Deep learning based adaptive control method for mobile phone ultra-fast wideband converter

Abstract

The application provides a mobile phone extreme charging wideband variable current adaptive regulation method based on deep learning, and relates to the technical field of mobile phone extreme charging. The method comprises collecting multi-dimensional operation data of the mobile phone extreme charging scene. A local linear embedding algorithm is used for feature extraction. A state of charge evaluation model based on a deep belief network is constructed to evaluate the operation state of the mobile phone extreme charging system in real time. An echo state network is used to construct a dynamic correlation relationship model of multi-dimensional data. A multi-objective optimization model is established, and a multi-objective artificial bee colony algorithm is used to solve the multi-objective optimization model to select the optimal variable current regulation strategy of the current charging condition. The variable current parameters are adjusted. The application is constructed through deep fusion of multi-dimensional data and a deep learning model, and the variable current regulation strategy is optimized by combining an intelligent algorithm, so that the charging efficiency is improved, the battery temperature rise and circuit loss are reduced, the intelligence and adaptability of the variable current regulation are enhanced, and the battery safety is ensured and the system service life is prolonged.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of mobile phone fast charging, and in particular to a mobile phone fast charging wide-frequency adaptive regulation method based on deep learning. BACKGROUND

[0002] With the diversified development of smart phone functions, users' demand for charging speed is increasingly urgent, and mobile phone fast charging technology has emerged as the times require. However, the current mobile phone fast charging field faces many technical bottlenecks. In terms of charging efficiency, the traditional fast charging scheme mostly adopts fixed variable conversion parameters, which cannot be adjusted in real time according to dynamic factors such as battery state and environmental conditions, resulting in large energy loss during the charging process and difficulty in continuously maintaining a high level of charging efficiency. For example, when the remaining battery capacity (SOC) is low, if the charging strategy of high voltage and low current is still adopted, the charging time will be greatly increased; and when the battery is close to the full charge state, high current charging is prone to cause battery overheating and other problems.

[0003] The battery internal resistance changes with the charging process and environmental temperature, and the traditional scheme lacks real-time and accurate monitoring of the battery internal resistance and variable conversion adjustment based thereon, so that the battery is prone to local overheating during fast charging, which will accelerate the battery aging and reduce the battery cycle life. At the same time, the switching tube loss, inductance loss and other circuit losses in the charging circuit have not been effectively optimized, which not only affects the charging efficiency, but also generates additional heat, further aggravating the battery temperature rise problem. Therefore, there is an urgent need for a mobile phone fast charging wide-frequency adaptive regulation method based on deep learning to solve the above technical problems. SUMMARY

[0004] The application provides a mobile phone fast charging wide-frequency adaptive regulation method based on deep learning to solve the defects in the prior art.

[0005] The application provides a mobile phone fast charging wide-frequency adaptive regulation method based on deep learning, comprising:

[0006] Real-time collection of multi-dimensional running data of the mobile phone fast charging scene, the multi-dimensional running data including charging circuit parameters, battery state data and environmental condition data.

[0007] Denoising of the multi-dimensional running data by using an empirical mode decomposition method, and feature extraction by using a local linear embedding algorithm to obtain circuit features, battery features and environmental features.

[0008] Construction of a charging state evaluation model based on a deep belief network, real-time evaluation of the running state of the mobile phone fast charging system according to the circuit features, battery features and environmental features, and preliminary variable conversion parameter adjustment according to the running state and a preset threshold rule.

[0009] The particle filtering algorithm is adopted to fuse multi-dimensional operation data, and an echo state network is used to construct a dynamic correlation relationship model of the multi-dimensional data.

[0010] A multi-objective optimization model of comprehensive charging efficiency, battery temperature rise and circuit loss is established according to the dynamic correlation relationship model, a multi-objective artificial bee colony algorithm is adopted to solve the multi-objective optimization model, a variable flow regulation strategy solution set is obtained, and the optimal variable flow regulation strategy of the current charging condition is selected through an intuitionistic fuzzy decision theory.

[0011] The optimal variable flow regulation strategy is converted into a control signal through a microcontroller, and the variable flow parameters of the polar charging system are adjusted in combination with a feedforward-feedback compound control.

[0012] According to the mobile phone polar charging broadband variable flow adaptive regulation method based on deep learning provided by the application, the charging circuit parameters include input voltage, output current, switch tube conduction time, filter capacitor voltage and inductance current.

[0013] According to the mobile phone polar charging broadband variable flow adaptive regulation method based on deep learning provided by the application, the process of feature extraction by using the local linear embedding algorithm includes:

[0014] A multi-dimensional operation data matrix D is constructed, and the data matrix D is standardized to eliminate the dimensional differences of different features.

[0015] Based on the Euclidean distance, the individual nearest neighbor sample of each sample is determined, a local neighborhood matrix is constructed, and the local reconstruction weight of the sample and the nearest neighbor is calculated.

[0016] By minimizing the reconstruction error of the sample and the nearest neighbor in the low-dimensional feature space, the high-dimensional standardized data is mapped to the low-dimensional space, and a low-dimensional feature matrix with a dimension of p is obtained, and each row in the low-dimensional feature matrix corresponds to a circuit feature, battery feature and environmental feature combined vector of a sample.

[0017] According to the mobile phone polar charging broadband variable flow adaptive regulation method based on deep learning provided by the application, the process of constructing a charging state evaluation model based on a deep belief network includes:

[0018] The historical operation data samples of the mobile phone polar charging system are collected to construct a feature matrix, and the charging state labels are corresponded, the charging state labels include normal charging, mild abnormality, moderate abnormality and severe abnormality.

[0019] The deep belief network structure is constructed, including a deep belief input layer, a restricted Boltzmann machine hidden layer and a deep belief output layer, the node number of the deep belief input layer is equal to the feature dimension, the deep belief output layer adopts a softmax activation function, and the node number is equal to the state of charge category number.

[0020] The energy function and the probability distribution of the restricted Boltzmann machine are defined.

[0021] The contrast divergence algorithm is adopted to pre-train the restricted Boltzmann machine of each layer, the weight and the bias are optimized layer by layer, the deep belief network is adjusted through the back propagation algorithm, and the cross-entropy loss function is minimized, so that the state of charge evaluation model is obtained.

[0022] According to the mobile phone extreme charging broadband variable flow adaptive regulation method based on deep learning provided by the application, the process of fusing multi-dimensional running data by adopting a particle filtering algorithm comprises:

[0023] N particles are generated, each particle represents an initial estimated value of the multi-dimensional running data, and the particle weight is initialized.

[0024] A state transition model is constructed, each particle at time t is predicted, and a predicted particle is obtained.

[0025] According to the multi-dimensional data actually collected at time t, the likelihood probability of each predicted particle is calculated, and the particle weight is updated.

[0026] A weight normalization factor is calculated, the weight is normalized, if the number of effective particles is less than a preset threshold, a system resampling method is used to screen high-weight particles and replace low-weight particles, and the fused data at time t is obtained.

[0027] According to the mobile phone extreme charging broadband variable flow adaptive regulation method based on deep learning provided by the application, the process of constructing a dynamic association relationship model of multi-dimensional data by using an echo state network comprises:

[0028] The multi-dimensional running data fused by the particle filtering is divided according to time sequence to form an input sequence with a length of T.

[0029] An echo state network structure is constructed, including an echo state input layer, a reservoir and an echo state output layer, the echo state input layer is used for receiving the multi-dimensional fused data of the time sequence input, the reservoir is used for performing nonlinear mapping and dynamic response on the input time sequence data through internal random connected neurons, and the echo state output layer is used for outputting the reconstruction result of the multi-dimensional data according to the state vector of the reservoir.

[0030] The ridge regression algorithm is used to train the output layer weight, and the prediction error is minimized, and after the training is completed, the reserve pool state vector at each time is taken as a node, and the time sequence correlation strength between the nodes is taken as a connection weight, and a dynamic correlation relationship model of multi-dimensional data is constructed.

[0031] According to the mobile phone extreme charging broadband variable current adaptive regulation method based on deep learning provided by the application, the process of establishing a multi-objective optimization model of comprehensive charging efficiency, battery temperature rise and circuit loss according to the dynamic correlation relationship model comprises:

[0032] The charging efficiency is defined as the ratio of the actual energy absorbed by the battery to the input energy of the charging system.

[0033] The battery temperature rise model is established based on the battery heat balance equation.

[0034] The circuit loss is defined as the sum of the conduction loss of the switch tube, the switching loss and the loss of the passive element.

[0035] Taking maximizing the charging efficiency, minimizing the battery temperature rise and minimizing the circuit loss as the target, a multi-objective optimization model is established in combination with the data constraint in the dynamic correlation relationship model.

[0036] According to the mobile phone extreme charging broadband variable current adaptive regulation method based on deep learning provided by the application, the process of solving the multi-objective optimization model by using the multi-objective artificial bee colony algorithm to obtain the variable current regulation strategy solution set comprises:

[0037] Initialize the bee colony and the food source, set the bee colony size to S, each food source corresponds to a group of optimization variables, and randomly generate initial food sources.

[0038] The target function value of each food source is calculated, the food sources are non-dominantly sorted based on the Pareto dominance relationship, and the non-dominant solution set is determined.

[0039] For each food source, a new food source is generated through neighborhood search.

[0040] The selection probability of each food source is calculated, the observation bee selects the food source according to the probability and repeats the neighborhood search, and the food source is updated.

[0041] If a food source is not updated for H consecutive times, it is determined to be discarded, and the scout bee randomly generates a new food source to replace the discarded food source.

[0042] The non-dominant solution of each iteration is recorded, and when the iteration number reaches the preset maximum value, the iteration is stopped, and the final non-dominant solution set is taken as the variable current regulation strategy solution set.

[0043] According to the mobile phone extreme charging broadband variable current adaptive regulation method based on deep learning provided by the application, the process of selecting the optimal variable current regulation strategy of the current charging working condition through the intuitionistic fuzzy decision theory comprises:

[0044] determining a decision factor set and a comment set, the factor set , for charging efficiency, for battery temperature rise, for circuit loss, the comment set , excellent, good, general, poor.

[0045] for the jth strategy in the strategy solution set of the converter regulation, according to the performance of the strategy on the ith factor, determining the membership degree and the non-membership degree, and forming a matrix.

[0046] according to the current working condition requirement, using the analytic hierarchy process to calculate a weight vector.

[0047] using an intuitionistic fuzzy weighted average operator, calculating a comprehensive evaluation value for the jth strategy.

[0048] based on a score function, ranking all strategies, and the strategy with the highest score is the optimal converter regulation strategy for the current working condition.

[0049] According to the mobile phone extreme charging broadband converter self-adaptive regulation method based on deep learning provided by the application, the process of combining feedforward-feedback composite control to adjust the converter parameters of the extreme charging system includes:

[0050] analyzing the optimal converter regulation strategy and extracting key parameters, the key parameters including a target switching frequency, a target duty cycle and a target input voltage.

[0051] based on the change trend of the battery remaining power and the environmental temperature, calculating a feedforward control amount.

[0052] using a proportional-integral-derivative controller, taking the deviation between an output voltage and a target voltage as input, calculating a feedback control amount.

[0053] superimposing the feedforward control amount and the feedback control amount to obtain a total control amount, a microcontroller generates a PWM control signal according to u to adjust the on and off timing of the switching tube in the broadband converter circuit, and realize the control of the converter parameters.

[0054] The mobile phone fast charging broadband variable frequency adaptive regulation method based on deep learning provided by the application can accurately evaluate the running state of the charging system in real time by using empirical mode decomposition and local linear embedding algorithm to denoise and extract features of multidimensional operation data, and can preliminarily adjust the variable frequency parameters according to the state. At the same time, the particle filtering algorithm is used to fuse multidimensional data, and the echo state network is used to construct a dynamic correlation model, and the multi-objective artificial bee colony algorithm is used to solve the multi-objective optimization model of comprehensive charging efficiency, battery temperature rise and circuit loss, so as to obtain the optimal variable frequency regulation strategy, greatly improve the charging efficiency, and adaptively adjust the switching frequency, duty ratio and input voltage and other parameters under different battery remaining capacity and environmental temperature conditions, effectively shorten the charging time, and meet the urgent needs of users for fast charging.

[0055] By establishing a battery temperature rise model based on the heat conduction equation and a circuit loss model integrating the losses of various elements in the circuit, the battery temperature rise and the circuit loss are taken as important indicators of multi-objective optimization, which effectively reduces the battery temperature rise and the circuit loss while improving the charging efficiency, greatly reduces the risk of battery overheating and circuit element aging, prolongs the service life of the battery and the charging system, and improves the safety and reliability of the mobile phone fast charging.

[0056] The echo state network is used to construct a dynamic correlation model of multidimensional data, which can more accurately capture the dynamic correlation between multidimensional data. At the same time, the multi-objective artificial bee colony algorithm can obtain the optimal variable frequency regulation strategy in a shorter time compared with the non-dominated sorting genetic algorithm, which meets the real-time requirements of the mobile phone fast charging system. The intuitionistic fuzzy decision theory is used to select the optimal variable frequency regulation strategy of the current charging condition, and the feedforward-feedback composite control is used to adjust the variable frequency parameters, so that the variable frequency regulation has high intelligence and adaptability, and can achieve optimal charging performance under various complex charging conditions. BRIEF DESCRIPTION OF DRAWINGS

[0057] In order to more clearly illustrate the technical solutions in the application or prior art, the following will briefly introduce 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 application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0058] Figure 1 is a flowchart of the mobile phone fast charging broadband variable frequency adaptive regulation method based on deep learning provided by the embodiment of the application. DETAILED DESCRIPTION

[0059] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below in conjunction with the drawings in the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work belong to the protection scope of the present application.

[0060] The present application is described below in conjunction with Figure 1 A deep learning-based adaptive control method for mobile phone extreme charging broadband variable current is described.

[0061] Figure 1 The present application is described below in conjunction with

[0062] As Figure 1 shown, the deep learning-based adaptive control method for mobile phone extreme charging broadband variable current provided by the embodiments of the present application comprises:

[0063] Real-time acquisition of multi-dimensional running data of the mobile phone extreme charging scene, the multi-dimensional running data comprising charging circuit parameters, battery state data and environmental working condition data.

[0064] The charging circuit parameters comprise input voltage, output current, switch tube conduction time, filter capacitor voltage and inductance current. The battery state data comprises battery remaining capacity, battery internal resistance, battery terminal voltage and battery cycle number. The environmental working condition data comprises environmental temperature, environmental humidity and charging interface contact resistance.

[0065] The charging interface contact resistance is calculated by acquiring the ratio of voltage drop and current in the charging loop, and the formula is expressed as:

[0066]

[0067] In the formula, is the charging interface contact resistance, is the voltage drop in the charging loop, is the charging current.

[0068] The empirical mode decomposition method is used to denoise the multi-dimensional running data, and the local linear embedding algorithm is used for feature extraction to obtain circuit features, battery features and environmental features.

[0069] The process of feature extraction using the local linear embedding algorithm comprises:

[0070] Constructing a multi-dimensional running data matrix Wherein m is the sample number, n is the feature dimension, represents the jth feature value of the ith sample.

[0071] The data matrix D is standardized to eliminate the difference in the dimensions of different features, and the standardization formula is:

[0072]

[0073] In the formula, is the standardized eigenvalue, is the mean of the jth feature, is the standard deviation of the jth feature.

[0074] Based on the Euclidean distance, determine the nearest neighbor sample of each sample, construct a local neighborhood matrix, and calculate the local reconstruction weight of the sample and the neighbor.

[0075] By minimizing the reconstruction error of the sample and the neighbor in the low-dimensional feature space, the high-dimensional standardized data is mapped to the low-dimensional space, and a low-dimensional feature matrix with dimension p ( ) is obtained Each row corresponds to a combination of circuit features, battery features, and environmental features of a sample.

[0076] A charging state evaluation model based on a deep belief network is constructed, which can evaluate the running state of the mobile phone extreme charging system in real time according to the circuit features, battery features and environmental features, and adjust the preliminary current conversion parameters according to the running state and the preset threshold rule.

[0077] The process of constructing a charging state evaluation model based on a deep belief network includes:

[0078] Collect historical running data samples of the mobile phone extreme charging system, a total of s samples, each sample contains features, and a feature matrix is constructed, where is the feature vector of the ith sample, corresponding to the charging state label , representing normal charging, mild abnormality, moderate abnormality and severe abnormality, respectively.

[0079] A deep belief network structure is constructed, including 1 deep belief input layer, 2-3 restricted Boltzmann machine hidden layers and 1 deep belief output layer. The number of nodes of the deep belief input layer is equal to the feature dimension, the deep belief output layer adopts a softmax activation function, and the number of nodes is equal to the number of charging state categories 4.

[0080] The energy function and probability distribution of the restricted Boltzmann machine are defined, and the energy function formula is:

[0081]

[0082] In the formula, Let be the energy function of the restricted Boltzmann machine, v be the visible layer vector (corresponding to the input features), h be the hidden layer vector, W be the connection weight matrix between the visible and hidden layers, a be the visible layer bias vector, and b be the hidden layer bias vector.

[0083] The activation probability of hidden layer nodes and the reconstruction probability of visible layer nodes are respectively:

[0084]

[0085]

[0086] In the formula, Given the visible layer vector v, let v be the probability that the j-th node in the hidden layer is activated (with a value of 1). Let be the bias of the j-th node in the hidden layer. Let be the value of the i-th node in the visible layer. Let be the connection weight between the i-th node in the visible layer and the j-th node in the hidden layer. The number of visible layer nodes. Given a hidden layer vector h, let h be the probability that the i-th node in the visible layer is activated (with a value of 1). Let be the bias of the i-th node in the visible layer. Let j be the value of the j-th node in the hidden layer. It is the sigmoid activation function. This represents the number of nodes in the hidden layer.

[0087] The contrastive divergence algorithm is used to pre-train the restricted Boltzmann machine at each layer, optimizing the weights and biases layer by layer. The deep belief network is then adjusted using the backpropagation algorithm, and the cross-entropy loss function is minimized. The formula is as follows:

[0088]

[0089] In the formula, Let be the true label (0 or 1) of the i-th sample corresponding to the i-th class of states. Let be the probability that the i-th sample is predicted to be in the c-th state. After training, the charging state evaluation model is obtained.

[0090] A particle filter algorithm is used to fuse multi-dimensional operational data, and an echo state network is used to construct a dynamic correlation model of the multi-dimensional data.

[0091] The process of fusing multi-dimensional operational data using the particle filter algorithm includes:

[0092] Generate N particles Each particle represents an initial estimate of the multi-dimensional operational data, and the particle weights are initialized to... , .

[0093] Construct a state transition model for each particle at time t. Make predictions and obtain predicted particles. The state transition model formula is expressed as:

[0094]

[0095] In the formula, Let A be the state vector at time t, A be the state transition matrix, and B be the control matrix. Let be the state vector at time t-1. This is the control variable at time t-1. It is Gaussian noise.

[0096] Based on the multi-dimensional data actually collected at time t, the likelihood probability of each predicted particle is calculated, and the particle weights are updated. The particle weight update formula is expressed as:

[0097]

[0098] The likelihood probability follows a Gaussian distribution, expressed by the formula:

[0099]

[0100] In the formula, Let be the weight of the i-th particle at time t. Let be the weight of the i-th particle at time t-1. Let be the likelihood probability of the i-th predicted particle at time t. Dimensions of the observed data The data represents the multi-dimensional data actually collected at time t. Let be the predicted state of the i-th particle at time t, C be the observation matrix, and R be the observation noise covariance matrix.

[0101] Calculate the weight normalization factor Normalize the weights; if the number of effective particles... If a preset threshold is used, a system resampling method is employed to filter high-weight particles and replace low-weight particles.

[0102] The data after fusion at time t is a weighted average of all particles, expressed by the formula:

[0103]

[0104] In the formula, Let be the system state estimate after fusion at time t. Let be the weight of the i-th particle at time t. Let t be the state of the i-th particle at time t.

[0105] The process of constructing a dynamic correlation model of multi-dimensional data using echo state networks includes:

[0106] The multi-dimensional operational data after particle filtering fusion is divided into time series to form an input sequence of length T. ,in Let be the fused data vector at time t. For data dimensions.

[0107] An echo-state network structure is constructed, comprising an echo-state input layer, a reservoir, and an echo-state output layer. The echo-state input layer receives multi-dimensional fused data from time-series inputs and maps it to the input space of the reservoir. The reservoir performs nonlinear mapping and dynamic response on the input time-series data through internally randomly connected neurons, capturing the short-term dependencies and long-term time-series characteristics of the data. The echo-state output layer outputs the reconstruction results of the multi-dimensional data based on the state vector of the reservoir, with the number of nodes equal to the data dimension.

[0108] Define the reserve pool state update formula:

[0109]

[0110] In the formula, Let be the state vector of the reservoir at time t. This is the connection weight matrix from the echo state input layer to the reservoir. The weight matrix for neuron connections within the reservoir. Let the reservoir bias vector be... It is the hyperbolic tangent activation function.

[0111] The output formula for the echo state output layer is:

[0112]

[0113] In the formula, Let be the network output vector at time t. The output layer weight matrix (trained via linear regression). This is the output layer bias vector.

[0114] The ridge regression algorithm is used to train the output layer weights to minimize the prediction error. After training, the state vectors of the reservoir at each time step are... As nodes, the temporal correlation strength between nodes (calculated) and Using the Pearson correlation coefficient as the connection weight, a dynamic correlation model of multi-dimensional data is constructed.

[0115] Based on the dynamic correlation model, a multi-objective optimization model is established to integrate charging efficiency, battery temperature rise, and circuit loss. The multi-objective artificial bee colony algorithm is used to solve the multi-objective optimization model to obtain the solution set of the current conversion control strategy. Finally, the optimal current conversion control strategy for the current charging condition is selected through intuitionistic fuzzy decision theory.

[0116] The process of establishing a multi-objective optimization model for comprehensive charging efficiency, battery temperature rise, and circuit loss based on a dynamic correlation model includes:

[0117] Charging efficiency is defined as the ratio of the actual energy absorbed by the battery to the energy input to the charging system, and the calculation formula is:

[0118]

[0119] In the formula, For charging efficiency, For charging time, , Let be the battery terminal voltage and charging current at time t, respectively. , Let be the input voltage and input current of the charging system at time t, respectively.

[0120] A battery temperature rise model is established based on the battery thermal balance equation, and the calculation formula is as follows:

[0121]

[0122] In the formula, For battery temperature rise, The internal resistance of the battery at any given time (which varies with the remaining charge and temperature). For the specific heat capacity of the battery, For battery quality.

[0123] Circuit loss is defined as the sum of the switching transistor's conduction loss, the switching loss, and the passive component loss. The calculation formula is:

[0124]

[0125]

[0126]

[0127]

[0128] In the formula, For circuit loss, For the conduction loss of the switching transistor, For switching losses, For passive component losses, Drain-source current, For conduction voltage drop, Duty cycle, For switching frequency, For switching time, , These are the inductor current and the equivalent resistance, respectively. , These are the capacitor current and the equivalent resistance, respectively.

[0129] With the objectives of maximizing charging efficiency, minimizing battery temperature rise, and minimizing circuit losses, a multi-objective optimization model is established by combining data constraints in the dynamic correlation model. The formula is expressed as follows:

[0130]

[0131]

[0132] In the formula, To optimize variables, , These are the lower and upper bound constraints for the optimization variables, respectively.

[0133] The process of solving the multi-objective optimization model using the multi-objective artificial bee colony algorithm to obtain the solution set of the flow regulation strategy includes:

[0134] Initialize the bee colony and food sources, setting the colony size to S, including hired bees, observation bees, and scout bees. Each food source corresponds to a set of optimization variables, which are randomly generated. An initial food source, satisfying .

[0135] Calculate the objective function value for each food source, perform non-dominated sorting of the food sources based on Pareto dominance, and determine the non-dominated solution set.

[0136] For each food source New food sources are generated through neighborhood search, as expressed by the formula:

[0137]

[0138] In the formula, For the k-th food source, For the kth new food source, A random number in the range [-1, 1] For other food sources selected randomly, if Dominate Then replace .

[0139] Calculate the selection probability of each food source, observe the bees select food sources according to the probability and repeat the neighborhood search, and update the food sources.

[0140] If a food source is not updated for H consecutive times, it is considered abandoned, and the scout bee randomly generates a new food source to replace the abandoned food source.

[0141] Record the non-dominated solutions for each iteration. Stop iterating when the number of iterations reaches a preset maximum value. Use the final set of non-dominated solutions as the solution set for the flow control strategy.

[0142] The process of selecting the optimal current regulation strategy for the current charging condition using intuitionistic fuzzy decision theory includes:

[0143] Determine the decision factor set and the evaluation set; factor set , For charging efficiency, For battery temperature rise, For circuit losses, comment collection , For excellence, For good, Generally speaking, It is considered poor.

[0144] For the j-th strategy in the solution set of the current regulation strategy, the membership degree is determined based on its performance on the i-th factor. (Degree of conformity to the comments) and degree of non-membership (Degree of non-compliance with comments), meets , forming a matrix , The number of strategies.

[0145] Based on the current operating conditions, the weight vector is calculated using the analytic hierarchy process (AHP). ,satisfy and .

[0146] The intuitionistic fuzzy weighted average operator is used to calculate the comprehensive evaluation value for the j-th strategy, expressed by the formula:

[0147]

[0148] In the formula, Let be the membership degree of the j-th strategy in intuitionistic fuzzy decision-making. Let be the degree of non-membership of the j-th strategy in intuitionistic fuzzy decision-making. Let i be the weight of the i-th decision factor. Let be the membership degree of the j-th strategy under the i-th decision factor. Let be the degree of non-membership of the j-th strategy under the i-th decision factor.

[0149] Based on the scoring function The strategy with the highest score among all strategies is the optimal variable current control strategy for the current operating condition.

[0150] The optimal converter regulation strategy is converted into a control signal using a microcontroller, and the converter parameters of the extreme charge system are adjusted by combining feedforward-feedback composite control. The process includes:

[0151] The optimal converter control strategy is analyzed, and key parameters are extracted, including the target switching frequency, target duty cycle, and target input voltage.

[0152] Based on the changing trends of remaining battery charge and ambient temperature, the feedforward control quantity is calculated using the following formula:

[0153]

[0154] In the formula, This is the feedforward control variable, and SOC is the remaining battery capacity. Based on the remaining battery power, For ambient temperature, Based on ambient temperature, and This represents the feedforward coefficient.

[0155] A proportional-integral-derivative (PI-DE) controller is used, with the deviation between the output voltage and the target voltage as input, to calculate the feedback control quantity. The formula is as follows:

[0156]

[0157] In the formula, For feedback control, , For the target output voltage, This is the actual output voltage. , , These are the proportional, integral, and differential coefficients, respectively.

[0158] The feedforward control quantity and the feedback control quantity are superimposed to obtain the total control quantity. The microcontroller generates a PWM control signal based on u, and adjusts the turn-on and turn-off timing of the switching transistors in the wideband converter circuit to achieve precise control of the converter parameters (switching frequency, duty cycle).

[0159] In summary, this embodiment provides a deep learning-based wideband variable current adaptive control method for mobile phone fast charging. By employing empirical mode decomposition and local linear embedding algorithms to denoise and extract features from multi-dimensional operating data, a charging state evaluation model based on a deep belief network is constructed. This model can accurately assess the operating state of the charging system in real time and make preliminary adjustments to the variable current parameters based on the state. Simultaneously, multi-dimensional data is fused using a particle filtering algorithm, and a dynamic correlation model is constructed using an echo state network. Combined with a multi-objective artificial bee colony algorithm, a multi-objective optimization model considering comprehensive charging efficiency, battery temperature rise, and circuit losses is solved to obtain the optimal variable current control strategy. This significantly improves charging efficiency and adaptively adjusts parameters such as switching frequency, duty cycle, and input voltage under different battery remaining capacity and ambient temperature conditions, effectively shortening charging time and meeting users' urgent demand for fast charging.

[0160] By establishing a battery temperature rise model based on the heat conduction equation and a circuit loss model that integrates the losses of various circuit components, battery temperature rise and circuit loss are used as important indicators for multi-objective optimization. While improving charging efficiency, battery temperature rise and circuit loss are effectively reduced, greatly reducing the risk of battery overheating and circuit component aging, extending the service life of the battery and charging system, and improving the safety and reliability of mobile phone charging.

[0161] By employing echo state networks to construct a dynamic correlation model for multi-dimensional data, the dynamic relationships between multi-dimensional data can be captured more accurately. Simultaneously, compared to non-dominated sorting genetic algorithms, the multi-objective artificial bee colony algorithm can obtain the optimal current regulation strategy in a shorter time, meeting the real-time requirements of mobile phone ultra-fast charging systems. Intuitive fuzzy decision theory is used to select the optimal current regulation strategy for the current charging condition, and combined with feedforward-feedback composite control to adjust the current regulation parameters, making the current regulation highly intelligent and adaptable, achieving optimal charging performance under various complex charging conditions.

[0162] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0163] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A deep learning-based adaptive control method for wideband converter current in mobile phone ultra-charging, characterized in that, include: Real-time collection of multi-dimensional operational data in mobile phone fast charging scenarios, including charging circuit parameters, battery status data, and environmental operating condition data; The multi-dimensional operational data is denoised using the empirical mode decomposition method, and features are extracted using the local linear embedding algorithm to obtain circuit features, battery features, and environmental features. A charging state evaluation model based on deep belief network is constructed. Based on the circuit characteristics, battery characteristics and environmental characteristics, the operating state of the mobile phone's ultra-fast charging system is evaluated in real time, and the initial current conversion parameters are adjusted according to the operating state and preset threshold rules. The multi-dimensional operational data is fused using a particle filter algorithm, and a dynamic correlation model of the multi-dimensional data is constructed using an echo state network. Based on the dynamic correlation model, a multi-objective optimization model for comprehensive charging efficiency, battery temperature rise and circuit loss is established. The multi-objective artificial bee colony algorithm is used to solve the multi-objective optimization model to obtain the solution set of the current conversion control strategy. The optimal current conversion control strategy for the current charging condition is selected by intuitionistic fuzzy decision theory. The optimal converter control strategy is converted into a control signal by a microcontroller, and the converter parameters of the polarization charging system are adjusted by combining feedforward-feedback composite control.

2. The method for adaptive control of wideband converter current in mobile phone ultra-charging based on deep learning according to claim 1, characterized in that, The charging circuit parameters include input voltage, output current, switching transistor conduction time, filter capacitor voltage, and inductor current; the battery status data includes remaining battery capacity, battery internal resistance, battery terminal voltage, and battery cycle count; the environmental condition data includes ambient temperature, ambient humidity, and charging interface contact resistance.

3. The method for adaptive control of wideband converter current in mobile phone ultra-charging based on deep learning according to claim 1, characterized in that, The process of feature extraction using the local linear embedding algorithm includes: Construct a multi-dimensional operational data matrix D, and standardize the data matrix D to eliminate differences in the dimensions of different features; Based on the Euclidean distance, determine the nearest neighbor samples of each sample, construct a local neighborhood matrix, and calculate the local reconstruction weights between the sample and its nearest neighbors; By minimizing the reconstruction error between samples and their nearest neighbors in the low-dimensional feature space, high-dimensional standardized data is mapped to a low-dimensional space to obtain a low-dimensional feature matrix of dimension p. Each row of the low-dimensional feature matrix corresponds to a combination vector of circuit features, battery features, and environmental features of a sample.

4. The method for adaptive control of wideband converter current in mobile phone ultra-charging based on deep learning according to claim 1, characterized in that, The process of constructing a charging state assessment model based on deep belief networks includes: Collect historical operation data samples of the mobile phone charging system to construct a feature matrix and correspond them to charging status labels, which include normal charging, slight abnormality, moderate abnormality and severe abnormality; A deep belief network structure is constructed, including a deep belief input layer, a restricted Boltzmann machine hidden layer, and a deep belief output layer. The number of nodes in the deep belief input layer is equal to the feature dimension, and the deep belief output layer adopts the softmax activation function, with the number of nodes equal to the number of charging state categories. Define the energy function and probability distribution of a restricted Boltzmann machine; The contrastive divergence algorithm is used to pre-train the restricted Boltzmann machine at each layer, and the weights and biases are optimized layer by layer. The deep belief network is adjusted by the backpropagation algorithm, and the cross-entropy loss function is minimized to obtain the state of charge assessment model.

5. The method for adaptive control of wideband converter current in mobile phone ultra-charging based on deep learning according to claim 1, characterized in that, The process of fusing multi-dimensional operational data using the particle filter algorithm includes: Generate N particles, each representing an initial estimate of the multi-dimensional running data, and initialize the particle weights; Construct a state transition model to predict each particle at time t, and obtain the predicted particle; Based on the multi-dimensional data actually collected at time t, calculate the likelihood probability of each predicted particle and update the particle weights. Calculate the weight normalization factor and normalize the weights. If the number of effective particles is less than the preset threshold, use the system resampling method to filter high-weight particles and replace low-weight particles to obtain the fused data at time t.

6. The method for adaptive control of wideband converter current in mobile phone ultra-charging based on deep learning according to claim 1, characterized in that, The process of constructing a dynamic correlation model of multi-dimensional data using echo state networks includes: The multi-dimensional operational data after particle filtering fusion is divided into time series to form an input sequence of length T; An echo state network structure is constructed, including an echo state input layer, a reservoir, and an echo state output layer. The echo state input layer is used to receive multi-dimensional fused data from time series input. The reservoir is used to perform nonlinear mapping and dynamic response on the input time series data through internally randomly connected neurons. The echo state output layer is used to output the reconstruction result of the multi-dimensional data based on the state vector of the reservoir. The ridge regression algorithm is used to train the output layer weights to minimize the prediction error. After training, the state vectors of the reservoir at each time step are used as nodes, and the temporal correlation strength between nodes is used as the connection weight to construct a dynamic correlation model of multi-dimensional data.

7. The method for adaptive control of wideband converter current in mobile phone ultra-charging based on deep learning according to claim 1, characterized in that, The process of establishing a multi-objective optimization model for comprehensive charging efficiency, battery temperature rise, and circuit loss based on the dynamic correlation model includes: Charging efficiency is defined as the ratio of the actual energy absorbed by the battery to the energy input to the charging system. A battery temperature rise model is established based on the battery thermal balance equation; Circuit loss is defined as the sum of the switching transistor's conduction loss, switching loss, and passive component loss; With the goals of maximizing charging efficiency, minimizing battery temperature rise, and minimizing circuit loss, a multi-objective optimization model is established by combining data constraints in the dynamic correlation model.

8. The deep learning-based adaptive control method for wideband converter current in mobile phone ultra-charging as described in claim 1, characterized in that, The process of solving the multi-objective optimization model using the multi-objective artificial bee colony algorithm to obtain the solution set of the flow regulation strategy includes: Initialize the bee colony and food sources, set the colony size to S, and assign a set of optimization variables to each food source, which are randomly generated. An initial food source; Calculate the objective function value for each food source, perform non-dominated sorting of the food sources based on Pareto dominance, and determine the non-dominated solution set. For each food source, generate new food sources through neighborhood search; Calculate the selection probability of each food source, observe the bees select food sources according to the probability and repeat the neighborhood search, and update the food sources; If a food source is not updated for H consecutive times, it is determined to be abandoned, and the scout bee randomly generates a new food source to replace the abandoned food source; Record the non-dominated solutions for each iteration. Stop iterating when the number of iterations reaches a preset maximum value. Use the final set of non-dominated solutions as the solution set for the flow control strategy.

9. The deep learning-based adaptive control method for wideband converter current in mobile phone charging as described in claim 1, characterized in that, The process of selecting the optimal current regulation strategy for the current charging condition using intuitionistic fuzzy decision theory includes: Determine the decision factor set and the evaluation set; factor set , For charging efficiency, For battery temperature rise, For circuit losses, comment collection , For excellence, For good, Generally speaking, Poor; For the j-th strategy in the solution set of the variable current control strategy, the membership degree and non-membership degree are determined according to its performance on the ith factor, and a matrix is ​​formed. Based on the current working conditions, the weight vector is calculated using the analytic hierarchy process (AHP). The intuitive fuzzy weighted average operator is used to calculate the comprehensive evaluation value of the j-th strategy; All strategies are ranked based on a scoring function, and the strategy with the highest score is the optimal variable current control strategy for the current operating condition.

10. The deep learning-based adaptive control method for wideband converter current in mobile phone charging according to claim 1, characterized in that, The process of adjusting the converter parameters of the extreme charge system using feedforward-feedback composite control includes: The optimal converter control strategy is analyzed, and key parameters are extracted, including the target switching frequency, target duty cycle, and target input voltage. Calculate the feedforward control quantity based on the changing trends of the remaining battery charge and ambient temperature. A proportional-integral-derivative controller is used, with the deviation between the output voltage and the target voltage as input, to calculate the feedback control quantity; The feedforward control quantity and the feedback control quantity are superimposed to obtain the total control quantity. The microcontroller generates a PWM control signal based on u to adjust the on and off timing of the switching transistors in the wideband converter circuit, thereby controlling the converter parameters.

Images