A wave energy power supply system intelligent switching method and system based on double SOC dynamic threshold

Abstract

This application discloses a smart switching method and system for wave energy power supply systems based on dual SOC dynamic thresholds, relating to the field of wave energy power generation and storage systems. The method includes: obtaining a predicted power generation sequence based on real-time output data from the wave energy generation unit; acquiring and calculating a comprehensive system state index based on real-time parameter data of the energy storage battery pack, current operating condition data, and a heterogeneous battery coupling aging model; inputting the low-frequency power component, high-frequency power component, predicted load power sequence, and comprehensive system state index into a hierarchical deep reinforcement learning decision engine to obtain an instantaneous power command for the energy storage battery pack; and generating a collaborative control command based on the instantaneous power command and adjusting the parameters of the energy storage converter and MPPT unit. This application improves the energy efficiency and stability of the wave energy system, enabling more efficient utilization of wave energy power generation resources, reducing energy waste, and significantly lowering system operating costs.

Description

Technical Field

[0001] This application relates to the field of wave energy power generation and energy storage system control technology, and in particular to a smart switching method and system for wave energy power supply system based on dual SOC dynamic thresholds. Background Technology

[0002] In the power generation field, wave energy, as a renewable energy source, is clean and sustainable. However, its power output fluctuates greatly, making it difficult to directly meet load demands. Current wave energy power generation systems typically employ a single energy storage solution, which struggles to cope with the complex demands of power fluctuations and load variations. Furthermore, existing control methods largely rely on traditional predictive models or simple threshold switching, lacking synergistic optimization for heterogeneous energy storage, such as energy-type and power-type energy storage, resulting in room for improvement in system efficiency and lifespan. For example, existing systems are prone to excessive power fluctuations under extreme sea conditions, leading to low system efficiency or equipment damage. Simultaneously, the lifespan and aging issues of energy storage batteries have not been effectively addressed, affecting the long-term stability and economic viability of the system. Moreover, existing technologies exhibit lag in power prediction and dynamic control, making it difficult to handle the instantaneous fluctuations in wave energy generation, resulting in insufficient energy utilization. Therefore, there is an urgent need for a novel control method that can effectively integrate multi-source information, optimize energy distribution, and improve system stability to overcome the current technical bottlenecks in energy efficiency, stability, and lifespan of wave energy power generation systems. Summary of the Invention

[0003] The purpose of this application is to address at least one of the aforementioned technical deficiencies.

[0004] On one hand, embodiments of this application provide a method for intelligent switching of wave energy power supply systems based on dual SOC dynamic thresholds, the method comprising: The real-time output data of the wave energy generation unit is obtained. Based on the real-time output data, the predicted power generation sequence is obtained through a generative adversarial network with integrated physical constraints. The predicted power generation sequence is then separated into low-frequency power components and high-frequency power components through a variational mode decomposition method with adaptive parameter adjustment. The real-time output data includes output voltage and output current. The system acquires real-time parameter data and current operating condition data corresponding to the energy storage battery pack, and calculates the comprehensive system status index based on the real-time parameter data, current operating condition data and heterogeneous battery coupling aging model. The real-time parameter data includes the SOC (State of Charge) of the energy storage battery pack and the SOH (State of Health) of the power storage battery pack. The predicted load power sequence is obtained, and the low-frequency power component, high-frequency power component, predicted load power sequence and system comprehensive state index are input into the hierarchical deep reinforcement learning decision engine to obtain the instantaneous power command of the energy storage battery pack. Based on the instantaneous power command, a coordinated control command is generated, and the parameters of the energy storage converter and MPPT (Maximum Power Point Tracking) unit are adjusted according to the coordinated control command to realize intelligent control of the wave energy power supply system.

[0005] Optionally, the generative adversarial network integrating physical constraints includes a generator and a discriminator. The generator is a temporal convolutional network structure, and the loss function corresponding to the discriminator includes a physical constraint loss term based on the hydrodynamic equations of the wave energy conversion device and the electrical equations of the generator.

[0006] Optionally, the number of modes and the second-order penalty factor corresponding to the variational mode decomposition method are determined after online adaptive adjustment based on the spectral entropy and zero-crossing rate of the historical power sequence.

[0007] Optionally, the current operating condition data includes normalized total energy storage and SOC balance. Based on real-time parameter data, current operating condition data, and the heterogeneous battery coupled aging model, the system's comprehensive state indicators are calculated, including: Based on real-time parameter data and a heterogeneous battery coupled aging model, the synergistic aging cost is obtained; We perform weighted fusion calculations on normalized total energy storage, co-aging cost, and SOC balance to obtain the comprehensive system status index.

[0008] Optionally, the hierarchical deep reinforcement learning decision engine includes an upper-layer policy network and a lower-layer policy network. The upper-layer policy network is trained using the soft actor-commentator algorithm, and the reward function corresponding to the upper-layer policy network includes electricity sales revenue power, switching loss, converter loss, and aging cost increment. The lower-layer policy network is trained using the deep deterministic policy gradient algorithm.

[0009] Optionally, the low-frequency power component, high-frequency power component, predicted load power sequence, and system comprehensive state index are input into the hierarchical deep reinforcement learning decision engine to obtain the instantaneous power command of the energy storage battery pack, including: The low-frequency power component, the predicted load power sequence, and the system comprehensive state index are input into the upper-level strategy network to obtain the reference charge-discharge power for the energy storage battery pack. The reference charge-discharge power includes the first reference charge-discharge power of the corresponding energy-type energy storage battery pack and the second reference charge-discharge power of the corresponding power-type energy storage battery pack. The Hilbert-Huang transform characteristics of the high-frequency power components are determined, and the first reference charge-discharge power, the second reference charge-discharge power, and the Hilbert-Huang transform characteristics are input into the lower-level policy network to obtain the instantaneous power command. The instantaneous power command includes the first instantaneous power command corresponding to the energy-type energy storage battery pack and the second instantaneous power command corresponding to the power-type energy storage battery pack. The Hilbert-Huang transform features include instantaneous frequency and instantaneous amplitude.

[0010] Optionally, the parameters of the MPPT unit include at least one of the perturbation step size and the perturbation period of the perturbation observation method; If the amplitude of the high-frequency power component exceeds a preset threshold, the perturbation step size is reduced and the perturbation period is shortened.

[0011] Optionally, the method further includes: The system acquires real-time charge-discharge cycle data, real-time temperature data, and real-time actual decay data corresponding to the wave energy generation unit. Based on the real-time charge-discharge cycle data, real-time temperature data, and real-time actual decay data, the system uses a recursive least squares algorithm to update the empirical parameters in the heterogeneous battery coupled aging model. The real-time power generation sequence of the wave energy generation unit is obtained, and the residual between the real-time power generation sequence and the predicted power generation sequence is determined. Based on the residual, the weight coefficient of the physical constraint loss term is adjusted by the gradient descent method.

[0012] Optionally, the method further includes: If the peak parameter of the predicted power generation sequence exceeds the preset extreme safety threshold, the objective function of the upper-layer strategy network will be switched to minimize equipment impact, and the MPPT unit will be set to constant voltage and current limiting mode, with the peak parameter being the peak value or fluctuation gradient.

[0013] In another aspect, embodiments of this application provide an intelligent switching system for wave energy power supply based on dual SOC dynamic thresholds. The system comprises a wave energy power generation and conversion module, an energy storage bidirectional DC / AC converter, and a central controller. The wave energy power generation and conversion module includes an energy-type first energy storage battery pack and a power-type second energy storage battery pack. The central controller has a built-in processor and a memory. The memory stores a computer program. When the processor executes the computer program, it implements any one of the methods in the intelligent switching method for wave energy power supply based on dual SOC dynamic thresholds.

[0014] The beneficial effects of the technical solutions provided in this application include at least the following: In this application, a generative adversarial network integrating physical constraints is used to obtain the predicted power generation sequence of wave energy power generation units. Then, a variational mode decomposition method is used to separate low-frequency and high-frequency power components, corresponding to the slow trend and instantaneous impact of system energy, respectively. These components are then differentiated and utilized by a subsequent hierarchical deep reinforcement learning decision engine to achieve dynamic and coordinated control of heterogeneous battery packs. Furthermore, based on the system's comprehensive state index calculated using a coupled aging model, the "health" and "economic cost" of the current energy storage system are quantified. These are used as key inputs by the upper-layer policy network, ensuring that the established baseline charge-discharge strategy meets the long-term economic requirements of the battery and optimizing energy allocation and switching strategies. This method significantly improves the energy efficiency and stability of the wave energy system, enabling more efficient use of wave energy power generation resources, reducing energy waste, and significantly lowering system operating costs.

[0015] Furthermore, by dynamically adjusting the coordinated control of the MPPT unit and the energy storage bidirectional DC (Direct Current) / AC (Alternating Current) converter, the system can maintain stable operation even under extreme sea conditions, effectively coping with the impact of power fluctuations. Its significant advantages in improving system energy efficiency, extending equipment life, and reducing maintenance costs have broad application prospects and economic value.

[0016] Furthermore, the upper-layer policy network in the hierarchical deep reinforcement learning decision engine utilizes low-frequency power components and system state indicators to formulate macroscopic, economically-oriented baseline commands; while the lower-layer policy network utilizes high-frequency power components and upper-layer baseline commands to generate instantaneous, dynamic-performance-oriented compensation commands. This "upper-layer planning, lower-layer execution" structure achieves precise coordination between energy-type and power-type energy storage in terms of time scale and function. The actual operating data generated after executing control commands is fed back and used to update aging model parameters and physical constraint weights online, enabling the entire system to have self-evolution capabilities, resist model drift, and ensure the sustainability of technical performance under long-term operation. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating an intelligent switching method for a wave energy power supply system based on dual SOC dynamic thresholds, provided in an embodiment of this application; Figure 2A schematic diagram of an intelligent switching system for wave energy power supply based on dual SOC dynamic thresholds provided in this application embodiment; Figure 3 A simplified flowchart illustrating the steps of an intelligent switching method for a wave energy power supply system based on dual SOC dynamic thresholds, provided for an embodiment of this application; Figure 4 A schematic diagram of the structure of an intelligent switching device for a wave energy power supply system based on dual SOC dynamic thresholds provided in this application embodiment; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0019] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting the invention.

[0020] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this application means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0022] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0023] Specifically, such as Figure 1 As shown, the method may include: Step S101: Obtain the real-time output data of the wave energy power generation unit. Based on the real-time output data, obtain the predicted power generation sequence through a generative adversarial network with integrated physical constraints. Then, separate the predicted power generation sequence into low-frequency power components and high-frequency power components through a variational mode decomposition method with adaptive parameter adjustment. The real-time output data includes output voltage and output current.

[0024] Optionally, the output voltage Uwg(t) and output current Iwg(t) of the wave energy generation unit within the historical time window Th are acquired in real time, and the wave energy generation power sequence within the historical time window Th is obtained based on the output voltage and output current. Then, using a Generative Adversarial Network (GAN) model that integrates physical constraints, the instantaneous output power sequence of the wave energy generation unit within the future time window Tp is output. That is, predicting the power generation sequence.

[0025] In an optional embodiment of this application, the generative adversarial network integrating physical constraints includes a generator and a discriminator. The generator is a temporal convolutional network structure, and the loss function corresponding to the discriminator includes a physical constraint loss term based on the hydrodynamic equations of the wave energy conversion device and the electrical equations of the generator.

[0026] Optionally, the physically constrained generative adversarial network includes a generator and a discriminator. The generator is composed of a time convolutional network with dilated convolutions. The input is the wave energy power generation sequence within a historical time window Th, and the output is the instantaneous output power sequence. The discriminator is a convolutional neural network, specifically used to determine the instantaneous output power sequence. The true / false cases. The loss function corresponding to the discriminator during training. L D It consists of two parts, specifically: ; in, L adv To combat the losses, L phy For physical constraint loss terms, λ phy As a weighting coefficient, this physical constraint loss term is constructed from the parameterized slope equation based on the wave energy conversion device and the dq-axis voltage equation of the permanent magnet synchronous generator. It is used to penalize prediction results that do not conform to physical laws, and its specific expression is as follows: in, P Wave energy power density, C g For group velocity, γHere is the dissipation coefficient. S For source terms, u d 、u q 、i d and i q For generator voltage and current, L d and L q For inductance, ψ f It is a permanent magnet flux linkage. t This represents a time variable, used to describe the rate of change of wave energy power density over time. R This refers to the stator resistance of a permanent magnet synchronous generator, measured in ohms. ω This represents the electric angular velocity of the generator, measured in radians per second, and is related to the generator's rotational speed. In practical applications, it is used to predict power generation sequences. The results need to be verified using the physical equations described above, and the resulting residuals should be included in the loss function.

[0027] Optionally, in practical applications, a training set and an initial generative adversarial network can be pre-established. The training sample set includes each sample pair, and each sample pair includes sample data and corresponding sample labels. The sample data refers to the wave energy power generation sequence within the historical time window Th, and the corresponding sample labels are the annotation results of the instantaneous output power sequence of the wave energy power generation unit within the future time window Tp corresponding to the input sample data.

[0028] Furthermore, each sample pair can be input into the initial generative adversarial network (GAN). The initial GAN ​​can then output the instantaneous output power sequence prediction result for each sample pair. Based on the instantaneous output power sequence prediction result and the corresponding annotation result for each sample pair, the value of the loss function corresponding to the model can be determined. If the determined loss function value does not converge, the network parameters of the initial GAN ​​can be adjusted. Then, each sample pair is input into the adjusted initial GAN ​​again, and the value of the loss function is determined again based on the instantaneous output power sequence prediction result and the corresponding annotation result for the sample pair. If it still does not converge, the network parameters of the initial GAN ​​are adjusted again until the value of the corresponding loss function converges, thus obtaining the generative adversarial network in this application. The loss function of the model represents the difference between the predicted instantaneous output power sequence of each sample pair and the labeled result of each sample pair. When the loss function converges, it means that the difference between the predicted instantaneous output power sequence of each sample pair and the labeled result of each sample pair meets the requirement, that is, the predicted instantaneous output power sequence of the sample pair output by the model is close to the labeled result of the sample pair.

[0029] Furthermore, after obtaining the predicted power generation sequence, the predicted power generation sequence is separated into low-frequency power components P, which characterize the steady energy trend, using a variational mode decomposition method with adaptive parameter adjustment. genLF(t) and the high-frequency power component P characterizing severe power surges genHF(t) .

[0030] In an optional embodiment of this application, the number of modes and the second-order penalty factor corresponding to the variational mode decomposition method are determined by online adaptive adjustment based on the spectral entropy and zero-crossing rate of the historical power sequence.

[0031] Optionally, the mode number K and the second-order penalty factor α in the variational mode decomposition (VMD) method are determined after online adaptive adjustment based on the spectral entropy and zero-crossing rate of the historical power sequence. This ensures that the low-frequency power components and high-frequency power components obtained from the decomposition can most accurately reflect the physical characteristics of wave energy.

[0032] Here, spectral entropy (H) is calculated by performing a Fourier transform on the predicted power generation sequence to obtain the power spectrum S(f), then calculating its probability distribution using p(f) = S(f) / ΣS(f), and finally applying the formula H = -Σp(f). The spectral entropy obtained by log2p(f) indicates that the higher the spectral entropy value, the more complex the frequency components of the power sequence. The zero-crossing rate (ZCR) is the number of times the power sequence crosses the zero level per unit time, and the specific formula is: in, For indicator functions, This represents the total number of data points in the power sequence. For the first t The power signal value at each sampling time. The zero-crossing rate is an indicator function. When the product of the power signal values ​​of two adjacent sampling points is less than zero, a sign change occurs. The signal is recorded as 1 when it crosses zero, and as 0 otherwise. The entire formula calculates the average number of times the power signal crosses zero per unit time, which reflects the frequency of power fluctuations. The higher the zero-crossing rate, the more frequent the power fluctuations.

[0033] Step S102: Obtain the real-time parameter data and current operating condition data corresponding to the energy storage battery pack, and calculate the comprehensive system status index based on the real-time parameter data, current operating condition data and heterogeneous battery coupling aging model. The real-time parameter data includes the SOC of the energy storage battery pack and the SOH of the power storage battery pack.

[0034] Optionally, the energy storage battery pack includes an energy-type energy storage battery pack (i.e., the first energy storage battery pack) and a power-type energy storage battery pack (i.e., the second energy storage battery pack). The acquired real-time parameter data includes the first real-time state of charge (SOC1) and the first state of health (SOH1) of the first energy storage battery pack, and the second real-time state of charge (SOC2) and the second state of health (SOH2) of the second energy storage battery pack. Furthermore, the system's comprehensive state index can be calculated based on the real-time parameter data, current operating condition data, and the heterogeneous battery coupling aging model.

[0035] In optional embodiments of this application, the current operating condition data includes normalized total energy storage and SOC balance. Based on real-time parameter data, current operating condition data, and a heterogeneous battery coupled aging model, the system comprehensive state index is calculated, including: Based on real-time parameter data and a heterogeneous battery coupled aging model, the synergistic aging cost is obtained; We perform weighted fusion calculations on normalized total energy storage, co-aging cost, and SOC balance to obtain the comprehensive system status index.

[0036] Optionally, the current operating condition data includes normalized total energy storage and SOC balance. Based on real-time parameter data and a heterogeneous battery coupled aging model, the co-aging cost is obtained. Then, the normalized total energy storage, co-aging cost, and SOC balance are weighted and fused using the following formula to obtain the system's comprehensive state index: in, F sys As a comprehensive system status indicator, E norm It is the normalized total system energy storage. B norm It is the SOC balance calculated based on the Gini coefficient. C aging It is the predicted cost of co-aging over the next N cycles. w e , w b , w c These are the weighting coefficients, initially set to... w e =0.4, w b =0.3, w c =0.3, and can be dynamically adjusted according to the system's operational stage, such as the early or late stages of its lifespan; the system's comprehensive status index F sys It integrates current energy storage levels, balance, and future... N The expected lifetime decay cost per cycle.

[0037] Optionally, the heterogeneous battery coupled aging model described above is constructed separately for lithium iron phosphate (LFP) energy type batteries and lithium titanate (LTO) power type batteries. For LFP batteries, the capacity decay Q loss-LFP The model is: ; For LTO batteries, the internal resistance increases R growth-LTO The model is: ; in, A , B,z , n , m , σ For the parameters to be identified, E a For activation energy, R 0 The gas constant is... T For temperature, Ah For the amount of ampoules, CRateThe charge / discharge rate is given, ΔSOC is the depth of cycle (calculated based on the SOC variation range of the energy storage battery pack), and cycles is the number of cycles (obtained from the SOC cycle history statistics of the energy storage battery pack). The SOH of the power storage battery pack is used to correct the baseline internal resistance value of the internal resistance growth model. At this point, the co-aging cost C... aging for: ; in, Q end-LFP and R end-LTO This is the battery life end threshold. w 1, w 2 represents the economic weighting coefficient.

[0038] Step S103: Obtain the predicted load power sequence, and input the low-frequency power component, high-frequency power component, predicted load power sequence and system comprehensive state index into the hierarchical deep reinforcement learning decision engine to obtain the instantaneous power command of the energy storage battery pack.

[0039] Optionally, the predicted load power sequence can be obtained from a load prediction model, which is also a generative adversarial network integrating physical constraints, including a generator and a discriminator. The generator is based on a temporal modeling unit and can be a network structure such as LSTM, Temporal Convolutional Network (TCN), or Transformer. The discriminator can be a convolutional discriminator or a recursive discriminator, used to determine whether the input sequence is a real historical sequence or a generated sequence, forcing the sequence output by the generator to be statistically indistinguishable from the real data.

[0040] Correspondingly, when it is necessary to obtain the predicted load power sequence, historical load sequences, environmental features, and random noise can be obtained. Then, the historical load sequences, environmental features, and random noise are input into the generator, which generates the predicted load sequence based on the historical load sequences, environmental features, and random noise. Furthermore, a discriminator is used to determine whether the generated predicted load sequence is a real historical sequence or a generated sequence, forcing the sequence output by the generator to be statistically indistinguishable from the real data.

[0041] Optionally, the load prediction model uses historical load data, historical wave resource characteristics, and energy storage status as inputs during training. First, the data is preprocessed by dividing it into time windows and normalizing it. Then, an alternating optimization strategy is adopted. In each iteration, the initial generator parameters are fixed, and the discriminator is trained using real historical sequences and generated sequences to enhance its discrimination ability. Then, the initial discriminator parameters are fixed again, and the model is iterated until the loss converges by minimizing the total loss function of the generator. The load prediction model can then be obtained.

[0042] The total loss function for minimizing the generator is composed of a weighted sum of adversarial loss, data fitting loss, and physical constraint loss, expressed by the following formula: in, To mitigate losses and ensure the generated sequence distribution is accurate; To monitor the weight of losses, To monitor losses and ensure prediction accuracy; The weights for the physical constraint loss, The physical constraint loss is calculated by using a pre-established wave energy conversion model and an energy storage system model to determine the energy balance deviation, the degree of equipment boundary violation, and the time sequence continuity penalty term. This forces the generator to learn a mapping relationship that conforms to both historical data distribution and strictly follows the physical laws of the system.

[0043] Optionally, to achieve this physical constraint, the generator of the load prediction model embeds a differentiable physical model layer, enabling the output to automatically satisfy physical relationships. This physical model layer includes a residual correction structure and a projection layer. The residual correction structure is used by the generator to first output intermediate variables and then calculate the final load through the physical model. The projection layer is used to add a mapping operation in the last layer of the generator, forcibly constraining the output within the range of maximum and minimum load power. Furthermore, the load prediction model adds a dedicated discriminator, which takes as input the degree of violation of the physical constraints of the sequence and outputs a physical rationality score. This score, along with the adversarial loss, optimizes the generator, thereby further improving the physical feasibility of the output.

[0044] In an optional embodiment of this application, the hierarchical deep reinforcement learning decision engine includes an upper-layer policy network and a lower-layer policy network. The upper-layer policy network is trained using the soft actor-commentator algorithm, and the reward function corresponding to the upper-layer policy network includes electricity sales revenue power, switching losses, converter losses, and aging cost increments. The lower-layer policy network is trained using the deep deterministic policy gradient algorithm.

[0045] Optionally, the hierarchical deep reinforcement learning decision engine includes an upper-layer policy network and a lower-layer policy network. The upper-layer policy network aims to maximize the long-term economic benefits of the system, namely, the combined energy gain and battery replacement cost. It is trained using the Soft Actor-Critic (SAC) algorithm, and its reward function is... R meta Designed as follows: - in, To sell or supply the effective power of a load to the grid, and These are switching losses and converter losses, respectively. The increase in aging cost per unit time caused by the current decision. , , For reward weighting.

[0046] In addition, the upper-layer policy network has a multi-layer fully connected neural network structure. The input layer receives the aforementioned state variables, the hidden layer performs nonlinear feature transformation, and the output layer outputs the reference power values ​​of the two sets of batteries.

[0047] The lower-level policy network (Controller) is trained using the Deep Deterministic Policy Gradient (DDPG) algorithm. Its network structure is also a multi-layer fully connected neural network. The input layer receives high-frequency time-frequency features and instructions from the upper layer, the hidden layer performs feature fusion, and the output layer outputs power correction values.

[0048] In an optional embodiment of this application, the low-frequency power component, high-frequency power component, predicted load power sequence, and system comprehensive state index are input into a hierarchical deep reinforcement learning decision engine to obtain the instantaneous power command of the energy storage battery pack, including: The low-frequency power component, the predicted load power sequence, and the system comprehensive state index are input into the upper-level strategy network to obtain the reference charge-discharge power for the energy storage battery pack. The reference charge-discharge power includes the first reference charge-discharge power of the corresponding energy-type energy storage battery pack and the second reference charge-discharge power of the corresponding power-type energy storage battery pack. The Hilbert-Huang transform characteristics of the high-frequency power components are determined, and the first reference charge-discharge power, the second reference charge-discharge power, and the Hilbert-Huang transform characteristics are input into the lower-level policy network to obtain the instantaneous power command. The instantaneous power command includes the first instantaneous power command corresponding to the energy-type energy storage battery pack and the second instantaneous power command corresponding to the power-type energy storage battery pack. The Hilbert-Huang transform features include instantaneous frequency and instantaneous amplitude.

[0049] Optionally, the low-frequency power component, predicted load power, and system comprehensive state index are input to the upper-level policy network. The actor network then generates a baseline charge / discharge power action, and the critic network evaluates the long-term value of this action. The optimal baseline charge / discharge power is output by maximizing the reward function. This baseline charge / discharge power includes the first baseline charge / discharge power P corresponding to the energy storage battery pack. b1 and the second reference charge / discharge power P of the corresponding power-type energy storage battery pack b2 .

[0050] Furthermore, the Hilbert-Huang transform characteristics of the high-frequency power components are determined. Then, the first and second reference charge-discharge powers output by the upper-level strategy network are used as inputs to the lower-level strategy network. The lower-level strategy network deterministically outputs instantaneous power correction values, which are then superimposed with the reference power to form the first instantaneous power command P1(t) and the second instantaneous power command P1(t) of the corresponding energy-type energy storage battery pack and the corresponding power-type energy storage battery pack, respectively.

[0051] Among them, the Hilbert-Huang transform features include the instantaneous frequency f inst (t) and instantaneous amplitude a inst (t), at this time, by analyzing the high-frequency power component P gen-HF (t) Empirical Mode Decomposition (EMD) is performed to obtain the Intrinsic Mode Function (IMF), and then Hilbert Transform is applied to the IMF to obtain the analytic signal, thereby determining the instantaneous frequency f. inst (t) and instantaneous amplitude a inst (t). The variational mode decomposition (VMD) adaptively decomposes the non-stationary wave energy power generation sequence into low-frequency trend components and high-frequency fluctuation components, enabling upper-level control to focus on energy management and lower-level control to focus on power smoothing, achieving hierarchical collaborative control on a time scale. The instantaneous frequency and amplitude of the high-frequency power components extracted by the Hilbert-Huang transform can provide early warning of millisecond-level power surges for the lower-level strategy network, enabling power-type energy storage to quickly respond to peak shaving and valley filling. Furthermore, the obtained instantaneous frequency and amplitude can be input into the lower-level strategy network, allowing it to respond quickly to millisecond-level power surges.

[0052] Step S104: Generate a coordinated control command based on the instantaneous power command, and adjust the parameters of the energy storage converter and MPPT unit according to the coordinated control command to realize intelligent control of the wave energy power supply system.

[0053] Optionally, based on the instantaneous power command output by the lower-level strategy network, a coordinated control command for the bidirectional DC / AC converter and MPPT unit of the energy storage can be generated, which can achieve impedance matching and optimal power transfer between the generation end and the energy storage end.

[0054] In an optional embodiment of this application, the parameters of the MPPT unit include at least one of the perturbation step size and the perturbation period of the perturbation observation method; wherein, if the amplitude of the high-frequency power component exceeds a preset threshold, the perturbation step size is reduced and the perturbation period is shortened.

[0055] Optionally, the coordinated control commands include dynamically adjusting the disturbance step size ΔU or disturbance period ΔT of the MPPT unit's disturbance observation method, such as when a drastic power input is predicted in the future, i.e., P... genHF(t) When the amplitude exceeds the threshold, the perturbation step size is temporarily reduced and the perturbation period is shortened to improve the dynamic tracking speed and acquisition efficiency of MPPT.

[0056] In optional embodiments of this application, the method further includes: The system acquires real-time charge-discharge cycle data, real-time temperature data, and real-time actual decay data corresponding to the wave energy generation unit. Based on the real-time charge-discharge cycle data, real-time temperature data, and real-time actual decay data, the system uses a recursive least squares algorithm to update the empirical parameters in the heterogeneous battery coupled aging model. The real-time power generation sequence of the wave energy generation unit is obtained, and the residual between the real-time power generation sequence and the predicted power generation sequence is determined. Based on the residual, the weight coefficient of the physical constraint loss term is adjusted by the gradient descent method.

[0057] Optionally, in practical applications, real-time charge-discharge cycle data, real-time temperature data, and real-time actual degradation data corresponding to the wave energy generation unit can also be acquired. Then, the acquired data is compared with the predicted values. If discrepancies exist, an extended Kalman filter algorithm is used to identify and update the empirical parameters in the heterogeneous battery coupled aging model online, making the model prediction closer to the actual degradation trend. For example, the central controller has an embedded digital twin module that runs the coupled aging model and a simplified version of the integrated physical constraint GAN, and performs this every 24 hours or after each complete charge-discharge cycle. Furthermore, the measured battery capacity and internal resistance data from the battery management system are compared with the model predictions. If discrepancies exist, an extended Kalman filter algorithm is used to update the empirical parameters in the model. A, B, z, n, m,σ Online identification and updates are performed to make the model predictions approximate the actual decay trend.

[0058] Furthermore, the digital twin module can calculate the real-time power generation sequence of the wave energy generation unit over the past 24 hours and determine the residual between the real-time power generation sequence and the predicted power generation sequence. Then, based on the residual, it adjusts the weight coefficients of the physical constraint loss term using the gradient descent method. For example, the digital twin module calculates the root mean square error (RMSE) between the predicted and actual power over the past 24 hours. If the RMSE continues to increase, the weight coefficients of the physical constraint loss term in the loss function are increased using the gradient descent method to enhance the constraint effect of physical laws on the prediction results. If the error decreases, the weights are appropriately reduced to improve model flexibility.

[0059] In optional embodiments of this application, the method further includes: If the peak parameter of the predicted power generation sequence exceeds the preset extreme safety threshold, the objective function of the upper-layer strategy network will be switched to minimize equipment impact, and the MPPT unit will be set to constant voltage and current limiting mode, with the peak parameter being the peak value or fluctuation gradient.

[0060] Optionally, this application also includes an extreme sea state safety mode: when the peak parameter of the predicted power generation sequence, i.e., the peak value or fluctuation gradient, exceeds a preset extreme safety threshold, the system is forced to enter this mode. In this mode, the objective function of the upper-layer policy network switches to minimizing equipment impact and ensuring battery safety. The MPPT unit is instructed to operate in constant voltage and current limiting mode, and the second energy storage battery pack is preferentially used to absorb all energy exceeding the system's rated processing capacity, and the excess energy is dissipated through the discharge resistor to protect the entire system.

[0061] In this application, a generative adversarial network integrating physical constraints is used to obtain the predicted power generation sequence of wave energy power generation units. Then, a variational mode decomposition method is used to separate low-frequency and high-frequency power components, corresponding to the slow trend and instantaneous impact of system energy, respectively. These components are then differentiated and utilized by a subsequent hierarchical deep reinforcement learning decision engine to achieve dynamic and coordinated control of heterogeneous battery packs. Furthermore, based on the system's comprehensive state index calculated using a coupled aging model, the "health" and "economic cost" of the current energy storage system are quantified. These are used as key inputs by the upper-layer policy network, ensuring that the established baseline charge-discharge strategy meets the long-term economic requirements of the battery and optimizing energy allocation and switching strategies. This method significantly improves the energy efficiency and stability of the wave energy system, enabling more efficient use of wave energy power generation resources, reducing energy waste, and significantly lowering system operating costs.

[0062] Furthermore, by dynamically adjusting the coordinated control of the MPPT unit and the bidirectional DC / AC converter for energy storage, the system can maintain stable operation even under extreme sea conditions, effectively coping with the impact of power fluctuations. Its significant advantages in improving system energy efficiency, extending equipment life, and reducing maintenance costs have broad application prospects and economic value.

[0063] Furthermore, the upper-layer policy network in the hierarchical deep reinforcement learning decision engine utilizes low-frequency power components and system state indicators to formulate macroscopic, economically-oriented baseline commands; while the lower-layer policy network utilizes high-frequency power components and upper-layer baseline commands to generate instantaneous, dynamic-performance-oriented compensation commands. This "upper-layer planning, lower-layer execution" structure achieves precise coordination between energy-type and power-type energy storage in terms of time scale and function. The actual operating data generated after executing control commands is fed back and used to update aging model parameters and physical constraint weights online, enabling the entire system to have self-evolution capabilities, resist model drift, and ensure the sustainability of technical performance under long-term operation.

[0064] like Figure 2As shown, this application also provides an intelligent switching system for wave energy power supply based on dual SOC dynamic thresholds. The system comprises a wave energy power generation and conversion module, an energy storage bidirectional DC / AC converter (i.e., energy storage converter), and a central controller. The wave energy power generation and conversion module includes an energy-type first energy storage battery pack (i.e., energy-type battery pack) and a power-type second energy storage battery pack (i.e., power-type battery pack). The energy storage converter is connected to the load / grid. The central controller has a built-in processor and memory. The memory stores a computer program. When the processor executes the computer program, it implements any one of the methods in the intelligent switching of the wave energy power supply system based on dual SOC dynamic thresholds.

[0065] The first energy storage battery pack consists of multiple lithium iron phosphate (LFP) battery packs connected in series and parallel, featuring high energy density, long cycle life, and a rated capacity greater than 1000 Ah. The second power storage battery pack consists of high-rate lithium titanate (LTO) batteries or supercapacitor modules, featuring high power density, ultra-long cycle life, and fast charge / discharge capabilities, with a rated power 5-10 times that of the first energy storage battery pack. Each battery pack is equipped with an independent battery management system (BMS) to monitor and upload voltage, current, temperature, and internal electrochemical state parameters.

[0066] The central controller further includes a digital twin module, which receives the system's operating data in real time and corrects the parameters in the heterogeneous battery coupling aging model and the physical constraint loss term mentioned above online to resist model drift and ensure the accuracy of long-term control.

[0067] Optionally, the execution steps of the method provided in this application can be simplified as follows: Figure 3 As shown, it specifically includes multi-domain power prediction, heterogeneous energy storage state assessment, multi-objective coordination strategy, control command generation and execution, and feedback optimization.

[0068] Optionally, to verify the effectiveness of the method in this application, this application also built a wave energy power supply system model on the MATLAB / Simulink and OPAL-RT real-time simulation platform and compared it with two benchmark methods: (1) the traditional fixed threshold switching method; (2) the hierarchical reinforcement learning method without aging model, i.e. the method provided in this application.

[0069] The simulation conditions used JONSWAP wave spectrum to simulate one year's sea state data for a certain sea area. The load consisted of mixed residential and communication base station loads. The LFP battery pack had a capacity of 1200Ah, and the LTO battery pack had a power capacity eight times that of the LFP battery pack. The comparison results are shown in the table below. Correspondingly, as shown in the table above, the method provided in this application improves energy utilization by approximately 19.4 percentage points compared to traditional methods. Compared to the HDRL method without an aging model, it further improves energy utilization while significantly slowing down the degradation rate of both types of batteries. It is expected to extend the overall lifespan of the battery pack by more than 35% and significantly reduce long-term operating costs.

[0070] This application provides an intelligent switching device for a wave energy power supply system based on dual SOC dynamic thresholds, such as... Figure 4 As shown, the device may include: a power component determination module 401, a status index determination module 402, a power command determination module 403, and a parameter adjustment module 404, wherein, The power component determination module is used to acquire the real-time output data of the wave energy generation unit. Based on the real-time output data, the predicted power generation sequence is obtained through a generative adversarial network with integrated physical constraints. The predicted power generation sequence is then separated into low-frequency power components and high-frequency power components through a variational mode decomposition method with adaptive parameter adjustment. The real-time output data includes output voltage and output current. The status index determination module is used to acquire real-time parameter data and current operating condition data corresponding to the energy storage battery pack, and calculate the comprehensive status index of the system based on the real-time parameter data, current operating condition data and heterogeneous battery coupling aging model. The real-time parameter data includes the SOC of the energy storage battery pack and the SOH of the power storage battery pack. The power command determination module is used to obtain the predicted load power sequence and input the low-frequency power component, high-frequency power component, predicted load power sequence and system comprehensive state index into the hierarchical deep reinforcement learning decision engine to obtain the instantaneous power command of the energy storage battery pack. The parameter adjustment module is used to generate coordinated control commands based on instantaneous power commands, and adjust the parameters of the energy storage converter and MPPT unit according to the coordinated control commands to realize intelligent control of the wave energy power supply system.

[0071] Optionally, the generative adversarial network integrating physical constraints includes a generator and a discriminator. The generator is a temporal convolutional network structure, and the loss function corresponding to the discriminator includes a physical constraint loss term based on the hydrodynamic equations of the wave energy conversion device and the electrical equations of the generator.

[0072] Optionally, the number of modes and the second-order penalty factor corresponding to the variational mode decomposition method are determined after online adaptive adjustment based on the spectral entropy and zero-crossing rate of the historical power sequence.

[0073] Optionally, the current operating condition data includes normalized total energy storage and SOC balance. When the state index determination module calculates the comprehensive system state index based on real-time parameter data, current operating condition data, and heterogeneous battery coupled aging model, it is specifically used for: Based on real-time parameter data and a heterogeneous battery coupled aging model, the synergistic aging cost is obtained; We perform weighted fusion calculations on normalized total energy storage, co-aging cost, and SOC balance to obtain the comprehensive system status index.

[0074] Optionally, the hierarchical deep reinforcement learning decision engine includes an upper-layer policy network and a lower-layer policy network. The upper-layer policy network is trained using the soft actor-commentator algorithm, and the reward function corresponding to the upper-layer policy network includes electricity sales revenue power, switching loss, converter loss, and aging cost increment. The lower-layer policy network is trained using the deep deterministic policy gradient algorithm.

[0075] Optionally, when the power command determination module inputs the low-frequency power component, high-frequency power component, predicted load power sequence, and system comprehensive state index into the hierarchical deep reinforcement learning decision engine to obtain the instantaneous power command of the energy storage battery pack, it is specifically used for: The low-frequency power component, the predicted load power sequence, and the system comprehensive state index are input into the upper-level strategy network to obtain the reference charge-discharge power for the energy storage battery pack. The reference charge-discharge power includes the first reference charge-discharge power of the corresponding energy-type energy storage battery pack and the second reference charge-discharge power of the corresponding power-type energy storage battery pack. The Hilbert-Huang transform characteristics of the high-frequency power components are determined, and the first reference charge-discharge power, the second reference charge-discharge power, and the Hilbert-Huang transform characteristics are input into the lower-level policy network to obtain the instantaneous power command. The instantaneous power command includes the first instantaneous power command corresponding to the energy-type energy storage battery pack and the second instantaneous power command corresponding to the power-type energy storage battery pack. The Hilbert-Huang transform characteristics include the instantaneous frequency and the instantaneous amplitude.

[0076] Optionally, the parameters of the MPPT unit include at least one of the perturbation step size and the perturbation period of the perturbation observation method; wherein, if the amplitude of the high-frequency power component exceeds a preset threshold, the perturbation step size is reduced and the perturbation period is shortened.

[0077] Optionally, the device also includes a model parameter update module, specifically used for: The system acquires real-time charge-discharge cycle data, real-time temperature data, and real-time actual decay data corresponding to the wave energy generation unit. Based on the real-time charge-discharge cycle data, real-time temperature data, and real-time actual decay data, the system uses a recursive least squares algorithm to update the empirical parameters in the heterogeneous battery coupled aging model. The real-time power generation sequence of the wave energy generation unit is obtained, and the residual between the real-time power generation sequence and the predicted power generation sequence is determined. Based on the residual, the weight coefficient of the physical constraint loss term is adjusted by the gradient descent method.

[0078] Optionally, the device also includes a mode update module, specifically used for: If the peak parameter of the predicted power generation sequence exceeds the preset extreme safety threshold, the objective function of the upper-layer strategy network will be switched to minimize equipment impact, and the MPPT unit will be set to constant voltage and current limiting mode, with the peak parameter being the peak value or fluctuation gradient.

[0079] The intelligent switching device for a wave energy power supply system based on dual SOC dynamic thresholds in this embodiment can execute the intelligent switching method for a wave energy power supply system based on dual SOC dynamic thresholds shown in the embodiment of this application. The implementation principle is similar and will not be repeated here.

[0080] This application provides an electronic device, which includes a processor and a memory configured to store machine-readable instructions that, when executed by the processor, cause the processor to perform a smart switching method for a wave energy power supply system based on dual SOC dynamic thresholds.

[0081] This application provides an electronic device, such as... Figure 5 As shown, Figure 5 The illustrated electronic device includes a processor 2001 and a memory 2003. The processor 2001 and the memory 2003 are connected, for example, via a bus 2002. Optionally, the electronic device 2000 may further include a transceiver 2004. It should be noted that in practical applications, the transceiver 2004 is not limited to one type, and the structure of this electronic device 2000 does not constitute a limitation on the embodiments of this application.

[0082] Processor 2001 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 2001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0083] Bus 2002 may include a pathway for transmitting information between the aforementioned components. Bus 2002 may be a PCI bus or an EISA bus, etc. Bus 2002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0084] The memory 2003 may be ROM or other type of static storage device capable of storing static information and instructions, RAM or other type of dynamic storage device capable of storing information and instructions, or EEPROM, CD-ROM or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0085] The memory 2003 stores the application code that executes the scheme of this application, and its execution is controlled by the processor 2001. The processor 2001 executes the application code stored in the memory 2003 to implement... Figure 4 The embodiment shown illustrates the operation of an intelligent switching device for a wave energy power supply system based on dual SOC dynamic thresholds.

[0086] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0087] The above description is only a partial embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A smart switching method for wave energy power supply systems based on dual SOC dynamic thresholds, characterized in that, include: The real-time output data of the wave energy generation unit is acquired. Based on the real-time output data, a predicted power generation sequence is obtained through a generative adversarial network with integrated physical constraints. The predicted power generation sequence is then separated into low-frequency power components and high-frequency power components using a variational mode decomposition method with adaptive parameter adjustment. The real-time output data includes output voltage and output current. The system acquires real-time parameter data and current operating condition data corresponding to the energy storage battery pack, and calculates the comprehensive system status index based on the real-time parameter data, the current operating condition data and the heterogeneous battery coupling aging model. The real-time parameter data includes the SOC of the energy storage battery pack and the SOH of the power storage battery pack. The predicted load power sequence is obtained, and the low-frequency power component, the high-frequency power component, the predicted load power sequence, and the system comprehensive state index are input into the hierarchical deep reinforcement learning decision engine to obtain the instantaneous power command of the energy storage battery pack. Based on the instantaneous power command, a coordinated control command is generated, and the parameters of the energy storage converter and MPPT unit are adjusted according to the coordinated control command to realize intelligent control of the wave energy power supply system.

2. The method according to claim 1, characterized in that, The integrated physical constraint generative adversarial network includes a generator and a discriminator. The generator is a temporal convolutional network structure, and the loss function corresponding to the discriminator includes a physical constraint loss term based on the hydrodynamic equations of the wave energy conversion device and the electrical equations of the generator.

3. The method according to claim 1, characterized in that, The mode number and the second-order penalty factor corresponding to the variational mode decomposition method are determined by online adaptive adjustment based on the spectral entropy and zero-crossing rate of the historical power sequence.

4. The method according to claim 1, characterized in that, The current operating condition data includes normalized total energy storage and SOC balance. Based on the real-time parameter data, the current operating condition data, and the heterogeneous battery coupled aging model, the system comprehensive state index is calculated, including: Based on the real-time parameter data and the heterogeneous battery coupled aging model, the synergistic aging cost is obtained; The normalized total energy storage, the collaborative aging cost, and the SOC balance are weighted and fused to obtain the comprehensive system status index.

5. The method according to claim 1, characterized in that, The hierarchical deep reinforcement learning decision engine includes an upper-layer policy network and a lower-layer policy network. The upper-layer policy network is trained using the soft actor-commentator algorithm, and the reward function corresponding to the upper-layer policy network includes electricity sales revenue power, switching losses, converter losses, and aging cost increments. The lower-layer policy network is trained using the deep deterministic policy gradient algorithm.

6. The method according to claim 5, characterized in that, The step of inputting the low-frequency power component, the high-frequency power component, the predicted load power sequence, and the system comprehensive state index into the hierarchical deep reinforcement learning decision engine to obtain the instantaneous power command of the energy storage battery pack includes: The low-frequency power component, the predicted load power sequence, and the system comprehensive state index are input to the upper-layer strategy network to obtain the reference charge-discharge power for the energy storage battery pack. The reference charge-discharge power includes a first reference charge-discharge power corresponding to the energy-type energy storage battery pack and a second reference charge-discharge power corresponding to the power-type energy storage battery pack. The Hilbert-Huang transform characteristics of the high-frequency power component are determined, and the first reference charge-discharge power, the second reference charge-discharge power, and the Hilbert-Huang transform characteristics are input into the lower-level policy network to obtain the instantaneous power command. The instantaneous power command includes a first instantaneous power command corresponding to the energy-type energy storage battery pack and a second instantaneous power command corresponding to the power-type energy storage battery pack. The Hilbert-Huang transform feature includes instantaneous frequency and instantaneous amplitude.

7. The method according to claim 1, characterized in that, The parameters of the MPPT unit include at least one of the perturbation step size and the perturbation period of the perturbation observation method. If the amplitude of the high-frequency power component exceeds a preset threshold, the perturbation step size is reduced and the perturbation period is shortened.

8. The method according to claim 2, characterized in that, The method further includes: The real-time charge-discharge cycle data, real-time temperature data, and real-time actual decay data corresponding to the wave energy power generation unit are obtained, and the empirical parameters in the heterogeneous battery coupling aging model are updated using the recursive least squares algorithm based on the real-time charge-discharge cycle data, the real-time temperature data, and the real-time actual decay data. The real-time power generation sequence of the wave energy generation unit is obtained, and the residual between the real-time power generation sequence and the predicted power generation sequence is determined. Based on the residual, the weight coefficient of the physical constraint loss term is adjusted using the gradient descent method.

9. The method according to claim 5, characterized in that, The method further includes: If the peak parameter of the predicted power generation sequence exceeds the preset extreme safety threshold, the objective function of the upper-layer strategy network is switched to minimize equipment impact, and the MPPT unit is set to constant voltage and current limiting mode. The peak parameter is the peak value or fluctuation gradient.

10. A smart switching system for wave energy power supply based on dual SOC dynamic thresholds, characterized in that, The system comprises a wave energy generation and conversion module, an energy storage bidirectional DC / AC converter, and a central controller. The wave energy generation and conversion module includes an energy-type first energy storage battery pack and a power-type second energy storage battery pack. The central controller has a built-in processor and a memory. The memory stores a computer program. When the processor executes the computer program, it implements the method of any one of claims 1 to 9.

Images