuation trend in advance, provide a decision basis for the early adjustment of the charging power, and reduce the adverse impact of the power grid fluctuation on the charging process.

[0086] The specific implementation manner of step S03 is to adjust the power output parameters of the high-power stack components through an adaptive power adjustment algorithm according to the power grid state warning signal generated in step S02, so as to achieve the adaptive matching of the charging power and the power grid state. The power output parameter adjustment adopts a fuzzy control strategy to establish the mapping relationship between the power grid state and the power adjustment amount. The input variables of the fuzzy control system include the voltage deviation rate δ V , the frequency deviation rate δ f and the prediction trend τ, and the output variables include the adjustment ratio α of the output current limit value, the voltage adjustment coefficient β, and the power change rate γ.

[0087] The fuzzy set of the voltage deviation rate δ V is {Negative Big (NB), Negative Small (NS), Zero (ZO), Positive Small (PS), Positive Big (PB)}, and the membership function adopts a combination of triangle and trapezoid. The frequency deviation rate δ fThe fuzzy sets are {Negative Big (NB), Negative Small (NS), Zero (ZO), Positive Small (PS), Positive Big (PB)}, and the membership functions adopt a combination of triangular and trapezoidal shapes. The fuzzy set of the prediction trend τ is {Decrease (D), Stable (S), Increase (U)}, and the membership function adopts a triangular shape. The fuzzy set of the adjustment ratio α of the output current limit is {Decrease a Lot (DL), Decrease a Little (DS), No Change (NC), Increase a Little (IS), Increase a Lot (IL)}, and the membership functions adopt a combination of triangular and trapezoidal shapes. The fuzzy set of the voltage adjustment coefficient β is {Decrease a Lot (DL), Decrease a Little (DS), No Change (NC), Increase a Little (IS), Increase a Lot (IL)}, and the membership functions adopt a combination of triangular and trapezoidal shapes. The fuzzy set of the power change rate γ is {Very Slow (VS), Slow (S), Medium (M), Fast (F), Very Fast (VF)}, and the membership functions adopt a combination of triangular and trapezoidal shapes.

[0088] The fuzzy rule base contains 27 rules, and some of the key rules are as follows:

[0089] 1. If δ V is NB and δ f is NB and τ is D, then α is DL, β is DL, and γ is VF;

[0090] 2. If δ V is NB and δ f is ZO and τ is S, then α is DS, β is DS, and γ is M;

[0091] 3. If δ V is ZO and δ f is ZO and τ is S, then α is NC, β is NC, and γ is S.

[0092] Defuzzification is performed using the centroid method, and the formula is as follows:

[0093]

[0094] In the formula, y is the output value after defuzzification; y i is the output value corresponding to the i-th rule; μ i is the triggering strength of the i-th rule; n is the total number of rules.

[0095] The response time of power adjustment does not exceed 100 milliseconds. The adjustment range of the output current limit is 60% - 120% of the rated value. The adjustment range of the voltage adjustment coefficient is 0.9 - 1.1. The adjustment range of the power change rate is 1% - 10% of the rated power per second. The function of this step is to adjust the charging power to match the grid state, ensuring both charging efficiency and reducing the impact on the grid, and improving the stability and reliability of the entire charging system.

[0096] The specific implementation of step S04 is to receive the data of the ship's container power battery management system transmitted by the communication interface device and calculate the optimal charging current curve through the battery characteristic analysis algorithm. The data of the ship's container power battery management system includes the state of charge (SOC) of the battery pack, the temperature distribution of the battery pack, and the internal resistance parameters of the battery pack. These data are transmitted to the communication interface device in real time through the CAN bus or Ethernet interface, and the sampling period is 1 second. The calculation of the optimal charging current curve adopts a multi-stage variable current rate charging strategy and is optimized in combination with the battery health state evaluation model.

[0097] The battery health state evaluation model is based on the battery internal resistance and the capacity attenuation rate, and is specifically expressed as follows:

[0098]

[0099] In the formula, SOH is the battery health state, and the value range is 0 to 1; R init is the initial internal resistance of the battery, with the unit of ohm (Ω); R now is the current internal resistance of the battery, with the unit of ohm (Ω); C init is the initial capacity of the battery, with the unit of ampere-hour (Ah); C now is the current capacity of the battery, with the unit of ampere-hour (Ah); w1 and w2 are weight coefficients, and w1 + w2 = 1. Usually, w1 = 0.4 and w2 = 0.6.

[0100] The multi-stage variable current rate charging current curve is adjusted according to the state of charge (SOC), temperature, and health state of the battery pack, and is specifically expressed as follows:

[0101]

[0102] In the formula, I(t) is the charging current, with the unit of ampere (A); I max is the maximum charging current, usually 1C to 1.5C of the battery rated capacity; I nom is the nominal charging current, usually 0.5C to 1C of the battery rated capacity; I min is the minimum charging current, usually 0.1C to 0.3C of the battery rated capacity; f1(SOC) is the state of charge correction function; f2(T) is the temperature correction function; f3(SOH) is the health state correction function.

[0103] Among them, the temperature correction function is specifically expressed as follows:

[0104]

[0105] In the formula, T is the highest temperature of the battery pack, with the unit of degree Celsius (℃).

[0106] The health state correction function is specifically expressed as follows:

[0107]

[0108] Wherein, is the increasing ratio of the battery internal resistance.

[0109] The function of this step is to formulate a personalized charging strategy according to the actual state of the battery, avoid overcharging or undercharging, and improve the charging efficiency and battery life.

[0110] The specific implementation manner of step S05 is based on the system temperature data collected by the temperature monitoring device and the battery heat diffusion equation, and adjusts the charging current and the operating parameters of the cooling system through the thermal management optimization algorithm. The temperature monitoring device collects the temperature data of each key part of the charging system at a sampling frequency of 1 time per second, including the battery surface temperature, the charging socket temperature, the power device temperature, and the ambient temperature. The battery heat diffusion equation is based on the heat conduction theory and describes the physical mechanism of heat generation and transfer inside the battery during the charging process, and is specifically expressed as follows:

[0111]

[0112] Wherein, ρ is the battery material density, with the unit of kilogram per cubic meter (kg / m 3 ); c p is the specific heat capacity, with the unit of joule per kilogram kelvin (J / (kg·K)); T is the temperature, with the unit of kelvin (K); t is the time, with the unit of second (s); k is the heat conduction coefficient, with the unit of watt per meter kelvin (W / (m·K)); is the gradient operator; q g is the internal heat generation rate, with the unit of watt per cubic meter (W / m 3 ).

[0113] The internal heat generation rate is mainly composed of joule heat and electrochemical reaction heat, and is specifically expressed as follows:

[0114] q g = I 2 R cell / V cell + I · (U ocv - U) / V cell ;

[0115] Wherein, I is the charging current, with the unit of ampere (A); R cell is the battery internal resistance, with the unit of ohm (Ω); V cell is the battery volume, with the unit of cubic meter (m 3 ); U ocv is the open circuit voltage, with the unit of volt (V); U is the terminal voltage, with the unit of volt (V).

[0116] The adjustment of the cooling system operating parameters adopts the proportional-integral-derivative (PID) control algorithm, which is specifically expressed as follows:

[0117]

[0118] In the formula, u(t) is the control output, including the coolant flow rate, coolant temperature, and the rotational speed of the cooling fan; e(t) is the error signal, that is, the deviation between the current temperature and the target temperature; K p is the proportionality coefficient; K i is the integral coefficient; K d is the derivative coefficient. The range of the proportionality coefficient K p is 0.5 to 2.0, the range of the integral coefficient K i is 0.1 to 0.5, and the range of the derivative coefficient K d is 0.01 to 0.1.

[0119] The temperature control target is set to 30°C to 35°C, and the maximum allowable temperature is 60°C. When the predicted temperature exceeds 55°C, the system automatically reduces the charging current and increases the cooling intensity. The function of this step is to avoid overheating of the battery through precise temperature prediction and control, ensure charging safety, and extend the service life of the device.

[0120] The specific implementation method of step S06 is to calculate the system harmonic content in real time and adjust the control strategy of the high-power stack components through the grid fluctuation compensation model to suppress harmonics. The system harmonic content calculation uses the fast Fourier transform (FFT) algorithm to perform spectral analysis on the current and voltage waveforms at the input and output ends of the system, extract the amplitudes and phases of each harmonic component, and calculate the total harmonic distortion rate. The total harmonic distortion rate (THD) is specifically expressed as follows:

[0121]

[0122] In the formula, THD is the total harmonic distortion rate; I h is the effective value of the hth harmonic component; I1 is the effective value of the fundamental wave component; H is the highest harmonic order considered, usually taking H = 25.

[0123] The sampling frequency is set to 10 kHz, the analysis window length is 20 ms, and the spectral resolution is 50 Hz. The total harmonic distortion rate threshold of the system is set to 5%, and when this threshold is exceeded, harmonic suppression control is triggered. Harmonic suppression uses the selective harmonic elimination (SHE) technology, and through a specific combination of switching angles, it focuses on suppressing the 3rd, 5th, 7th, and 11th harmonics. The formula for solving the switching angles of selective harmonic elimination is as follows:

[0124]

[0125] In the formula, V dcis the DC bus voltage; N is the number of pulses per cycle; α i is the i-th switching angle; n is the harmonic order; M is the modulation index, where 0 < M < 1.

[0126] Meanwhile, through the active filtering technology, a compensation current with a phase opposite to that of the system harmonics is injected to further reduce the total harmonic content of the system. The calculation formula for the compensation current is as follows:

[0127]

[0128] In the formula, i c (t) is the compensation current; I h is the amplitude of the h-th harmonic component; ω is the fundamental angular frequency; φ h is the phase of the h-th harmonic component.

[0129] The update frequency of the control strategy is 10 times per second to ensure that the system can respond promptly to grid fluctuations and load changes. The role of this step is to reduce the harmonic pollution generated by the system, improve the power quality, reduce the adverse effects on the power grid and equipment, and enhance the electromagnetic compatibility performance of the system.

[0130] The specific implementation of step S07 is to monitor the power factor during the charging process and adjust the reactive power compensation parameters of the high-power stack components through the elastic reactive power compensation equation. The power factor monitoring uses the voltage-current phase difference detection method, calculates the phase angle through the time difference between the voltage zero point and the current zero point, and then obtains the power factor. The calculation formula for the power factor is as follows:

[0131] PF = cos(θ) = cos(2πfΔt);

[0132] In the formula, PF is the power factor; θ is the phase angle between the voltage and the current, in radians (rad); f is the fundamental frequency of the system, in hertz (Hz); Δt is the time difference between the voltage zero point and the current zero point, in seconds (s).

[0133] The sampling frequency is set to 10 kHz to ensure the accuracy of phase detection. The elastic reactive power compensation equation is based on the least squares method and the Lagrange multiplier method to solve the optimal value of the reactive power to be compensated under the condition of meeting the target power factor. The elastic reactive power compensation equation is specifically expressed as follows:

[0134]

[0135]

[0136] In the formula, Q c is the reactive power compensation value, in vars (var); Q c0is the initial reactive power compensation value; P is the active power, in watts (W); Q is the reactive power, in vars (var); PF target is the target power factor, usually set to be above 0.98.

[0137] The calculation formula for the reactive power compensation value is:

[0138] Q c = Q - P · tan(arccos(PF target ));

[0139] The calculation formula for the phase adjustment parameter is:

[0140]

[0141] In the formula, φ is the phase adjustment parameter, in radians (rad).

[0142] The target power factor is set to be above 0.98. When the current power factor is lower than 0.95, the system will trigger the reactive power compensation operation. The reactive power compensation adopts a combination of a static var compensator and an active power filter. The compensation capacity range is 0 - 30% of the system rated capacity, and the adjustment accuracy is 1%. The update frequency of the compensation parameters is 5 times per second to ensure that the system can adapt to load changes. The function of this step is to improve the system power factor, reduce the reactive power transmission, reduce the line loss, improve the power utilization efficiency, and at the same time improve the power quality and reduce the adverse impact on the power grid.

[0143] The specific implementation method of step S08 is to calculate the power distribution scheme for multi - path charging based on the ship's container power capacity and charging demand, combined with the battery charging optimization model. First, obtain the basic information of the ship's container power sources connected to each charging circuit, including battery type, rated capacity, current state of charge, and expected charging time. The power distribution uses a weighted proportional distribution algorithm and combines the dynamic programming method to solve the optimal distribution scheme. The optimization objective function for power distribution is:

[0144]

[0145] In the formula, m is the number of charging circuits; p i is the priority of the i - th circuit; P i is the charging power allocated to the i - th circuit, in watts (W); C i is the rated capacity of the i - th circuit, in watts (W); η i is the charging efficiency of the i - th circuit; w1, w2, and w3 are weight coefficients, and w1 + w2 + w3 = 1, where w1 = 0.5, w2 = 0.3, and w3 = 0.2.

[0146] The constraint conditions are:

[0147]

[0148] 0 ≤ P i ≤ 0.95·C i ;

[0149]

[0150] Wherein, P total is the total available power of the system, in watts (W); SOC i is the current state of charge of the marine container power supply connected to the i-th circuit; E i is the rated capacity of the marine container power supply connected to the i-th circuit, in watt-hours (Wh); T i is the expected charging completion time of the i-th circuit, in hours (h).

[0151] The load balancing control threshold is set such that the maximum difference in the ratio of the output power to the rated capacity of each circuit does not exceed 15%, that is:

[0152]

[0153] The update frequency of the allocation scheme is once per minute, or it is updated immediately when a new charging demand is connected. The function of this step is to reasonably allocate the charging power, avoid overloading of individual circuits, improve the overall charging efficiency, meet the needs of multiple users charging simultaneously, and enhance the service capacity and user experience of the charging station.

[0154] The specific implementation of step S09 is that when it is detected that the grid frequency fluctuation exceeds the threshold, the change rate of the charging power is adjusted through the grid frequency response equation. First, the grid frequency fluctuation threshold is set to ±0.2 Hz. When the grid quality monitoring device detects that the frequency fluctuation exceeds this threshold, the frequency response regulation is triggered. The grid frequency response equation is based on the principle of primary frequency regulation of the power system and describes the relationship between the grid frequency change and the load power adjustment, which is specifically expressed as follows:

[0155]

[0156] Wherein, ΔP is the power adjustment amount, in watts (W); Δf is the frequency deviation value, in hertz (Hz); is the frequency change rate, in hertz per second (Hz / s); K f is the static frequency regulation coefficient, in watts per hertz (W / Hz); K d is the dynamic frequency regulation coefficient, in watt-seconds per hertz (W·s / Hz).

[0157] The system inertia constant is usually set to 5 - 8 seconds, and the load damping coefficient is usually set to 1% - 3%. The calculation formula for the adjusted charging power is:

[0158] P new = P old + ΔP;

[0159] The calculation formula for the power change rate limit value is:

[0160]

[0161] In the formula, P new is the adjusted charging power, with the unit of watt (W); P old is the charging power before adjustment, with the unit of watt (W); is the power change rate, with the unit of watt per second (W / s); P rated is the system rated power, with the unit of watt (W); T resp is the response time, with the unit of second (s), usually set to 0.2 seconds.

[0162] The charging power adjustment adopts a proportional regulation strategy. When the grid frequency drops, the charging power is reduced; when the grid frequency rises, the charging power is increased. The adjustment coefficient is set to adjust the charging power by 5% - 10% for every 0.1 Hz deviation, and the maximum adjustment amplitude is ±30% of the original charging power. The power change rate limit value is set to not exceed 5% of the system rated power per second to avoid overly drastic power adjustment causing system instability. The response time does not exceed 200 milliseconds to ensure that the system can respond to grid frequency fluctuations in a timely manner. The function of this step is to improve the stability of the power grid by actively adjusting the charging power, coordinating with the grid frequency regulation, reducing the adverse effects of frequency fluctuations on the power grid and equipment, and reflecting the supporting role of the charging station as an adjustable load on the power grid.

[0163] The specific structure of the grid fluctuation compensation model is a multi-layer temporal convolutional neural network structure, including four convolutional layers, two pooling layers, three fully connected layers, and one output layer. The mathematical expression of the convolutional layer is:

[0164] Z l = f(W l * Z l-1 + b l );

[0165] In the formula, Z l is the output feature map of the l-th layer; W l is the convolutional kernel weight matrix of the l-th layer; Z l-1 is the output feature map of the (l - 1)-th layer; b l is the bias vector of the l-th layer; * represents the convolutional operation; f is the activation function, and in this embodiment, the ReLU function is adopted, that is, f(x) = max(0, x).

[0166] The convolution kernel size of the first convolutional layer is 5×1, the number of convolution kernels is 32, and the stride is 1; the convolution kernel size of the second convolutional layer is 3×1, the number of convolution kernels is 64, and the stride is 1; the convolution kernel size of the third convolutional layer is 3×1, the number of convolution kernels is 128, and the stride is 1; the convolution kernel size of the fourth convolutional layer is 3×1, the number of convolution kernels is 256, and the stride is 1. The pooling layer uses max pooling, and the mathematical expression of the pooling operation is:

[0167]

[0168] In the formula, is the output of the l-th pooling layer at position (i, j); R i，j is the pooling window area centered at position (i, j); is the output of the (l-1)-th layer at position (m, n).

[0169] The pooling kernel size of the first pooling layer is 2×1, and the stride is 2; the pooling kernel size of the second pooling layer is 2×1, and the stride is 2. The mathematical expression of the fully connected layer is:

[0170] Z l = f(W l Z l-1 + b l );

[0171] In the formula, Z l is the output vector of the l-th layer; W l is the weight matrix of the l-th layer; Z l-1 is the output vector of the (l-1)-th layer; b l is the bias vector of the l-th layer; f is the activation function.

[0172] The number of neurons in the first fully connected layer is 512, the activation function is ReLU, batch normalization and Dropout regularization are used, and the Dropout rate is 0.3; the number of neurons in the second fully connected layer is 256, the activation function is ReLU, batch normalization and Dropout regularization are used, and the Dropout rate is 0.3; the number of neurons in the third fully connected layer is 128, the activation function is ReLU, batch normalization and Dropout regularization are used, and the Dropout rate is 0.3. The number of neurons in the output layer matches the dimension of the control parameter, usually 10 - 20, and the activation function is Sigmoid or Tanh, which is selected according to the parameter range.

[0173] The model training uses the stochastic gradient descent algorithm for parameter optimization, and the loss function is the cross-entropy loss function, which is specifically expressed as follows:

[0174]

[0175] Where L is the value of the loss function; N is the number of samples; C is the number of classes; y i，j is the true label that sample i belongs to class j; p i，j is the probability that the model predicts sample i belongs to class j.

[0176] The learning rate adopts an adaptive adjustment algorithm, which is specifically expressed as follows:

[0177]

[0178] Where η t is the learning rate at the t-th step; η0 is the initial learning rate, usually set to 0.01; α is the decay rate, usually set to 0.0001; t is the number of iteration steps.

[0179] The specific structure of the battery charging optimization model is a deep reinforcement learning network, which consists of a policy network and a value network. The policy network adopts a three-layer long short-term memory (LSTM) network structure and two-layer fully connected layers. The mathematical expression of the long short-term memory unit is:

[0180] f t = σ(W f · [h t-1 , x t + b f );

[0181] i t = σ(W i · [h t-1 , x t + b i );

[0182]

[0183] o t = σ(W o · [h t-1 , x t + b o );

[0184] h t = o t * tanh(C t );

[0185] Where f t is the forget gate vector; i t is the input gate vector; is the candidate memory cell vector; C t is the memory cell vector; o t is the output gate vector; h t is the hidden state vector; x t is the input vector at time step t; Wf , W i , W C and W o are weight matrices; b f , b i , b C and b o are bias vectors; σ is the Sigmoid activation function; tanh is the hyperbolic tangent activation function; * represents element-wise multiplication.

[0186] The number of units in the first long short-term memory layer is 128, the number of units in the second long short-term memory layer is 256, and the number of units in the third long short-term memory layer is 128. The goal of the policy network is to maximize the expected return, and the optimization objective is:

[0187]

[0188] where J(θ) is the performance metric of the policy network parameters θ; π θ is the policy function with parameters θ; denotes the expectation under the policy π θ ; γ is the discount factor, with a value range of 0 to 1, usually set to 0.95 to 0.99; r t is the reward at time step t; T is the episode length.

[0189] The value network adopts a four-layer fully connected neural network structure, and the goal is to estimate the state value function. The optimization objective is:

[0190]

[0191] where L(φ) is the loss function of the value network parameters φ; V φ (s t ) is the estimate of the state s t by the value network with parameters φ; V target is the target value, usually using the temporal difference (TD) target, i.e., V target = r t + γV φ (s t+1 ).

[0192] During the training process, the proximal policy optimization (PPO) algorithm is used to train the policy network, and its objective function is:

[0193]

[0194] where L CLIP (θ) is the clipped objective function; r t (θ) is the probability ratio, i.e., A tis the advantage function; ∈ is the clipping parameter, usually set to 0.1 - 0.3.

[0195] The advantage function is calculated by Generalized Advantage Estimation (GAE):

[0196]

[0197] In the formula, λ is the GAE parameter, with a value range of 0 - 1, usually set to 0.9 - 0.95; δ t is the TD error, i.e., δ t = r t + γV φ (s t+1 ) - V φ (s t ).

[0198] The mathematical expression of the adaptive weight adjustment function is:

[0199]

[0200] In the formula, w i is the weight of the i-th sensor data; v i is the change amplitude of the i-th sensor data, and the calculation formula is where x i is the current value, is the historical average value, σ i is the standard deviation; f i is the change frequency of the i-th sensor data, and the calculation formula is where n i is the number of data changes, N is the total number of sampling times; β is the weight adjustment coefficient, usually set to 0.5 - 2.0.

[0201] The adjustment point is obtained by calculating through the data distribution dynamic threshold function, and the mathematical expression of this function is:

[0202]

[0203] In the formula, T i is the adjustment threshold of the i-th sensor data; k is the basic threshold coefficient, usually set to 2.0 - 3.0; α is the load coefficient, usually set to 0.1 - 0.5; β is the stability coefficient, usually set to 0.2 - 0.8; L is the system load level, i.e., where P out is the current output power, P rated is the rated power; S is the power grid stability index, and the calculation formula is where δ V，max 、δ f，max 、σ V，max and σf，max They are the maximum allowable values of voltage deviation rate, frequency deviation rate, voltage fluctuation rate, and frequency stability respectively.

[0204] The specific implementation of the hardware part is as follows: The transformer assembly adopts a dry-type transformer design with a rated capacity of 1500 kVA, a voltage transformation ratio of 10 kV / 400 V, an insulation class of F, a temperature rise limit of 100 K, an impedance voltage of 6%, and a noise level not exceeding 65 dB. The AC circuit breaker assembly uses a vacuum circuit breaker with a rated current of 2500 A and a breaking capacity of 50 kA. It adopts an electronic release and has overcurrent, short-circuit, undervoltage, and overvoltage protection functions. The high-power stack assembly uses a fully controlled IGBT power module with a rated current of 1000 A, a withstand voltage of 1200 V, a switching frequency of 5 kHz to 20 kHz, and uses a water-cooled heat dissipation method with a thermal resistance of 0.01 °C / W. The DC circuit breaker assembly uses a hybrid DC circuit breaker, combining the advantages of mechanical switches and electronic switches, with a breaking time less than 2 milliseconds, a rated current of 1500 A, and a breaking capacity of 80 kA. The high-power liquid-cooled charging socket assembly uses a sealed structure with a protection level of IP67, a rated current of 1000 A, a working voltage range of 200 V to 1000 V, a coolant flow rate of 20 L / minute, and a heat exchange efficiency greater than 90%.

[0205] The temperature monitoring device uses a PT100 platinum resistance temperature sensor with a measurement range of -50 °C to 200 °C and an accuracy of ±0.5 °C. The voltage monitoring device uses a Hall voltage sensor with a measurement range of 0 to 1500 V, an accuracy of ±0.5%, and a bandwidth of 10 kHz. The current monitoring device uses a Hall current sensor with a measurement range of 0 to 2000 A, an accuracy of ±0.5%, and a bandwidth of 100 kHz. The power grid quality monitoring device uses a power quality analyzer chip with a sampling rate of 25.6 kHz, a resolution of 16 bits, and can simultaneously monitor voltage, current, frequency, harmonics, flicker, and unbalance. The communication interface device supports multiple interfaces such as CAN bus, Ethernet, RS-485, and wireless communication, with a communication rate of 1 Mbps to 100 Mbps. The human-machine interaction device uses a 10.4-inch color touch screen with a resolution of 1024×768, has anti-glare and anti-fouling treatments, and can work normally in an environment of -20 °C to 60 °C.

[0206] Optionally, the power quality analyzer chip can also use an electric metering chip such as ATT7022B, etc.

[0207] To better understand and implement the present invention, the following provides Example 2 of a specific application scenario of the present invention: Researchers implemented a 5MW marine high-power containerized charging station system for the rapid charging of large container ships. The system is designed using the standard 40-foot container specification and can be moved between ports by a tractor to meet the charging needs of different berths. As Figure 2 shown, the entire power input X1 of the charging station is connected to the AC660V shore power system through a multi-interface plug-and-play method. The charging station is divided into two parts: the charging branch and the auxiliary branch. The charging branch consists of 5 charging circuits, and each branch includes a transformer, an AC circuit breaker, a high-power stack, a DC circuit breaker, and a high-power liquid-cooled socket, etc.; the auxiliary branch includes an AC circuit breaker F2, an auxiliary transformer T1, a control system (V6 + U1), a fire protection system U2, a lighting system U3, and a ventilation system U4. This charging station can be installed either fixedly or movably. If it is installed movably, it can be moved beside the shore power interface, docked with the shore power system through a high-power plug. After connecting the cables, connect to the on-board box-type power supply through the X2 - X6 interfaces. After the cable connection is completed, start charging by setting the main parameters such as the input power of the charging station, the charging power of each circuit, and the battery capacity of the box-type power supply of each circuit through the human-machine interaction module of the control system. The system adjusts the charging output in real time according to the data requested by the battery management system inside the box-type power supply; monitors the device data in real time (such as temperature, voltage, current, etc.) to ensure the safe and reliable operation of the system.

[0208] The main technical parameters of this system are shown in Table 1:

[0209] Table 1 Main Technical Parameters of Marine High-Power Containerized Charging Station System

[0210]

[0211]

[0212] During the use of the system, researchers collected the charging current and battery pack temperature data of each circuit under the condition that the 5 charging circuits were working simultaneously, as shown in Table 2:

[0213] Table 2 Parameter Records of Each Charging Circuit during Multi-Circuit Charging

[0214]

[0215] During the actual operation process, researchers recorded the fluctuations of the shore power system and the system response data during the charging process, as shown in Table 3:

[0216] Table 3 Grid Fluctuation Conditions and System Response Records

[0217]

[0218] During the test, when the grid voltage deviation rate reaches 3.5% at the 15 - minute mark, the power stability regulation module calculates and predicts through the grid fluctuation compensation model that the grid voltage may continue to rise within the next 10 seconds, triggering a voltage fluctuation warning. The system automatically reduces the total charging power by 5%, and the response time is 98 milliseconds. When the grid voltage deviation rate reaches 6.2% and the frequency deviation rate reaches 0.30% at the 45 - minute mark, the system determines that the grid stability is poor and automatically reduces the total charging power by 12%, with the response time shortened to 76 milliseconds. Throughout the charging process, the power adjustment of the system fully complies with the requirements indicated by the grid status warning signal, and the maximum response time does not exceed 100 milliseconds.

[0219] In terms of harmonic control, the researchers used a grid quality analyzer to measure the harmonic content at the input and output ends of the system in detail, as shown in Table 4:

[0220] Table 4 Comparison data of the system's harmonic content

[0221]

[0222]

[0223] From the data in Table 4, it can be seen that the selective harmonic elimination technology and active filtering technology adopted by the system effectively suppress each harmonic, reducing the total harmonic distortion rate of the system output from 6.94% before compensation to 1.45%, which is lower than the system - designed 3% index requirement, proving that the system has excellent power quality control capabilities.

[0224] In terms of temperature control, the researchers recorded the temperature change during the highest load (loop 3, power 980 kW) in the charging process, and the results are shown in Table 5:

[0225] Table 5 Temperature control data during high - load charging

[0226]

[0227] According to Table 5, it can be seen that as the charging process progresses, both the battery surface temperature and the predicted internal temperature increase. The system predicts the internal temperature through the battery heat diffusion equation and dynamically adjusts the coolant flow rate, coolant temperature, and the rotational speed of the cooling fan based on the prediction results to ensure that the battery internal temperature is always controlled within a safe range, with the highest not exceeding 45.2°C, far lower than the system - set 60°C safety threshold.

[0228] During the system commissioning phase, the researchers collected the battery management system data of the ship - mounted box - type power supply through the CAN bus and generated an optimal charging curve using the battery charging optimization model, as shown in Table 6:

[0229] Table 6 Optimal Charging Curve Data Based on BMS Data

[0230]

[0231] The charging curve in Table 6 fully considers the battery health state and temperature influence, and adopts a multi-stage variable current rate charging strategy, which not only ensures the charging speed but also extends the battery service life. During the actual charging process, this curve can be dynamically adjusted according to the real-time temperature and internal resistance to further optimize the charging effect.

[0232] Traditional ship charging systems use a fixed power charging method, lacking the ability to sense and respond to grid fluctuations, resulting in poor system stability when the grid fluctuates greatly; at the same time, the traditional system is also relatively simple in harmonic suppression and temperature control, only using passive filtering and fixed-flow cooling methods, with low charging efficiency and potential safety hazards. The present invention realizes the prediction and adaptive adjustment of grid fluctuations by introducing a grid fluctuation compensation model and a battery charging optimization model, and adopting advanced temporal convolutional neural network and deep reinforcement learning technologies. The system response time is shortened from the traditional 500 milliseconds to no more than 100 milliseconds; through selective harmonic elimination technology and active filtering technology, the harmonic content of the system is reduced by about 7.5%; by adopting accurate temperature prediction and dynamic cooling control based on the battery heat diffusion equation, the charging efficiency is increased by about 8%, and at the same time, the battery service life is significantly extended, providing a safer, more efficient and intelligent charging service for ship container power supplies.

[0233] It should be noted that the detailed explanations of the variables involved in the present invention are shown in Tables 7, 8, and 9 below.

[0234] Table 7 Variable Explanation Table (Part 1)

[0235]

[0236] Table 8 Variable Explanation Table (Part 2)

[0237]

[0238]

[0239] Table 9 Variable Explanation Table (Part 3)

[0240]

[0241]

[0242] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention.

Claims

1. A high-power containerized charging station for ships, comprising a control chip, a transformer assembly, an AC circuit breaker assembly, a high-power stack assembly, a DC circuit breaker assembly, a high-power liquid-cooled charging socket assembly, a monitoring device, and an auxiliary system, characterized in that The control chip is provided with a power stability adjustment module, which is used to receive grid monitoring data to calculate grid quality evaluation indicators, construct a grid prediction model to predict the grid fluctuation trend, and adjust the charging power according to the grid status warning signal to achieve stable charging control under grid fluctuations; the control chip is also provided with a grid fluctuation compensation model and a battery charging optimization model, which realize grid fluctuation compensation and charging strategy optimization through a multi-layer time series convolutional neural network and a deep reinforcement learning network structure.

2. The charging station according to claim 1, wherein The power stability adjustment module is used to perform the following steps: receive the shore power system voltage data and frequency data collected by the grid quality monitoring device, calculate the voltage deviation rate and the frequency deviation rate, and establish grid quality evaluation indicators; Based on the grid quality evaluation indicators, construct a grid prediction model to predict the grid fluctuation trend within the next 10 seconds, and generate a grid status warning signal; adjust the power output parameters of the high-power stack component according to the grid status warning signal to achieve an adaptive match between the charging power and the grid status.

3. The charging station according to claim 2, characterized in that, The power stability adjustment module is also used to perform the following steps: receive the data of the shipborne box-type power battery management system transmitted by the communication interface device, and calculate the optimal charging current curve; based on the system temperature data collected by the temperature monitoring device and the battery thermal diffusion equation, adjust the charging current and the operating parameters of the cooling system.

4. The charging station according to claim 3, characterized in that, The power stability adjustment module is also used to perform the following steps: calculate the system harmonic content in real time, and adjust the control strategy of the high-power stack component through the grid fluctuation compensation model to suppress harmonics; monitor the power factor during the charging process, and adjust the reactive power compensation parameters of the high-power stack component through the flexible reactive power compensation equation.

5. The charging station according to claim 4, characterized in that, The power stability adjustment module is also used to perform the following steps: based on the shipborne box-type power capacity and charging demand, calculate the power distribution scheme for multi-channel charging in combination with the battery charging optimization model; when it is detected that the grid frequency fluctuation exceeds the threshold, adjust the change rate of the charging power through the grid frequency response equation.

6. The charging station according to claim 5, wherein The voltage deviation rate refers to the ratio of the difference between the actual voltage value and the rated voltage value to the rated voltage value; the frequency deviation rate refers to the ratio of the difference between the actual frequency value and the rated frequency value to the rated frequency value; the grid quality evaluation indicators include the voltage deviation rate, the frequency deviation rate, the voltage volatility and the frequency stability.

7. The charging station according to claim 6, characterized in that, The grid prediction model adopts the time series analysis method, and predicts the change trend of the grid within a certain period of time in the future through the analysis of historical grid data; the grid status warning signal includes three levels: voltage fluctuation warning, frequency fluctuation warning and comprehensive stability warning.

8. The charging station according to claim 7, wherein, When the grid status warning signal indicates that the grid voltage is low, the power stability adjustment module reduces the power output parameters of the high-power stack component; when the grid status warning signal indicates that the grid voltage is high, the power stability adjustment module increases the power output parameters of the high-power stack component; the power output parameters include the output current limit value, the voltage adjustment coefficient and the power change rate.

9. The charging station according to claim 8, characterized in that, The input data weights of the battery charging optimization model and the grid fluctuation compensation model are adjusted using an adaptive weight adjustment function. When the change rate of the data collected by various sensors exceeds the adjustment point, the weight adjustment is triggered; the adjustment point is obtained by calculating through a data distribution dynamic threshold function, and this function determines a reasonable threshold range based on the historical statistical characteristics of the data and the current system operating state.

10. The charging station according to claim 9, characterized in that, The harmonic content of the system refers to the sum of the non-fundamental components in the current and voltage waveforms at the input and output ends of the system; when the harmonic content of the system exceeds the preset threshold, the power stability adjustment module outputs harmonic suppression control parameters through the grid fluctuation compensation model; the harmonic suppression control parameters are used to adjust the switching control signals of the high-power stack components.

Images