Bidirectional inverter coordination current conversion method and system based on ai prediction

Abstract

The application provides a bidirectional inversion collaborative conversion method and system based on AI prediction, and relates to the technical field of bidirectional inversion collaborative conversion. The method comprises the following steps: collecting multi-dimensional operation data in real time, and extracting data features; predicting the load demand change trend in a future preset time period; constructing a dynamic correlation network of multi-dimensional data by using a graph neural network; establishing a multi-objective collaborative optimization model, and obtaining a variable conversion strategy solution set to screen an optimal variable conversion strategy; converting the optimal variable conversion strategy into a PWM control signal, adjusting the switching state of a bidirectional inversion circuit, and realizing collaborative conversion control. Through AI prediction and multi-objective collaborative optimization, the application improves the battery charging conversion efficiency of an electric two-wheeled vehicle, shortens the charging time, and reduces energy consumption; the application perceives multi-scene load changes in advance, quickly adjusts the strategy to prevent charging interruption, and has strong adaptability; the application reduces charging harmonics, reduces power grid and equipment interference, protects the battery from overcharging and overheating, and prolongs the service life.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of bidirectional inverter cooperative current conversion technology, and in particular to a bidirectional inverter cooperative current conversion method and system based on AI prediction. BACKGROUND

[0002] Electric two-wheeled vehicles, as convenient short-distance travel tools, play an important role in urban transportation. The core power system relies on the cooperative work of batteries and charging equipment, and bidirectional inverter cooperative current conversion technology is the key to realizing efficient conversion of grid energy and safe charging and discharging of batteries, directly affecting charging efficiency, battery life, and stability of the charging process.

[0003] The conventional current conversion technology of electric two-wheeled vehicles mostly adopts a relatively simple control strategy, such as conventional pulse width modulation (PWM) control. In terms of energy conversion, the efficiency is low, and there is a lot of energy loss. During the charging process, the conventional technology cannot dynamically adapt to the needs of different SOC (state of charge) stages of the battery, for example, it is difficult to match the power when a large current is needed at the initial stage of charging, and energy is easily wasted when the battery is close to full at the later stage of charging, which not only prolongs the charging time, but also generates additional heat due to energy loss, adversely affecting the life of the battery and the current conversion components. When the grid voltage fluctuates frequently, the battery temperature is abnormal, and other situations occur, the conventional current conversion system responds slowly and cannot quickly adjust the control parameters, which easily leads to problems such as interruption of charging, unstable voltage and current output, and even may cause safety hazards such as overcharging and bulging of the battery. At the same time, the conventional current conversion technology has insufficient harmonic suppression capability, and the harmonics generated during the charging process will interfere with the stability of the grid, affect the normal work of surrounding electrical equipment, and increase internal losses of the current conversion system, further reducing charging efficiency and component reliability.

[0004] With the increasing requirements of people for charging speed, safety, and compatibility of electric two-wheeled vehicles, the conventional current conversion technology has been difficult to meet the needs. Especially in the new charging scenarios such as "light storage charging" integration and regenerative braking energy recovery, the conventional technology cannot adapt to the multi-mode charging needs, and the energy conversion efficiency and scenario compatibility shortcomings are increasingly prominent. There is an urgent need for a more advanced bidirectional inverter cooperative current conversion method and system to improve the energy conversion efficiency during the charging process, enhance the stability and safety of the charging process, suppress harmonic interference, and adapt to multi-scenario charging needs, and ensure efficient and reliable operation of the electric two-wheeled vehicle charging system. SUMMARY

[0005] The present application provides a bidirectional inverter cooperative current conversion method and system based on AI prediction to solve the defects in the prior art.

[0006] In one aspect, the present application provides a bidirectional inverter cooperative current conversion method based on AI prediction, comprising:

[0007] Real-time acquisition of multi-dimensional running data of the bidirectional inverter system.

[0008] Denoising processing is performed on the multi-dimensional running data, and kernel principal component analysis is adopted to extract data characteristics.

[0009] A load prediction model based on a space-time graph convolution network is constructed, data characteristics are combined, data space-time correlation rules are mined, and a load demand change trend in a preset time period in the future is predicted.

[0010] An adaptive weighted fusion algorithm is adopted to fuse the multi-dimensional running data, a graph neural network is used to construct a dynamic correlation network of multi-dimensional data, and the coupling relationship between data is mined.

[0011] According to the load demand change trend and the dynamic correlation network, a multi-objective collaborative optimization model is established, with the maximum conversion efficiency, the minimum harmonic distortion rate and the minimum switching loss as the objectives.

[0012] A multi-objective particle swarm optimization algorithm is adopted to solve the multi-objective collaborative optimization model, and a conversion strategy solution set is obtained.

[0013] The optimal conversion strategy is converted into a PWM control signal by using a digital signal processor, and the switching state of the bidirectional inverter circuit is adjusted according to the PWM control signal, so that collaborative conversion control is realized.

[0014] According to the bidirectional inverter collaborative conversion method based on AI prediction provided by the application, the multi-dimensional running data includes DC side voltage, DC side current, AC side voltage, AC side current and switch tube temperature.

[0015] According to the bidirectional inverter collaborative conversion method based on AI prediction provided by the application, the process of extracting data characteristics by using kernel principal component analysis includes:

[0016] A multi-dimensional running data matrix is constructed, and the data matrix is standardized.

[0017] A Gaussian kernel function is selected to construct a kernel matrix, and the kernel matrix is centralized.

[0018] The eigenvalues and eigenvectors of the kernel matrix are calculated, the eigenvectors corresponding to the first k1 largest eigenvalues are selected as principal components, and voltage characteristics, current characteristics and temperature characteristics are obtained.

[0019] According to the bidirectional inverter collaborative conversion method based on AI prediction provided by the application, the process of constructing a load prediction model based on a space-time graph convolution network includes:

[0020] Collect historical operation data samples of the bidirectional inverter system, divide each sample into T time steps according to time sequence, and each time step contains voltage characteristics, current characteristics and temperature characteristics.

[0021] A space-time data structure is constructed, the space-time data structure takes multi-dimensional characteristics of each time step as a graph convolution node, and the association between time steps is taken as a graph convolution edge, a time series graph convolution sequence is formed, and a corresponding load value of each time series graph sequence is labeled, and a sample set is formed.

[0022] A space-time graph convolution network model is designed, the space-time graph convolution network model comprises a space-time convolution layer, a time convolution layer and a prediction output layer. The space-time convolution layer adopts graph convolution to capture the spatial association between characteristics in the same time step. The time convolution layer is used to capture the time sequence association between different time steps. The prediction output layer is used to output the load demand change trend in the future preset time period.

[0023] The space-time convolution layer and the time convolution layer are stacked to extract deep space-time features and input the full connection prediction layer.

[0024] The mean square error is used as the loss function, and the model parameters are iteratively updated by the Adam optimization algorithm until the loss function converges, and the load prediction model is obtained.

[0025] According to the bidirectional inverter collaborative variable flow method based on AI prediction provided by the application, the process of constructing a multi-dimensional data dynamic association network by using a graph neural network comprises:

[0026] The multi-dimensional data fused by adaptive weighting is used as a graph neural node.

[0027] The similarity matrix between the graph neural nodes is calculated and used as the adjacency matrix of the graph neural network.

[0028] A graph neural network model is constructed, including an input layer, a graph convolution layer and an output layer, the input layer receives the graph neural node features. The graph convolution layer aggregates the neighbor graph neural node information through the adjacency matrix. The output layer outputs the association strength between the graph neural nodes.

[0029] The graph neural network model is trained, the graph neural node feature reconstruction error is used as the loss function, the model parameters are optimized, and the dynamic association network of the multi-dimensional data is obtained.

[0030] According to the bidirectional inverter collaborative variable flow method based on AI prediction provided by the application, the process of establishing a multi-objective collaborative optimization model comprises:

[0031] According to the power loss of the bidirectional inverter system, the variable flow efficiency is defined as the ratio of the output active power to the input active power.

[0032] Based on Fourier decomposition, the AC side current harmonic distortion rate is defined.

[0033] In combination with the switching frequency and the energy consumption characteristics of the switching tube, the switching loss is defined as the product of the switching frequency and the total energy consumption in a single switching cycle.

[0034] The constraint coupling modeling combines the topology characteristics and the operation safety requirements of the bidirectional inverter system, establishes a plurality of constraint coupling relationships, including electrical constraints, control constraints and coupling constraints.

[0035] The optimization function integration combines the weighting coefficient method and the target normalization, constructs a unified multi-objective optimization function, and embeds the plurality of constraint coupling relationships to obtain a multi-objective collaborative optimization model.

[0036] According to the bidirectional inverter collaborative variable flow method based on AI prediction provided by the application, the process of solving the multi-objective collaborative optimization model by using the multi-objective particle swarm optimization algorithm includes:

[0037] N1 particles are randomly generated, each particle corresponds to a group of decision variables, satisfies the constraint condition, initializes the velocity vector and individual optimal solution of the particle, and sets the external archive set to store the non-dominated solution.

[0038] S11, the target function value of each particle is calculated, the particle is compared with the individual optimal solution, if the particle dominates the individual optimal solution, the individual optimal solution is updated.

[0039] S12, the particle is compared with the solution in the external archive set, if the particle is a non-dominated solution and is not in the archive set, the particle is added to the archive set, if the size of the archive set exceeds the preset value, the redundant solution is deleted by using the crowding degree sorting.

[0040] S13, the particle velocity and position are updated.

[0041] Steps S11-S13 are repeated until the iteration number reaches the preset value, and the solution in the external archive set is output as the variable flow strategy solution set.

[0042] According to the bidirectional inverter collaborative variable flow method based on AI prediction provided by the application, the process of screening the optimal variable flow strategy by using the analytic hierarchy process includes:

[0043] The variable flow efficiency, the harmonic distortion rate and the switching loss are taken as the criterion layer, the strategies in the variable flow strategy solution set are taken as the scheme layer, and the judgment matrix is constructed.

[0044] The maximum eigenvalue and the corresponding eigenvector of the judgment matrix are calculated, and the consistency test is performed.

[0045] The performance of the strategy under each criterion is normalized to obtain a normalized matrix, the comprehensive score is calculated, and the strategy with the highest score is selected as the optimal variable flow strategy.

[0046] The application provides a bidirectional inversion collaborative variable flow method based on AI prediction.

[0047] S21, a discrete state space model of the bidirectional inversion circuit is established, and the formula is as follows:

[0048]

[0049]

[0050] In the formula, is a state vector at the moment, is a switch control signal, is a disturbance, is an output vector, A, B, C and D are system matrices, is measurement noise.

[0051] S22, a target function is defined, and the formula is as follows:

[0052]

[0053] In the formula, J is a target function value of model prediction control, is a prediction time domain, is a control time domain, is a predicted value of system output at the moment to a future moment, is a reference output at the moment is a predicted value of control input at the moment to a future moment, and Q and R are weight matrices.

[0054] S23, in each control cycle, a quadratic programming problem with constraints is solved to obtain an optimal control sequence , and the PWM control signal at the current moment is output to the switch tube as a PWM control signal.

[0055] In the next control cycle, the state vector is updated, and steps S21-S23 are repeated to realize dynamic adjustment.

[0056] On the other hand, the application also provides a bidirectional inversion collaborative variable flow system based on AI prediction, which comprises:

[0057] A data acquisition module is configured to acquire multi-dimensional operation data of the bidirectional inversion system in real time.

[0058] ​​​The feature extraction module is configured to perform denoising processing on the multi-dimensional operation data and extract data features by using kernel principal component analysis.

[0059] The load prediction module is configured to construct a load prediction model based on a spatio-temporal graph convolution network, combine data features, mine spatio-temporal correlation rules of the data, and predict a load demand change trend in a preset time period in the future.

[0060] The correlation mining module is configured to fuse the multi-dimensional operation data by using an adaptive weighted fusion algorithm, construct a dynamic correlation network of multi-dimensional data by using a graph neural network, and mine coupling relationships between the data.

[0061] The strategy output module is configured to establish a multi-objective collaborative optimization model with the objectives of maximizing variable current efficiency, minimizing harmonic distortion rate, and minimizing switching loss based on the load demand change trend and the dynamic correlation network, solve the multi-objective collaborative optimization model by using a multi-objective particle swarm optimization algorithm, obtain a variable current strategy solution set, and screen an optimal variable current strategy under a current operating condition by using an analytic hierarchy process.

[0062] The control execution module is configured to convert the optimal variable current strategy into a PWM control signal by using a digital signal processor and adjust a switching state of a bidirectional inverter circuit according to the PWM control signal to realize collaborative variable current control.

[0063] The bidirectional inverter collaborative variable current method and system based on AI prediction provided by the application can greatly improve the variable current efficiency of battery charging of an electric two-wheeled vehicle through AI prediction and multi-objective collaborative optimization, so that grid energy can be more efficiently converted into battery storage capacity, energy loss in the variable current link during charging is reduced, the problems of difficult large current matching in the initial charging stage and invalid energy dissipation in the later stage in the traditional technology are avoided, the charging time is effectively shortened, the charging energy consumption is reduced, the user's charging waiting time is saved, and the charging convenience is improved.

[0064] Through the AI prediction model based on the spatio-temporal graph convolution network, the load demand change trend under different charging scenarios such as grid voltage fluctuation, battery temperature change, and "light storage and charging" integration and regenerative braking energy recovery can be perceived in advance. When the scenario is switched or an external disturbance occurs, the variable current strategy can be quickly adjusted, the charging system can be quickly adapted to different operating conditions, the problem of charging interruption caused by slow response in the traditional technology is avoided, and the stability and compatibility of charging in multiple scenarios are ensured.

[0065] By taking the minimization of harmonic distortion rate as an important optimization target through a multi-objective collaborative optimization model, and combining with the accurate PWM control strategy generated by the digital signal processor, the harmonics generated in the charging current conversion process can be effectively reduced. Not only can it reduce the interference to the stability of the power grid and avoid affecting the normal work of the surrounding electrical equipment, but also can reduce the additional loss inside the current conversion system; at the same time, the battery side voltage, current and switch tube temperature are monitored in real time, and the control parameters are dynamically adjusted to avoid safety hazards such as overcharging and overheating of the battery, and to ensure the stability of the battery charging and discharging process, further prolonging the service life of the battery. BRIEF DESCRIPTION OF DRAWINGS

[0066] In order to more clearly illustrate the technical solutions in the present application or prior art, below will be a brief introduction to the drawings needed to be used in the embodiments or prior art description, obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0067] Figure 1 is a flowchart of a bidirectional inverter collaborative current conversion method based on AI prediction provided by an embodiment of the present application;

[0068] Figure 2 is a structural diagram of a bidirectional inverter collaborative current conversion system based on AI prediction provided by an embodiment of the present application. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical solutions and advantages of the present application more clear, below will be a clear and complete description of the technical solutions in the present application, combined with the drawings in the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the scope of protection of the present application.

[0070] Below will be a description of the bidirectional inverter collaborative current conversion method and system based on AI prediction of the present application. Figures 1-2

[0071] Figure 1 is a flowchart of a bidirectional inverter collaborative current conversion method based on AI prediction provided by an embodiment of the present application;

[0072] As Figure 1 shown, the bidirectional inverter collaborative current conversion method and system based on AI prediction provided by an embodiment of the present application, the execution subject can be a bidirectional inverter collaborative current conversion method based on AI prediction, the method comprising:

[0073] Real-time acquisition of multi-dimensional running data of the bidirectional inverter system. ​

[0074] The multi-dimensional operation data includes DC side voltage, DC side current, AC side voltage, AC side current and switch tube temperature.

[0075] The multi-dimensional operation data is collected by the sensor group, the DC side voltage and current are collected by a Hall voltage sensor and a Hall current sensor, the AC side voltage and current are collected by a voltage transformer and a current transformer, and the switch tube temperature is collected by a patch type thermocouple sensor.

[0076] The multi-dimensional operation data is denoised, the data features are extracted by kernel principal component analysis, and the process includes:

[0077] A multi-dimensional operation data matrix is constructed where n is the number of samples, m is the feature dimension, represents the jth feature value of the ith sample.

[0078] The data matrix is standardized, and the formula is:

[0079]

[0080] In the formula, is the mean of the jth feature, is the standard deviation of the jth feature.

[0081] A Gaussian kernel function is selected to construct a kernel matrix, and the formula is:

[0082]

[0083] In the formula, is the kernel matrix, is the kernel function bandwidth parameter, is the Euclidean distance between the standardized sample i and sample j.

[0084] The kernel matrix is centered, the eigenvalues and eigenvectors of the kernel matrix are calculated, the eigenvectors corresponding to the first k1 largest eigenvalues are selected as principal components, and voltage features, current features and temperature features are obtained.

[0085] A load prediction model based on a spatio-temporal graph convolution network is constructed, data features are combined, data spatio-temporal correlation rules are mined, and future load demand change trends in a preset time period are predicted.

[0086] The process of constructing the load prediction model based on the spatio-temporal graph convolution network includes:

[0087] Collect historical operation data samples of the bidirectional inverter system, set the total number of samples as N, divide each sample into T time steps according to time sequence, each time step contains voltage characteristics, current characteristics and temperature characteristics, construct a space-time data structure, the space-time data structure takes multi-dimensional characteristics of each time step as a graph convolution node, and the association between time steps as a graph convolution edge, forming a time series graph convolution sequence wherein , is a set of graph convolution nodes at time t, corresponding to each dimension of characteristics, is a set of edges between graph convolution nodes at time t, corresponding to the coupling relationship of characteristics, and the load value y corresponding to each time series graph sequence is labeled at the same time, forming

[0088]

[0089] A space-time graph convolution network model is designed, which includes a space-time convolution layer, a time convolution layer and a prediction output layer.

[0090] The space-time convolution layer adopts graph convolution to capture the spatial association between characteristics within the same time step, and the formula is represented as:

[0091]

[0092] In the formula, is the input feature of the graph convolution node i at time t, is the adjacency matrix element of the graph convolution node i and j at time t, is the neighbor set of the graph convolution node i, is the neighbor set of the graph convolution node j, is the spatial convolution weight, is the spatial convolution bias, is the ReLU activation function, is the spatial convolution output.

[0093] The time convolution layer is used to capture the time sequence association between different time steps, and the formula is represented as:

[0094]

[0095] In the formula, L is the size of the time convolution kernel, is one-dimensional convolution, is the spatial convolution weight, is the spatial convolution bias, is the joint output of space-time convolution.

[0096] The prediction output layer is used to output the load demand change trend in the future preset time period.

[0097] Stacking 3-5 layers of spatio-temporal convolutional layers and temporal convolutional layers to extract deep spatio-temporal features, input into the fully connected prediction layer, and the formula is represented as:

[0098]

[0099] In the formula, is the prediction layer weight, is the prediction layer bias, is the deep spatio-temporal feature, is the load prediction value.

[0100] The mean square error is used as the loss function, and the model parameters are updated iteratively by the Adam optimization algorithm until the loss function converges, obtaining the load prediction model.

[0101] An adaptive weighted fusion algorithm is used to fuse multi-dimensional operation data, and a dynamic association network of multi-dimensional data is constructed using a graph neural network to mine the coupling relationship between data.

[0102] The process of constructing a dynamic association network of multi-dimensional data using a graph neural network includes:

[0103] The multi-dimensional data after adaptive weighted fusion is used as a graph neural node, and the set of graph neural nodes is , where corresponds to the i-th dimension of the operation data.

[0104] The similarity matrix between graph neural nodes is calculated as the adjacency matrix A of the graph neural network, and the formula is represented as:

[0105]

[0106] In the formula, is the i-th dimension of the operation data, is the j-th dimension of the operation data, is the number of elements in the graph neural node set, is the k-th dimension of the operation data, and are the maximum and minimum values in the graph neural node set, respectively.

[0107] The graph neural network model is constructed, including the input layer, the graph convolutional layer, and the output layer, and the input layer receives the graph neural node features. The graph convolutional layer aggregates neighbor graph neural node information through the adjacency matrix, and the formula is represented as:

[0108]

[0109] In the formula, is the feature of the i-th layer of graph neural nodes, is the Sigmoid activation function, is a neighbor set of graph neural node i, is a neighbor set of graph neural node j, is the l-th layer weight matrix, is a bias term.

[0110] The output layer outputs the correlation strength between graph neural nodes.

[0111] The graph neural network model is trained to obtain a dynamic correlation network of multi-dimensional data by taking the graph neural node feature reconstruction error as a loss function and optimizing the model parameters.

[0112] According to the load demand change trend and the dynamic correlation network, a multi-objective collaborative optimization model is established to maximize the conversion efficiency, minimize the harmonic distortion rate, and minimize the switching loss, and a multi-objective particle swarm optimization algorithm is used to solve the multi-objective collaborative optimization model to obtain a set of variable conversion strategy solutions. The optimal variable conversion strategy under the current working condition is selected by the analytic hierarchy process.

[0113] The process of establishing a multi-objective collaborative optimization model includes:

[0114] According to the power loss of the bidirectional inverter system, the conversion efficiency is defined as the ratio of the output active power to the input active power, which is expressed as:

[0115]

[0116] In the formula, is the conversion efficiency, is the output active power, is the input active power, is the AC side voltage, is the current effective value, is the power factor, is the DC side voltage, is the DC side current, is the switching tube conduction loss, is the switching loss.

[0117] Based on Fourier decomposition, the AC side current harmonic distortion rate is defined, which is expressed as:

[0118]

[0119] In the formula, is the AC side current harmonic distortion rate, is the fundamental current effective value, is the nth harmonic current effective value, and the voltage harmonic distortion rate is constrained to ensure power quality.

[0120] ​Combining the switching frequency and the switching tube energy consumption characteristics, the switching loss is defined as the product of the switching frequency and the total energy consumption of a single switching cycle, which is expressed as:

[0121]

[0122] In the formula, is the switching loss, is the switching frequency, is the number of switching tubes, and is the turn-on and turn-off energy consumption of the ith switching tube with the DC side current (through device manual curve fitting function relationship).

[0123] The constraint coupling modeling combines the topology characteristics and operation safety requirements of the bidirectional inverter system, establishes multiple constraint coupling relationships, including electrical constraints, control constraints and coupling constraints.

[0124] Among them, the electrical constraints include: the DC side voltage satisfies the capacitor voltage withstand limit , the AC side voltage satisfies the grid access standard , the switching tube current , is the rated current of the switching tube.

[0125] The control constraints include: the modulation coefficient (PWM modulation wave and carrier amplitude ratio) satisfies , to ensure the stability of the modulation. The switching frequency satisfies , to balance the switching loss and the filtering effect.

[0126] The coupling constraints include: based on the feature correlation strength of the dynamic association network output, the coupling relationship between the decision variable and the target is established, which is expressed as:

[0127]

[0128] In the formula, is the decision variable vector, is the target vector, is the association strength matrix element of the association network output.

[0129] The optimization function integration adopts the combination of the weighting coefficient method and the target normalization to construct a unified multi-objective optimization function, which is expressed as:

[0130]

[0131] In the formula, , and These are the maximum converter efficiency, the upper limit of allowable harmonic distortion rate, and the upper limit of allowable switching loss, respectively. , For the target weight (satisfying) (Determined through the analytic hierarchy process), while embedding multiple constraint coupling relationships to obtain a multi-objective collaborative optimization model.

[0132] The process of solving a multi-objective collaborative optimization model using a multi-objective particle swarm optimization algorithm includes:

[0133] N1 particles are randomly generated, each particle corresponds to a set of decision variables that satisfy the constraints. At the same time, the velocity vector of the particles and the individual optimal solution are initialized, and an external archive set is set to store the non-dominated solutions.

[0134] S11. Calculate the objective function value of each particle, compare the particle with its own optimal solution, and update the optimal solution if the particle dominates the optimal solution.

[0135] S12. Compare the particle with solutions in the external archive set. If the particle is a non-dominated solution and is not in the archive set, add it to the archive set. If the size of the archive set exceeds a preset value, use crowding sorting to delete redundant solutions.

[0136] S13. Update particle velocity and position using the following formulas:

[0137]

[0138]

[0139] In the formula, for The velocity of particle i at time i for The velocity of particle i at time i For inertial weights, and As a learning factor, and A random number in the interval [0,1]. Let be the individual optimal solution for particle i. The global bootstrap solution is randomly selected from an external archive set. for The position of particle i at time i for The position of particle i at time i.

[0140] Repeat steps S1 to S3 until the number of iterations reaches the preset value, and output the solution in the external archive set as the solution set of the variable flow strategy.

[0141] The process of selecting the optimal current conversion strategy using the analytic hierarchy process includes:

[0142] Using converter efficiency, harmonic distortion rate, and switching loss as the criteria layer, and the strategies in the converter strategy solution set as the scheme layer, a judgment matrix is ​​constructed. ,in This indicates the importance of criterion i relative to criterion j.

[0143] Calculate the largest eigenvalue and corresponding eigenvector of the judgment matrix, and perform a consistency check. The formula is as follows:

[0144]

[0145]

[0146] In the formula, CI is the consistency index. To determine the largest eigenvalue of a matrix, RI is a random consistency index. If the matrix satisfies consistency, then the matrix is ​​determined to be consistent.

[0147] The performance of the strategy under each criterion is normalized to obtain the normalized matrix. ( (Number of strategies), calculate the overall score. , The weight of the l-th criterion, Given the performance value of the k-th conversion strategy after normalization under the l-th criterion, select the strategy with the highest score as the optimal conversion strategy.

[0148] The optimal converter strategy is converted into a PWM (Pulse Width Modulation) control signal using a digital signal processor, and the switching state of the bidirectional inverter circuit is adjusted according to the PWM control signal to achieve coordinated converter control. The process includes:

[0149] S21. Establish the discrete state-space model of the bidirectional inverter circuit, expressed by the formula:

[0150]

[0151]

[0152] In the formula, for The state vector at any given time (including inductor current and capacitor voltage). For switch control signals, For disturbances (including load changes). Let A, B, C, and D be the output vectors, and let A, B, C, and D be the system matrices. For measuring noise.

[0153] S22, define the target function, the formula is expressed as:

[0154]

[0155] In the formula, J is the target function value of the model predictive control, is the prediction time domain, is the control time domain, is the predicted value of the system output at the future time at the time , is the reference output at the time , is the predicted value of the control input at the future time at the time , Q and R are weighting matrices.

[0156] S23, in each control period, solve the quadratic programming problem with constraints to obtain the optimal control sequence , and output to the switch tube as the current time PWM control signal.

[0157] In the next control period, update the state vector, repeat steps S21-S23 to realize dynamic adjustment.

[0158] In summary, the embodiment provides a bidirectional inverter collaborative conversion method based on AI prediction. Through AI prediction and multi-objective collaborative optimization, the conversion efficiency of electric two-wheeler battery charging can be greatly improved. The grid energy can be more efficiently converted into battery storage power, reducing energy loss in the conversion link during charging, avoiding the problems of large current matching difficulty in the initial charging stage and invalid energy dissipation in the later stage in traditional technology, thereby effectively shortening the charging time, reducing the charging energy consumption, saving the user's charging waiting time, and improving the charging convenience.

[0159] Through the AI prediction model based on the space-time graph convolution network, the load demand change trend under different charging scenarios such as grid voltage fluctuation, battery temperature change, "light storage and charging" integration, and regenerative braking energy recovery can be perceived in advance. In the scene switching or external disturbance, the conversion strategy can be quickly adjusted to make the charging system quickly adapt to different working conditions, avoid the charging interruption problem caused by slow response in traditional technology, and ensure the stability and compatibility of charging in multiple scenarios.

[0160] By minimizing the harmonic distortion rate as an important optimization target through a multi-objective collaborative optimization model, combined with the accurate PWM control strategy generated by the digital signal processor, the harmonics generated in the charging current conversion process can be effectively reduced. Not only can it reduce the interference to the stability of the power grid and avoid affecting the normal work of the surrounding electrical equipment, but also can reduce the additional loss inside the current conversion system; at the same time, the battery side voltage, current and switch tube temperature are monitored in real time, and the control parameters are dynamically adjusted to avoid safety hazards such as overcharging and overheating of the battery, and to ensure the stability of the battery charging and discharging process, further prolonging the service life of the battery.

[0161] Based on the same overall inventive concept, the application also protects a bidirectional inverter collaborative current conversion system based on AI prediction. The following describes a bidirectional inverter collaborative current conversion system based on AI prediction provided by the application. The bidirectional inverter collaborative current conversion system described below can be mutually corresponding and referred to with the bidirectional inverter collaborative current conversion method described above.

[0162] Figure 2 is a structural schematic diagram of a bidirectional inverter collaborative current conversion system based on AI prediction provided by an embodiment of the application.

[0163] As shown in Figure 2 , a bidirectional inverter collaborative current conversion system based on AI prediction includes a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor includes a data acquisition module, a feature extraction module, a load prediction module, an association mining module, a strategy output module, and a control execution module.

[0164] The data acquisition module is used to acquire multi-dimensional running data of the bidirectional inverter system in real time.

[0165] The feature extraction module is used to denoise the multi-dimensional running data and extract data features using kernel principal component analysis.

[0166] The load prediction module is used to construct a load prediction model based on a spatio-temporal graph convolution network, combine data features, mine data spatio-temporal association rules, and predict the load demand change trend in a future preset time period.

[0167] The association mining module is used to fuse multi-dimensional running data using an adaptive weighted fusion algorithm, construct a dynamic association network of multi-dimensional data using a graph neural network, and mine the coupling relationship between data.

[0168] The strategy output module is used for establishing a multi-objective collaborative optimization model with the maximum variable current efficiency, the minimum harmonic distortion rate and the minimum switching loss as the targets according to the load demand change trend and the dynamic correlation network, solving the multi-objective collaborative optimization model by using a multi-objective particle swarm optimization algorithm, obtaining a variable current strategy solution set, and screening the optimal variable current strategy under the current working condition by using the analytic hierarchy process.

[0169] The control execution module is used for converting the optimal variable current strategy into a PWM control signal by using a digital signal processor, adjusting the switching state of the bidirectional inverter circuit according to the PWM control signal, and realizing the collaborative variable current control.

[0170] Through the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course, can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, and the computer software product can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the method described in each embodiment or some parts of the embodiment.

[0171] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A bidirectional inverter coordination current conversion method based on AI prediction, characterized in that, The application relates to a method for realizing collaborative variable-current control of a bidirectional inverter system. Real-time acquisition of multidimensional operation data of the bidirectional inverter system; Denoising processing of the multidimensional operation data, and extraction of data characteristics by using kernel principal component analysis; Construction of a load prediction model based on a spatio-temporal graph convolution network, combination of the data characteristics, mining of spatio-temporal correlation rules of the data, and prediction of a load demand change trend in a future preset time period; Adaptive weighted fusion algorithm is used to fuse the multidimensional operation data, a graph neural network is used to construct a dynamic correlation network of the multidimensional data, and coupling relationships between the data are mined; According to the load demand change trend and the dynamic correlation network, a multi-objective collaborative optimization model is established, with the maximum variable-current efficiency, the minimum harmonic distortion rate and the minimum switching loss as the targets; A multi-objective particle swarm optimization algorithm is used to solve the multi-objective collaborative optimization model, a variable-current strategy solution set is obtained, and the optimal variable-current strategy under the current working condition is screened by using an analytic hierarchy process; The optimal variable-current strategy is converted into a PWM control signal by using a digital signal processor, the switching state of the bidirectional inverter circuit is adjusted according to the PWM control signal, and collaborative variable-current control is realized. The process of adjusting the switching state of the bidirectional inverter circuit according to the PWM control signal comprises the following steps: S21, a discrete state space model of the bidirectional inverter circuit is established, and the formula is as follows: wherein is the time instant state vector, is the switching control signal, is the disturbance, is the output vector, A, B, C, D are system matrices, is the measurement noise; S22, a target function is defined, and the formula is as follows: wherein J is a value of an objective function of model predictive control, is a prediction horizon, is a control horizon, is a predicted value of a system output at a future time from a time t, is a predicted value of a system output at a future time from a time t, is a reference output at a time t, is a predicted value of a control input at a future time from a time t, Q, R are weighting matrices; S23, in each control cycle, solve the constrained quadratic programming problem to obtain the optimal control sequence , the PWM control signal of the current time is output to the switch tube. , the PWM control signal of the current time is output to the switch tube. In the next control period, the state vector is updated, and steps S21-S23 are repeated to realize dynamic adjustment.

2. The AI prediction-based bidirectional inversion coordination and current conversion method according to claim 1, characterized in that, The multidimensional operation data comprises a DC side voltage, a DC side current, an AC side voltage, an AC side current and a switching tube temperature. 3.The AI prediction based bidirectional inverter coordination and variable current method according to claim 1, characterized in that, The process of extracting data characteristics by using kernel principal component analysis comprises the following steps: A multidimensional operation data matrix is constructed, and the data matrix is standardized; A Gaussian kernel function is selected to construct a kernel matrix, and the kernel matrix is centralized; The eigenvalues and eigenvectors of the kernel matrix are calculated, the eigenvectors corresponding to the first k1 largest eigenvalues are selected as principal components, and voltage characteristics, current characteristics and temperature characteristics are obtained.

4. The AI prediction-based bidirectional inversion coordination and current conversion method according to claim 1, characterized in that, The process of constructing the load prediction model based on the spatio-temporal graph convolution network comprises the following steps: Historical operation data samples of the bidirectional inverter system are collected, each sample is divided into T time steps according to a time sequence, and each time step contains voltage characteristics, current characteristics and temperature characteristics; A spatio-temporal data structure is constructed, the spatio-temporal data structure takes the multidimensional characteristics of each time step as a graph convolution node, and the correlation between time steps as a graph convolution edge, a time sequence graph convolution sequence is formed, and each time sequence graph sequence is labeled with a corresponding load value to form a sample set; A spatio-temporal graph convolution network model is designed, the spatio-temporal graph convolution network model comprises a spatio-temporal convolution layer, a time convolution layer and a prediction output layer; the spatio-temporal convolution layer uses graph convolution to capture the spatial correlation between characteristics in the same time step; the time convolution layer is used to capture the time sequence correlation between different time steps; and the prediction output layer is used to output a load demand change trend in a future preset time period; The spatio-temporal convolution layer and the time convolution layer are stacked to extract deep spatio-temporal characteristics, and the deep spatio-temporal characteristics are input into a full connection prediction layer; Adopting mean square error as loss function, the model parameters are iteratively updated by Adam optimization algorithm until the loss function converges, and the load prediction model is obtained.

5. The AI prediction-based bidirectional inverter coordination and variable current method according to claim 1, characterized in that, The process of constructing a multi-dimensional data dynamic correlation network by using a graph neural network includes: Adaptive weighted fusion of multi-dimensional data as a graph neural node; Calculate the similarity matrix between graph neural nodes as the adjacency matrix of the graph neural network; Construct a graph neural network model, including an input layer, a graph convolution layer, and an output layer, the input layer receives the graph neural node features; the graph convolution layer aggregates neighbor graph neural node information through the adjacency matrix; the output layer outputs the correlation strength between graph neural nodes; Train the graph neural network model, taking the graph neural node feature reconstruction error as the loss function, and optimize the model parameters to obtain the dynamic correlation network of multi-dimensional data.

6. The AI prediction-based bidirectional inversion coordination and current conversion method according to claim 1, characterized in that, The process of establishing a multi-objective collaborative optimization model includes: According to the power loss of the bidirectional inverter system, the conversion efficiency is defined as the ratio of the output active power to the input active power; Based on Fourier decomposition, the AC side current harmonic distortion rate is defined; Combined with the switching frequency and the energy consumption characteristics of the switching tube, the switching loss is defined as the product of the switching frequency and the total energy consumption in a single switching cycle; The constraint coupling modeling combines the topology characteristics and operation safety requirements of the bidirectional inverter system to establish a multi-constraint coupling relationship, including electrical constraints, control constraints, and coupling constraints; The optimization function integration adopts a weighted coefficient method combined with target normalization to construct a unified multi-objective optimization function, which is embedded in the multi-constraint coupling relationship to obtain a multi-objective collaborative optimization model.

7. The AI prediction-based bidirectional inversion coordination and current conversion method according to claim 1, characterized in that, The process of solving the multi-objective collaborative optimization model by using a multi-objective particle swarm optimization algorithm includes: Randomly generate N1 particles, each particle corresponds to a set of decision variables that meet the constraint conditions, and initialize the particle's speed vector and individual optimal solution, and set up an external archive set to store non-dominated solutions; S11, calculate the objective function value of each particle, compare the particle with its individual optimal solution, if the particle dominates the individual optimal solution, update the individual optimal solution; S12, compare the particle with the solutions in the external archive set, if the particle is a non-dominated solution and is not in the archive set, add it to the archive set; if the size of the archive set exceeds the preset value, delete the redundant solutions by using the crowding degree sorting; S13, update the particle speed and position; Repeat steps S11-S13 until the iteration count reaches the preset value, and output the solutions in the external archive set as the conversion strategy solution set.

8. The AI prediction-based bidirectional inversion coordination and current conversion method according to claim 1, characterized in that, The process of selecting the optimal conversion strategy by using the analytic hierarchy process includes: Taking the conversion efficiency, harmonic distortion rate and switching loss as the criterion layer, and the strategies in the conversion strategy solution set as the scheme layer, a judgment matrix is constructed; Calculate the maximum eigenvalue and corresponding eigenvector of the judgment matrix, and perform consistency check; The performance of the strategy under each criterion is normalized to obtain a normalized matrix, and the comprehensive score is calculated, and the strategy with the highest score is selected as the optimal conversion strategy. 9.A bidirectional inverter coordinated variable current system based on AI prediction, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the AI prediction-based bidirectional inverter collaborative conversion method of any one of claims 1-8.

Images