Auxiliary power intelligent starting control system and method based on battery health state

Abstract

The application discloses an auxiliary power intelligent starting control system and method based on battery health state, and particularly relates to the field of vehicle auxiliary power control, which comprises the following steps: obtaining the measured internal resistance, measured cold starting current and static voltage of a rescued battery, converting and weighting to obtain a health state value, and reading the corresponding first charging action from a charging action table; after the auxiliary power and the rescued battery are connected in parallel to a vehicle direct current power supply network, an electronic control device adjusts the duty cycle of a power converter according to the current charging action, controls the charging current or the charging voltage, forms a starting record, and adds the current charging action to the action sequence; the charging current or the charging voltage control parameters of the current action position are updated according to the transportation weighting of the historical starting records with similar states before the action, and the influence of the battery state and the given output mode on the starting result is distinguished by using a cause-effect forest.

Description

Technical Field

[0001] This invention relates to the field of vehicle auxiliary power control technology, and more specifically, to an intelligent auxiliary power start-up control system and method based on battery health status. Background Technology

[0002] Existing auxiliary starting devices mostly collect parameters such as static voltage, internal resistance, cold start current and ambient temperature of the battery being rescued to calculate the battery's health status, and select control methods such as direct parallel output, pre-charge before starting, slow-up output or prohibition of starting based on this, and then use the results of previous starts to correct the health level classification and output rules. During emergency road start-up, the same battery can only perform one start-up method in one rescue. Moreover, once the battery has undergone pre-charging, high-current discharge, or slow-rise output, its state of charge, polarization state, and temperature will change, making it impossible to restore it to the exact same initial state for repeated verification of another start-up method. At the same time, due to start-up timeliness and battery safety conditions, multiple output methods cannot be tested on the same battery in parallel. Therefore, the starting result under a certain health level is inevitably affected by both the battery status and the predetermined output mode: a Class C battery that starts successfully after pre-charging cannot be considered suitable for direct full power output; a Class D battery that fails to start during the gradual increase process cannot be considered to have an overestimation of its health status. If the health level boundary is still adjusted directly based on such results, it is easy to have situations where the pre-charging strategy is effective but is mistakenly considered to be under-classified, or the slow-rise output is insufficient but is mistakenly considered to be over-aged, which will cause continuous deviations in the subsequent charging current and charging voltage control rules. The technical problem this application aims to solve is: when different auxiliary start-up methods are executed at different health levels and the same initial battery state cannot be reproduced, how to eliminate the interference of the predetermined output method on the start-up result, so that historical start-up data can be accurately used for updating the subsequent charging current and charging voltage control rules. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an intelligent auxiliary power supply start-up control system and method based on battery health status. By performing transport weighting on historical start-up records with similar pre-action states, and using causal forests to distinguish the influence of battery status and predetermined output mode on the start-up result, the charging current or charging voltage control parameters at the current action position are updated accordingly to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an intelligent start-up control method for auxiliary power supply based on battery health status, comprising: S1. Obtain the measured internal resistance, measured cold start current and static voltage of the battery being rescued, convert and weight them to obtain the health status value, and read the corresponding first charging action from the charging action table. S2. After the auxiliary power supply and the rescued battery are connected in parallel to the vehicle's DC power supply network, the electronic control device adjusts the duty cycle of the power converter step by step according to the current charging action, controls the charging current or charging voltage, forms a start record, and adds the current charging action to the action sequence. S3. Read historical startup records from the industrial big data record library according to battery type and action sequence, group them according to the original charging action and another charging action, generate transportation costs according to the state difference before the action, and use the unbalanced Sinkhorn algorithm to iteratively obtain the transportation matrix, and form transportation weights by row sums. S4. Based on the transport weight, the causal forest algorithm is used to divide the split record and the estimated record. The record is recursively split according to the midpoint of the state field. In the leaf node corresponding to the start record, the protection trigger difference, start success difference and response burden difference of the other charging action relative to the original charging action are calculated to determine the next charging action. S5. Update the current action position according to the next charging action and keep other action positions unchanged. The electronic control device adjusts the duty cycle of the power converter and repeats S2 to S4 until the vehicle starts successfully or reaches the nominal termination state. Output the auxiliary start result and write it into the industrial big data record library.

[0005] In a preferred embodiment, S1 includes: S11. After the static voltage sampling is completed, the electronic control device controls the power converter to output the first verification charging segment and the second verification charging segment in sequence, so that the charging current of the second verification charging segment is twice that of the first verification charging segment and the number of execution steps is half that of the first verification charging segment, and stops outputting the same number of steps after each verification charging segment to form a voltage response sequence. Static voltage refers to the voltage across the battery being rescued, measured at the end of the static phase when the auxiliary power supply stops outputting and the vehicle's starting load is not connected. Electronic control devices refer to controllers that are connected to voltage sampling terminals, current sampling terminals, temperature sampling terminals, and power converter control terminals, and read sampled values, calculate output differences, and update the switching duty cycle according to a unified control sequence. A power converter is a DC-DC converter circuit that is connected in series between the auxiliary power supply and the battery being rescued, and changes the charging current or charging voltage according to the switching duty cycle output by the electronic control device. The number of execution steps refers to the number of control steps occupied by the charging action from the first output step to the last output step. Each control step sequentially completes current or voltage sampling, setpoint comparison and switch duty cycle update. The process of outputting the first verification charging stage and the second verification charging stage is as follows: The electronic control device first drives the power converter to run the first execution step number with the first charging current, then stops the power converter to run the current-stopping step number, and then runs the second execution step number with the second charging current that is twice the first charging current, half of the first execution step number, and stops the power converter to run the same current-stopping step number again. The voltage at the end of the rescued battery is recorded at each control step to form a voltage response sequence. S12. Based on the voltage response sequence, divide the voltage difference between the first current-on beat and the last current-off beat of each verification charging segment by the corresponding charging current to obtain two dynamic internal resistance values. Correct the measured internal resistance with the arithmetic mean of the two dynamic internal resistance values. Then normalize the absolute values ​​of the internal resistance health value, cold start health value and voltage health value deviating from one to weights respectively, and sum them according to the weights to obtain the health status value. The first control step after the power converter switches from stopping output to outputting charging current refers to the first control step after the power converter stops outputting and starts outputting charging current. The last control step after the power converter stops outputting current refers to the last control step of the current-stopping segment when the power converter maintains zero output. The internal resistance health value is the ratio of the nominal internal resistance to the calibrated measured internal resistance; the cold start health value is the ratio of the measured cold start current to the rated cold start current; and the voltage health value is the ratio of the static voltage to the rated voltage. S13. Locate the value segment in the charging action table according to the health status value, read the current setpoint or voltage setpoint, duty cycle step and execution number recorded in the value segment, form the first charging action and write it into the electronic control device. The charging action table refers to a data table that records charging control parameters according to the health status value range. Each value range corresponds to a charging action and records the charging control method, set value, duty cycle step distance and number of execution steps. The current setpoint is the target value used by the electronic control device to adjust the output current of the power converter during the charging operation, and the voltage setpoint is the target value used by the electronic control device to adjust the output voltage of the power converter during the charging operation. One of the setpoints is selected for each charging operation. Duty cycle step refers to the fixed change in the duty cycle of a power converter switch when the electronic control device increases or decreases it once within one control step.

[0006] In a preferred embodiment, S2 includes: S21. After the auxiliary power supply and the rescued battery are connected in parallel to the vehicle's DC power supply network, the set value, starting duty cycle, duty cycle step distance and execution number of the current charging action are read. In the first two steps, the electronic control device drives the power converter with the starting duty cycle and the starting duty cycle plus the duty cycle step distance respectively. The difference between the feedback quantities of the two steps is divided by the duty cycle step distance to obtain the initial response slope. The feedback quantity and the set value are both charging current or charging voltage. The auxiliary power supply refers to an auxiliary starting power supply that has a DC energy storage unit and outputs charging current or charging voltage to the DC power supply network of the vehicle where the battery being rescued is located via a power converter. The vehicle DC power supply network refers to a DC circuit that uses the battery being rescued as an energy storage branch and connects the starter power supply branch to the vehicle's DC bus. After the auxiliary power supply is connected, it forms a parallel power supply relationship with the battery being rescued. The initial duty cycle refers to the duty cycle written to the power converter switch control terminal in the first frame of the current charging action, which is taken from the duty cycle of the last frame of the previous charging action; when the current charging action is the first charging action, the duty cycle of the corresponding record in the charging action table is taken. S22. Starting from the third beat, subtract the current beat feedback value from the given value to obtain the control difference. Divide the control difference by the current response slope to obtain the duty cycle correction value. Add the duty cycle correction value to the current beat duty cycle. Write zero when the sum is less than zero, write one when the sum is greater than one, and retain the sum when the sum is between zero and one to form the next beat duty cycle. Then, divide the difference between the feedback values ​​of two adjacent beats by the difference between the duty cycles of two adjacent beats to update the current response slope. Execute beat by beat until the number of beats is reached, and output the duty cycle sequence. S23. Drive the power converter to adjust the charging current or charging voltage of the auxiliary power supply according to the duty cycle sequence, calculate the voltage drop and battery temperature rise with the start and end of the action, form a start record, and add the current charging action to the action sequence. The process of calculating the voltage drop and battery temperature rise is as follows: read the battery terminal voltage and battery temperature at the start of the action, then read the battery terminal voltage and battery temperature at the end of the action. Subtract the battery terminal voltage at the start of the action from the battery terminal voltage at the end of the action to obtain the voltage drop, and subtract the battery temperature at the start of the action from the battery temperature at the end of the action to obtain the battery temperature rise. Write both calculation results into the startup record.

[0007] In a preferred embodiment, S3 includes: S31. Freeze the same data version from the industrial big data record library according to the battery type and action sequence. Form a source group by taking the historical start record of the original charging action for the next action, and form a target group by taking the historical start record of each candidate charging action for the next action. Then, splice the health status value and ambient temperature to the terminal voltage and battery temperature at the start of the action to form a status column, and output the source status matrix and the target status matrix. S32. Calculate the median and median absolute deviation of each state field based on the source state matrix and the target state matrix. Subtract the median from the field value and divide by the median absolute deviation to form a dimensionless state matrix. Write zero for fields with a median absolute deviation of zero. Then perform singular value decomposition on the covariance matrix of the dimensionless state matrix and take the reciprocal of the non-zero singular values ​​to form a generalized inverse matrix. Multiply the difference vector between the source state column and the target state column on both sides of the generalized inverse matrix to obtain the transportation cost matrix.

[0008] In a preferred embodiment, S3 further includes: S33. The source marginal quantity and target marginal quantity are formed by the reciprocal of the number of records in the source group and the reciprocal of the number of records in the target group. The median of the positive values ​​in the transportation cost matrix is ​​used as the entropy regularization quantity, and the transportation kernel is formed by dividing the negative values ​​of the transportation cost by the exponent of the entropy regularization quantity. When the transportation cost matrix is ​​all zero, the transportation kernel is written as a matrix of all ones. Then, the unbalanced Sinkhorn algorithm is used to alternately update the source scaling quantity by dividing the source marginal quantity by the weighted row of the transportation kernel and the square root of the result, and update the target scaling quantity by dividing the target marginal quantity by the weighted column of the transportation kernel and the square root of the result, until the transportation matrices of two adjacent rounds are consistent bit by bit under the fixed-point format of the electronic control device, and the initial transportation matrix is ​​output. The number of records in the source group and the number of records in the target group refer to the number of historical startup records contained in the source group and the number of historical startup records contained in the target group within the current data version, after filtering by battery type and action sequence and deduplication by historical record number. The process of using the unbalanced Sinkhorn algorithm is as follows: Initialize each element of the source scaling and the target scaling to one. In each round of operation, first calculate the weighted sum of each row of the transport kernel with the target scaling. Then, divide the source scaling by the corresponding weighted sum and take the square root to update the source scaling. Next, calculate the weighted sum of each column of the transport kernel with the updated source scaling. Then, divide the target scaling by the corresponding weighted sum and take the square root to update the target scaling. Subsequently, multiply the source scaling, the transport kernel, and the target scaling in sequence to form the transport matrix until the element encoding of the transport matrix in this round is consistent with that of the transport matrix in the previous round after fixed-point format conversion. S34. The Dixtra projection algorithm is used to alternately perform row quality projection and column quality projection on the initial transportation matrix, so that the quality of each row does not exceed the corresponding source margin and the quality of each column does not exceed the corresponding target margin. The projection stops when the projection matrices of two adjacent rounds are consistent bit by bit under the fixed-point format of the electronic control device. The transportation weights of the projection matrix rows and the historical start record are formed for S4 to calculate the difference between the original charging action and the candidate charging action on the charging current or charging voltage adjustment results. The process of performing row quality projection and column quality projection using the Dixtra projection algorithm is as follows: The initial transport matrix is ​​used as the current projection matrix, and the row correction matrix and column correction matrix are initialized to all-one matrices. During row quality projection, the current projection matrix is ​​multiplied element-wise by the row correction matrix. If the sum of elements in a row does not exceed the corresponding source margin, the row is retained. If the sum of elements in a row exceeds the corresponding source margin, the source margin is divided by the sum of elements in that row to obtain the row coefficient, which is then multiplied by each element in that row. The row correction matrix is ​​then updated with the ratio of corresponding elements before and after projection. During column quality projection, the row quality projection result is multiplied element-wise by the column correction matrix. If the sum of elements in a column does not exceed the corresponding target margin, the column is retained. If the sum of elements in a column exceeds the corresponding target margin, the target margin is divided by the sum of elements in that column to obtain the column coefficient, which is then multiplied by each element in that column. The column correction matrix is ​​then updated with the ratio of corresponding elements before and after projection. Row quality projection and column quality projection are performed alternately until the element codes of adjacent projection matrices after fixed-point format conversion are consistent bit-by-bit.

[0009] In a preferred embodiment, S4 includes: S41. For each candidate charging action, read the positive transport element in the transport matrix, form a pairing record by connecting the original charging action record and the candidate charging action record with the positive transport element, form a pairing state by the arithmetic mean of the state fields before the action of the two records, form an action difference vector by subtracting the protection trigger value, start success value and response burden value of the original charging action record from the candidate charging action record, and form a pairing weight with the positive transport element. The positive transport element refers to the matrix element with a value greater than zero in the transport matrix after S34 projection. The row where the element is located corresponds to an original charging action record, and the column where the element is located corresponds to a candidate charging action record. The element value represents the transport quality allocated between the two records. Charging action records refer to historical start records stored in the industrial big data record library corresponding to a specific action position. These records include at least the pre-action status field, charging action, protection trigger value, start success value, and response burden value. The original charging action record uses the charging action recorded in the charging action table at that action position, while the candidate charging action record uses a charging action that is different from the original charging action. The protection trigger value is recorded as one when the terminal voltage reaches the discharge termination voltage or the battery temperature reaches the charging cut-off temperature during the charging operation, and as zero when neither occurs; the start success value is recorded as one when the vehicle starts after the charging operation is executed, and as zero when the start is not completed; the response burden value is obtained by adding the value obtained by dividing the terminal voltage drop by the nominal allowable drop and the value obtained by dividing the battery temperature rise by the nominal allowable rise. S42. Circularly shift the pairing record number to form the record order. In each record order, take the odd-numbered pairing records as split records and the even-numbered pairing records as estimated records. Use the honest causal forest algorithm to split the records sequentially by the midpoint of the adjacent values ​​of each field of the pairing state. Calculate the mean of the pairing weights of the action difference vectors in the two child nodes respectively. Divide the product of the sum of the pairing weights of the two child nodes by the square of the sum of the pairing weights of the parent node, and then multiply by the sum of the squares of the differences between the two mean vectors to obtain the splitting value. Read the first and second midpoints in descending order of the splitting value to form tree nodes until there are no midpoints in the node. Output the action difference tree. The process of using the honest causal forest algorithm is as follows: First, multiple record sequences are generated by cyclically shifting the pairing record numbers. In each record sequence, odd-numbered pairing records are fixed for tree node splitting, and even-numbered pairing records are fixed for leaf node action difference estimation. Tree node splitting uses only split records. The midpoint of adjacent values ​​of each field of the pairing state is read sequentially. The split record is divided into two child nodes according to the midpoint. The average pairing weight of the action difference vector of the two child nodes and the corresponding splitting value are calculated. The midpoint with the first splitting value is selected to establish the tree node. After the splitting is completed, the estimated record enters the leaf node along the established tree node, and the action difference vector is calculated only from the estimated record in the leaf node, so that node splitting and action difference estimation use different records.

[0010] In a preferred embodiment, S4 further includes: S43. Send the current start-up record's pre-action state into each action difference tree. Calculate the mean of the action difference vector in the estimated record falling into the leaf node according to the pairing weight. Then calculate the median and median absolute deviation for the same component output by each action difference tree. Use the median minus the median absolute deviation to form the lower bound of the action difference, and use the median plus the median absolute deviation to form the upper bound of the action difference. Output the action difference interval of the candidate charging action. The leaf node refers to the terminal node in the action difference tree where further segmentation stops. The leaf node corresponds to a set of estimated records that satisfy the conditions for all field values ​​from the root node to the node, and is used to calculate the action difference vector of the currently started record within the corresponding state range. A component refers to the numerical position in the action difference vector corresponding to a certain start-up result. The three components correspond to the protection trigger difference, start-up success difference, and response burden difference, respectively. The median and median absolute deviation are calculated for the same component output by each action difference tree. S44. First, arrange the candidate charging actions in ascending order according to the upper limit of the protection trigger difference. Then, arrange the candidate charging actions in the same position in descending order according to the lower limit of the start success difference. Then, arrange the candidate charging actions that are still in the same position in ascending order according to the upper limit of the response burden difference. Read the first candidate charging action as the next charging action. When there are fewer than two positive transport elements, retain the original charging action and have the electronic control device read the current setpoint or voltage setpoint of the next charging action to update the charging current or charging voltage at the current action position. The upper bound of the protection trigger difference is the sum of the median and the median absolute deviation of the protection trigger differences obtained from each action difference tree. It is used to represent the upper result of the candidate charging action increasing the protection trigger compared to the original charging action. The lower bound of the start success difference is the median of the start success difference obtained from each action difference tree minus the median absolute deviation. It is used to represent the lower result of the candidate charging action increasing the start success compared to the original charging action. The upper bound of the response burden difference refers to the sum of the median and the median absolute deviation of the response burden differences obtained from each action difference tree. It is used to represent the upper bound of the candidate charging action relative to the original charging action, which increases the terminal voltage drop and the battery temperature rise.

[0011] In a preferred embodiment, S5 includes: S51. Read the next charging action and the current action position, copy the charging action table to the execution action table, replace the original charging action at the current action position in the execution action table with the next charging action, restore the other action positions to the corresponding values ​​in the charging action table, and output the updated execution action table. S52. Read the given value and duty cycle step of the current action position according to the action table. Divide the difference between the given value and the current feedback amount by the current response slope to obtain the duty cycle correction amount. Limit the duty cycle correction amount to between positive and negative duty cycle step sizes and add it to the current duty cycle. Write zero when the sum is less than zero, write one when the sum is greater than one, and retain the sum when the sum is between zero and one to form a candidate duty cycle.

[0012] In a preferred embodiment, S5 further includes: S53. The electronic control device drives the power converter to run one control step with the candidate duty cycle. It calculates the predicted feedback quantity corresponding to the candidate duty cycle based on the current response slope. When the direction of change of the predicted feedback quantity is consistent with the direction of change of the measured feedback quantity, the candidate duty cycle is retained. When the two change directions are opposite, the current duty cycle is restored and the duty cycle step size is halved before S52 is executed again until the two change directions are consistent or the duty cycle step size is zero after fixed-point encoding. The duty cycle sequence is then output. The process of calculating the predicted feedback quantity corresponding to the candidate duty cycle is as follows: read the current feedback quantity, the current duty cycle, the candidate duty cycle and the current response slope, multiply the difference between the candidate duty cycle and the current duty cycle by the current response slope, and then add it to the current feedback quantity to obtain the predicted feedback quantity corresponding to the candidate duty cycle. The current feedback quantity and the predicted feedback quantity are uniformly expressed as charging current value or charging voltage value. S54. Adjust the charging current or charging voltage of the auxiliary power supply to the battery being rescued according to the duty cycle sequence, add the next charging action to the action sequence and repeat S2 to S4 until the vehicle start signal appears, or either the terminal voltage or the battery temperature reaches the corresponding termination value in the nominal parameter table, output the auxiliary start result, and associate the start record with the auxiliary start result and write it into the industrial big data record library. The nominal parameter table refers to a data table that records auxiliary start and stop parameters according to battery type. Each record is associated with the corresponding discharge stop voltage and charging stop temperature by the battery type code, and records the parameter unit and effective version.

[0013] An intelligent start-up control system for auxiliary power supply based on battery health status, comprising a health assessment module, a charging regulation module, a weight generation module, an action decision module, and a start-up execution module: The health assessment module is used to obtain the measured internal resistance, measured cold start current and static voltage of the rescued battery, convert and weight them to obtain the health status value, and read the corresponding first charging action from the charging action table. The charging adjustment module is used to adjust the duty cycle of the power converter step by step according to the current charging action after the auxiliary power supply and the rescued battery are connected in parallel to the vehicle's DC power supply network, control the charging current or charging voltage, form a start record, and add the current charging action to the action sequence. The weight generation module is used to read historical start-up records from the industrial big data record library according to battery type and action sequence, group them according to the original charging action and another charging action, generate transportation costs according to the state difference before the action, and use the unbalanced Sinkhorn algorithm to iteratively obtain the transportation matrix, and form transportation weights by row sums. The action decision module uses a causal forest algorithm based on transportation weight to divide split records and estimation records. It recursively divides the records according to the midpoint of the status field. In the leaf node corresponding to the start record, it calculates the protection trigger difference, start success difference, and response burden difference of another charging action relative to the original charging action to determine the next charging action. The start-up execution module is used to update the current action position according to the next charging action and keep other action positions unchanged. The electronic control device adjusts the duty cycle of the power converter and repeats S2 to S4 until the vehicle starts successfully or reaches the nominal termination state. The auxiliary start-up result is output and written to the industrial big data record library.

[0014] The technical effects and advantages of this invention are as follows: 1. This scheme matches the pre-state of similar actions by transport weights and uses causal forests to distinguish the impact of different charging actions, thereby reducing the confusion between historical startup results and the predetermined output method. 2. The dynamic internal resistance obtained from the verification charging is used to correct the measured internal resistance, and the health status value is calculated by combining the cold start current and static voltage, so that the first charging action is more in line with the current battery status. 3. The duty cycle of the power converter is corrected step by step according to the unified control step, and the response slope is updated according to the feedback, so that the charging current or charging voltage gradually approaches the given value and alleviates the sudden output change. 4. Based on the protection trigger difference, start success difference, and response burden difference, an action difference range is formed, and candidate charging actions are screened according to a fixed order, so that the action selection takes into account both the start result and the battery burden. 5. Control is performed by replacing only the current motion position while retaining other motion positions, while recording the actual trial shots and start-up results, so that the motion adjustment range is constrained and a consistent data basis is formed. Attached Figure Description

[0015] Figure 1 This is a flowchart of the auxiliary start-up control process of the present invention; Figure 2 This is a diagram of the auxiliary start-up control system of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Refer to the instruction manual appendix Figure 1-2 The present invention provides an intelligent start-up control method for auxiliary power supply based on battery health status, comprising: S1. Obtain the measured internal resistance, measured cold start current and static voltage of the battery being rescued, convert and weight them to obtain the health status value, and read the corresponding first charging action from the charging action table. Before the assisted start begins, a unified measurement benchmark is established for the battery being rescued; the electronic control device freezes the versions of the nominal parameter table, verification parameter table, and charging action table used for this assisted start, and reads the unified control tap duration; With the auxiliary power supply stopped and the vehicle starting load disconnected, the vehicle was kept stationary according to the number of stationary cycles recorded in the verification parameter table. At the end of the stationary cycle, the static voltage was collected from the positive and negative terminals of the rescued battery, and the ambient temperature was read. Then, the measured internal resistance was obtained through the AC injection circuit, and the measured cold start current was obtained through the conductivity test circuit, so that the static voltage was not affected by the polarization changes caused by the subsequent test process. The verification charging is used to obtain the terminal voltage response under different charging currents; the electronic control device reads the first charging current, the first execution step number, and the current stop step number from the verification parameter table. When the first charging current exceeds the rated output current of the power converter, the allowed charging current of the rescued battery, or the current output capacity of the auxiliary power supply, the lower value of the three limits is read; when the first execution step number is odd, one is subtracted to form an even step number. After completing the parameter verification, execute S11. First, drive the power converter to run the first execution step number according to the first charging current, then stop the output current stop step number, and then execute the second verification charging segment according to twice the first charging current. The second execution step number is half of the first execution step number, and the current is stopped again for the same number of steps after the second verification charging segment ends. When the second charging current exceeds any output limit, the verification is terminated and written into the verification failure record; otherwise, each control step collects the voltage at the end of the rescued battery after the duty cycle is updated and the power converter is running, writes it into the voltage response sequence according to the step number, and terminates the current verification when the sampling is missing and retains the obtained record. The terminal voltage response is used to form the corrected internal resistance and health status value; when S12 is executed, the absolute value of the difference between the terminal voltage of the first current-carrying phase of the first verification charging section and the static voltage of the last static phase of the resting section is divided by the first charging current to obtain the first dynamic internal resistance value. The absolute value of the difference between the voltage at the first tap of the second verification charging section and the voltage at the last tap of the first stopping section is divided by the second charging current to obtain the second dynamic internal resistance value. When both dynamic internal resistance values ​​are valid, the arithmetic mean is written as the corrected internal resistance; when only one dynamic internal resistance value is valid, the valid value is written as the corrected internal resistance; when both dynamic internal resistance values ​​are missing, the measured internal resistance is retained and a verification failure flag is written. The nominal internal resistance is divided by the corrected internal resistance to form the internal resistance health value; the measured cold start current is divided by the rated cold start current to form the cold start health value; and the static voltage is divided by the full-charge static reference voltage to form the voltage health value. The three health values ​​are subtracted from each other and their absolute values ​​are taken. Each absolute value is then divided by the sum of the three absolute values ​​to form a corresponding weight. The health status value is obtained by summing the weights. When the sum of the three absolute values ​​is zero, all three weights are written as one-third, and the health status value is written as one; when any denominator is missing or zero, the calculation stops and a parameter error record is output. After the health status value is formed, the electronic control device reads the charging action table according to the frozen version; the numerical segments in the charging action table adopt the rule that the left endpoint belongs to the current segment and the right endpoint belongs to the next segment, and the last numerical segment includes the right endpoint. When the health status value is lower than the lower end of the first value segment, it is assigned to the first value segment; when it is higher than the upper end of the last value segment, it is assigned to the last value segment. After locating the numerical segment, execute S13 to read the first action position from the charging actions arranged by action position, and read the current setpoint or voltage setpoint according to the charging control method, while reading the duty cycle step and the number of execution steps. If a charging action records both current and voltage setpoints, both setpoints are missing, the duty cycle step exceeds zero to one, or the number of execution steps is not a positive integer, writing will stop and an error record in the charging action table will be generated; if the parameters are complete, the first charging action will be written to the current action register of the electronic control device for subsequent charging adjustment to read. Taking a rescued battery with a rated voltage of 12V and a rated cold start current of 600A as an example, after freezing the three parameter table versions, the electronic control device obtains the static voltage, ambient temperature, measured internal resistance and measured cold start current in sequence. The verification parameter table records that when the first charging current is 5A, the first execution step number is 20 steps, and the current stop step number is 8 steps, the power converter sequentially executes 5A charging for 20 steps, current stop for 8 steps, 10A charging for 10 steps, and current stop for 8 steps. The electronic control device generates two dynamic internal resistance values ​​based on this, writes the correction internal resistance, calculates the health status value, and reads the first charging action according to the numerical segment assignment rules, so as to keep the sampling timing, verification output, internal resistance correction, weight calculation and action reading consistent.

[0018] S2. After the auxiliary power supply and the rescued battery are connected in parallel to the vehicle's DC power supply network, the electronic control device adjusts the duty cycle of the power converter step by step according to the current charging action, controls the charging current or charging voltage, forms a start record, and adds the current charging action to the action sequence. After the auxiliary power supply completes the verification of positive and negative polarity and voltage platform, it is connected in parallel with the rescued battery to the vehicle's DC power supply network. The electronic control device freezes the current charging action, the nominal parameter table and the charging action table version, and adjusts the power converter according to the unified control step. When the current charging action is current-controlled, the feedback measurement is taken of the output current of the power converter, and the voltage sample value is used to limit the output voltage; when the current charging action is voltage-controlled, the feedback measurement is taken of the terminal voltage of the battery being rescued, and the current sample value is used to limit the output current. Each control step sequentially completes the duty cycle writing, power converter operation, and end-of-step feedback acquisition, ensuring a fixed timing relationship between the duty cycle and the feedback quantity. After the parallel state is established, first verify whether the current charging action can form a valid first two-step response; the electronic control device reads the given value, the initial duty cycle, the duty cycle step size and the number of execution steps. If the number of execution steps is less than two, the duty cycle step size fixed-point code is zero, or the initial duty cycle plus the duty cycle step size exceeds zero to one, the output stops and an error record of the charging action table is generated. When the parameters are valid, in S21, the power converter is driven to run the first two cycles with the initial duty cycle and the initial duty cycle plus the duty cycle step size, respectively, and the feedback quantity is read at the end of each cycle; the feedback quantity of the second cycle minus the feedback quantity of the first cycle is divided by the duty cycle step size to obtain the initial response slope. When the initial response slope fixed-point code is zero or the sign is negative, the current charging action is stopped and a no-response record is written; when the terminal voltage reaches the charging voltage limit recorded in the nominal parameter table during current control, or when the output current reaches the output value jointly limited by the rated current of the power converter and the allowable charging current of the rescued battery during voltage control, the power converter is stopped. When feedback is missing, repeated, or delayed, the first complete record is retained according to the control beat number, and the remaining records are not included in the response slope calculation. The current charging action is terminated when the current beat lacks a complete record. After entering the third control phase, the duty cycle is corrected step by step based on the measured response. When executing S22, the control difference is formed by subtracting the current phase feedback from the given value, and then the duty cycle correction is obtained by dividing the control difference by the current response slope. When the duty cycle correction exceeds the positive or negative duty cycle step size, it is truncated according to the corresponding duty cycle step size, added to the current duty cycle, and then limited to between zero and one to form the duty cycle of the next beat. After the next cycle ends, the current response slope is updated by dividing the difference between the feedback values ​​of the two adjacent cycles by the difference between the duty cycles of the two adjacent cycles; if the duty cycle clipping results in the same duty cycle for the two adjacent cycles, the previous current response slope is retained. When the fixed-point code of the newly calculated current response slope is zero or negative, the previous positive response slope is retained and written to the slope exception flag. Output stops when the previous positive response slope does not exist. The above calculations are performed until the actual number of operating cycles reaches the execution cycle number. The duty cycles actually written to the power converter are then connected according to the cycle number to form a duty cycle sequence. The execution results of the action are used to form the start record required for subsequent industrial big data calculation; after the power converter completes the output according to the duty cycle sequence, the battery terminal voltage and battery temperature of the start and actual end of the action are read in S23. When the difference between the initial voltage and the actual final voltage is positive, the difference is written as the voltage drop. When the difference is zero or negative, the voltage drop is written as zero, and the absolute value of the negative value is written as the voltage rise. When the difference between the actual battery temperature at the end of the action and the battery temperature at the beginning of the action is positive, write the difference as the amount of battery temperature rise; when the difference is zero or negative, write the amount of battery temperature rise as zero. After the current charging action is fully executed and the output termination is not triggered, the electronic control unit issues a start permission signal at the next control step boundary and reads the engine operating position output by the engine controller via the vehicle bus within the observation step count after the starter is connected; when the engine operating position is written to 1, it is recorded as a successful start, and when the observation step count is still zero at the end, it is recorded as a failed start. All current charging actions that have actually been output by the power converter through one control step are added to the action sequence, and the complete execution or protection termination status is written into the startup record; current charging actions that have not been output during the parameter verification phase are not included in the action sequence. In a 12V lead-acid jump starter battery rescue mission, the current charging action is controlled by current, with an initial duty cycle of 0.20, a duty cycle step of 0.02, and 20 execution steps. The electronic control unit calculates the initial response slope based on the feedback current in the first two steps. From the third step onwards, it corrects the duty cycle based on the difference between the current setpoint and the feedback current at the end of the step, while using the terminal voltage to limit the output. After the action is completed, it calculates the non-negative terminal voltage drop and battery temperature rise, and reads the engine operating position within the observation period after the starter is engaged. The resulting startup record simultaneously retains the duty cycle sequence, actual termination beat, action status, and startup result, which can be used for subsequent action sequence matching and charging action comparison.

[0019] S3. Read historical startup records from the industrial big data record library according to battery type and action sequence, group them according to the original charging action and another charging action, generate transportation costs according to the state difference before the action, and use the unbalanced Sinkhorn algorithm to iteratively obtain the transportation matrix, and form transportation weights by row sums. Before industrial big data records are compared with charging actions, it is necessary to eliminate the impact of version changes, record anomalies, and misaligned action timing on transportation relationships. When an auxiliary start-up task begins, the electronic control device simultaneously freezes the versions of the industrial big data record library, the charging action table, and the nominal parameter table; the current action position is taken as the action sequence length plus one, the current state before the next action is taken as the terminal voltage and battery temperature at the end of the previous charging action, and the first action position is taken as the state before the first charging action is formed by S1. Historical startup records are sorted by startup task number, action position, and historical record number. When there are duplicate records for the same startup task number and action position, the record with the earliest time and complete fields is retained. Records that are missing battery type, action sequence, pre-action status, or next charging action are not included in this calculation. Late records are written to subsequent data versions. After completing the record organization, execute S31 to form a source group of historical start records that have the same battery type and action sequence and whose next action uses the original charging action, and form a target group of historical start records that use each candidate charging action. Health status value, ambient temperature, initial voltage of action, and initial battery temperature of action are written into the status column in a fixed column order to form a source status matrix and a target status matrix; when the source group is empty, the original charging action is retained and S3 ends; when the target group is empty, the corresponding candidate charging action is deleted. Dimensional differences and correlations in state fields are processed using the same statistical benchmark. For a source group and a target group, the field values ​​of the two groups in the same state field are first merged, and the median is obtained in ascending order. Then, the absolute difference between the field value and the median is used to obtain the median absolute deviation according to the same rules. S32 uses the same median and median absolute deviation to transform the source state matrix and the target state matrix respectively, and writes zeros for the entire column of fields whose median absolute deviation is fixed-point encoded as zero. The dimensionless state matrix after merging is subtracted from the column mean, and the covariance matrix is ​​formed by using the number of merged records as the divisor; when the number of merged records is one, the covariance matrix is ​​written as a zero matrix. After performing singular value decomposition on the covariance matrix, the singular values ​​remain zero when they are encoded as zero in the fixed-point format of the electronic control device, and the reciprocals of the other singular values ​​are taken. The generalized inverse matrix is ​​then formed by multiplying the right singular vector matrix, the reciprocal diagonal matrix, and the left singular vector matrix in that order. The difference between each source state column and each target state column is calculated. The corresponding transportation cost is obtained by multiplying the transpose of the difference vector, the generalized inverse matrix, and the difference vector in sequence. All transportation costs are written into the transportation cost matrix according to the source record rows and target record columns. When the covariance matrix is ​​all zero, the transportation cost matrix is ​​written as a matrix of all zeros. The allocation of transport quality simultaneously constrains state difference, matrix entropy, and marginal deviation; each element of the source marginal quantity is the reciprocal of the number of records in the source group, and each element of the target marginal quantity is the reciprocal of the number of records in the target group. When the transportation cost matrix has positive values, the median of the positive values ​​is read as the entropy regularization. When all transportation costs are zero, the transportation kernel is written as a one matrix. S33 uses the unbalanced Sinkhorn algorithm to reduce the sum of transportation costs, the entropy of the transportation matrix, and the KL divergence of the row quality and column quality deviations from the corresponding marginal quantities. When the entropy regularization and the two KL divergence coefficients take the same value, the source scaling and target scaling are updated using the square root. When the transport kernel element is lower than the minimum positive code of the fixed-point format, it is written as the minimum positive code. When the weighted row sum or column sum of the transport kernel is zero, the corresponding record is deleted and the marginal quantity is re-formed. When the multiplication result exceeds the fixed-point bit width, the source scaling amount and the target scaling amount are simultaneously shifted to the right by the same number of bits to maintain the proportion of each element in the transport matrix. In each round, the source scaling, target scaling, and transport matrix are updated sequentially. When the fixed-point codes of the transport matrices of adjacent rounds are consistent bit by bit, the initial transport matrix is ​​output. When non-adjacent duplicate codes occur, the matrix with the lower transport target value is read from each transport matrix in the loop and copied for the next round. The process stops after the codes of adjacent rounds are consistent. The upper bounds of row quality and column quality are further corrected by constrained projection. After the initial transport matrix enters S34, the Dixtra projection algorithm uses the initial transport matrix as a reference and reduces the KL divergence of the current projection matrix relative to the initial transport matrix under the constraint that the quality of each row does not exceed the corresponding source margin and the quality of each column does not exceed the corresponding target margin. The row correction matrix and column correction matrix are initialized as all-one matrices. For row projection, the current projection matrix is ​​multiplied element by element by the row correction matrix. When the row sum exceeds the source margin, the entire row is scaled according to the ratio of the source margin to the row sum. Then, the row correction matrix is ​​updated by dividing the elements before projection by the elements after projection. The column projection processes the column quality and column correction matrix in the same way. When both elements before and after projection are zero, write a correction value of 1; when the elements after projection are zero and the elements before projection are positive, write the upper limit of the fixed-point format allowed encoding. Row projection and column projection are performed alternately. When the projection matrices of two adjacent rounds are identical, the projection matrix at the time of stopping is recorded as the final transport matrix. When an encoding loop occurs, the projection matrix with the lower KL divergence in the loop is read and copied to the next round before stopping. The rows of the final transport matrix form the transport weights of the source group's historical startup record. The final transport matrix and transport weights are written into the calculation record of the current candidate charging action for S4 to read. Taking 120 original charging action records and 80 candidate charging action records under the same battery type and the same action sequence as an example, the electronic control device first cleans up duplicate and missing records according to the frozen version, and then completes dimensionless transformation and covariance correction with two sets of common statistics. The unbalanced Sinkhorn algorithm allows two sets of records with different numbers of records and allocates transport mass based on the state before the action. Then, it limits the transport mass of a single record by projecting the row and column mass. The resulting transportation weights maintain the same sequence of actions and the same state at the same time, and are not affected by differences in new records, field units, and sample numbers. They can be directly used to compare the results of the original charging actions with those of the candidate charging actions.

[0020] S4. Based on the transport weight, the causal forest algorithm is used to divide the split record and the estimated record. The record is recursively split according to the midpoint of the state field. In the leaf node corresponding to the start record, the protection trigger difference, start success difference and response burden difference of the other charging action relative to the original charging action are calculated to determine the next charging action. After completing the final transport matrix, this implementation method uses the transport quality to limit the comparison relationship between the original charging action record and the candidate charging action record, and then establishes an action difference tree with mutually isolated historical start records, so that the differences caused by the candidate charging action to protection triggering, start success and response burden can be estimated separately. The current state before the next action is uniformly taken from the health status value, ambient temperature, terminal voltage and battery temperature of the previous charging action termination time; when the current action position is the first action position, S1 is read to form the corresponding state before the first charging action. Non-zero connections in transport quality are used to establish action pairs with similar states. For a candidate charging action, the electronic control device reads the final transport matrix row by row and identifies elements with fixed-point codes greater than zero as positive transport elements in S41. Pairing records are formed by the original charging action record corresponding to the row where the element is located and the candidate charging action record corresponding to the column where the element is located. The pairing record number is generated by concatenating two historical record numbers and the current action position. The pairing state is the arithmetic mean of the state fields before the action of the two records. The pairing weight is the positive transport element. The protection trigger value reads the termination status during the charging action. When the terminal voltage reaches the discharge termination voltage or the battery temperature reaches the charging cut-off temperature, it is written as 1. The start success value reads the engine running position in the observation panel after the charging action is completed. When the engine running position is written as 1, it is recorded as 1. The allowable drop is the voltage at the start of the action minus the discharge termination voltage; the allowable rise is the charging cutoff temperature minus the battery temperature at the start of the action; the response burden value is formed by adding the voltage drop divided by the allowable drop and the battery temperature rise divided by the allowable rise. When any permissible value is not positive, the protection trigger value is written and the corresponding paired record is deleted; the valid paired records form the action difference vector according to the direction of the candidate charging action record minus the original charging action record; When the number of transported elements increases, the pairing records are written to the temporary record area in the final transport matrix row order, and the matrix row buffer is released after a row is completed. Tree node partitioning and leaf node estimation use mutually isolated historical start records; paired records are sorted in ascending order by paired record number and then cyclically shifted. S42 first lists odd-numbered paired records as initial split records and even-numbered paired records as initial estimation records. Paired records that share the same original charging action record or candidate charging action record are grouped into the same association group. The association group as a whole is grouped into the set of the first paired record in the group, to prevent the same historical start record from participating in node division and action difference estimation at the same time. Records that form the same group repeatedly are retained only once; if any set is empty, the corresponding record order is deleted. Before segmentation, the median absolute deviation of the three action difference components in the split record is calculated respectively, and the segmentation vector is formed by dividing each component by the corresponding median absolute deviation. The component with a median absolute deviation of zero is written as zero. Then, the midpoints of adjacent different values ​​are read according to the column order of the pairing status field, and the split record is divided into two child nodes. The segmentation value is calculated based on the average and sum of the pairing weights of the two child nodes. When the segmentation values ​​are the same, the segmentation position is determined by the column order of the status field and the ascending order of the midpoints within the field. If a midpoint fails to enable both child nodes to obtain a split record, it is not included in the comparison. The segmentation stops when there is no valid midpoint within the node, forming an action difference tree. The current state obtains the local results of the candidate charging action along the action difference tree; when executing S43, the electronic control device sequentially sends the current state before the next action into each action difference tree, and enters the corresponding leaf node according to the state field and midpoint recorded in the tree node. If a leaf node has no estimated record, the process returns to the parent node level by level until a node containing an estimated record is found. If the root node still has no estimated record, the corresponding action difference tree is deleted. When the pairing weight and fixed-point code of the estimated record within a node are zero, the corresponding action difference tree is also deleted; otherwise, the mean of the original action difference vector is calculated using the pairing weight. For the protection trigger difference, start success difference, and response burden difference output by the retained action difference tree, calculate the median and median absolute deviation respectively. Subtract the median absolute deviation from the median to form the lower bound of the action difference, and add the median absolute deviation to the median to form the upper bound of the action difference, thus obtaining the action difference interval of the candidate charging action. The action difference interval only represents the discrete range of each action difference tree output and does not include statistical confidence meaning. Candidate charging actions are assigned output parameters for the current action position according to a fixed result order. After entering S44, the electronic control device first converts the upper limit of the protection trigger difference into a fixed-point code and arranges them in ascending order. Then, for candidate charging actions with the same code, they are arranged in descending order according to the lower limit of the start success difference. If they are still in the same position, they are arranged in ascending order according to the upper limit of the response burden difference. When all three codes are the same, the first record is read in ascending order according to the candidate charging action number. If a candidate charging action generates fewer than two valid positive transport elements, fails to form a valid action difference tree, or has a missing action difference interval, the candidate charging action is removed from the sorting process; if all candidate charging actions are removed, the original charging action is retained. After determining the next charging action, the electronic control device reads the corresponding current setpoint or voltage setpoint and writes it into the current action position. Other action positions retain the original values ​​in the charging action table, which are used by S5 to perform duty cycle adjustment. Taking the case where there is one original charging action and two candidate charging actions at the same action position, the final transport matrix first forms a pair of historical start records with similar states, and the pair of records that share the same historical start record are fixed in the same data set. Based on this, the action difference tree estimates the protection trigger difference, start-up success difference, and response burden difference of the two candidate charging actions relative to the original charging action; When sorting, first exclude candidate charging actions with later fixed-point codes corresponding to the upper limit of protection trigger difference, then compare the lower limit of start success difference with the upper limit of response burden difference, thereby obtaining the next charging action that only acts on the current action position, and hand over the corresponding current setpoint or voltage setpoint to the subsequent charging control for reading.

[0021] S5. Update the current action position according to the next charging action and keep other action positions unchanged. The electronic control device adjusts the duty cycle of the power converter and repeats S2 to S4 until the vehicle starts successfully or reaches the nominal termination state. Output the auxiliary start result and write it into the industrial big data record library. After the next charging action is formed, the electronic control device uses the version of the charging action table and nominal parameter table that has been frozen in this auxiliary start, and determines the current action position by adding one to the action sequence length; When the current action position exceeds the range of action positions recorded in the charging action table, the generation of charging action is stopped and the vehicle start status is checked; when the current action position is still within the action position range, only the charging control parameters corresponding to the current action position are updated, and the duty cycle adjustment direction is checked based on the actual test feedback. The partial replacement of the action position is used to isolate the impact of the next charging action on the subsequent control position; when S51 is executed, the electronic control device copies the frozen version of the charging action table item by item into the execution action table, replaces the original charging action of the current action position with the next charging action, and rereads the original charging actions of the other action positions from the charging action table and writes them into the execution action table. If the next charging action lacks a given value, duty cycle step, or number of execution steps, the original charging action at the current action position is retained and an action missing record is generated; after the update is completed, the execution action table and the current action position are written to the current control register for duty cycle calculation. The feedback relationship determines the direction of the candidate duty cycle value; when the current charging action and the previous charging action use the same control method, the current response slope is the difference between the feedback quantities of the last two steps of the previous charging action divided by the difference in duty cycles; when the control method is switched or a positive response slope is not obtained, the first two steps of response identification are re-executed. After entering S52, the electronic control device reads the setpoint and duty cycle step of the current action position from the action table, subtracts the current feedback amount from the setpoint to form the control difference, and then divides the control difference by the current response slope to form the duty cycle correction amount. When the duty cycle correction exceeds the positive or negative duty cycle step, it is truncated according to the corresponding boundary, added to the current duty cycle, and then limited to the range of zero to one to form a candidate duty cycle; When the current response slope fixed-point code is zero or the sign is negative, the division is stopped, the current duty cycle is retained and written into the slope anomaly record; when the duty cycle step fixed-point code is zero and the current feedback quantity has not reached the given value, the current charging action is terminated. Direction verification is used to identify duty cycle misadjustment caused by local response changes. In S53, the electronic control device first subtracts the current duty cycle from the candidate duty cycle, then multiplies it by the current response slope and adds the current feedback quantity to obtain the predicted feedback quantity. Subsequently, the power converter is driven to run a control cycle with the candidate duty cycle, and the measured feedback quantity is read at the end of the cycle. When the difference between the predicted feedback quantity and the current feedback quantity and the difference between the measured feedback quantity and the current feedback quantity have the same sign, the candidate duty cycle is retained; when the two differences have opposite signs, the current duty cycle is restored in the next beat, the duty cycle step size is halved, and the duty cycle calculation is re-executed using the feedback quantity at the end of the trial beat as the new current feedback quantity. When both differences are zero, the current duty cycle is retained. When the predicted difference is non-zero but the measured difference is zero, it is treated as an inconsistency in direction. When the predicted difference is zero but the measured difference is non-zero, the current response slope is recalculated using a trial shot. Each run trial step is written into the duty cycle sequence according to the step number and marked as either retained or withdrawn. The trial step is counted in the execution count of the current charging action. When the duty cycle step is zero after fixed-point encoding, the formed duty cycle sequence is output and the current charging action ends. After the duty cycle direction is checked, the power converter continues to execute the remaining number of cycles from the cycle after the last actual running cycle, and the trial cycles that have already been run are not output repeatedly. During S54 execution, current control uses the current setpoint to adjust the duty cycle and synchronously limit the terminal voltage, while voltage control uses the voltage setpoint to adjust the duty cycle and synchronously limit the output current. When the terminal voltage reaches the discharge termination voltage or the battery temperature reaches the charging cutoff temperature, the electronic control device stops the power converter after the current cycle ends. After the current charging action has actually passed one control frame, the next charging action is added to the action sequence, and a start record is formed based on all actual running frames; After the current charging action ends, the next control step outputs a start-allowing signal to the vehicle start control terminal and reads the engine operating position sent by the engine controller through the vehicle bus. When the engine operating position is written to 1, the vehicle is considered to have started successfully. If the vehicle is not started and has not reached the termination value, repeat S2 to S4 with the updated action sequence to determine the next charging action at the next action position. The resulting execution process only replaces the current action position and retains the changes in terminal voltage, feedback quantity, and battery temperature caused by each probe in the startup record; Taking the current action position as the third position and the next charging action as current control as an example, after the electronic control device replaces the third parameter of the execution action table, it calculates the candidate duty cycle based on the current response slope; when the candidate duty cycle generates a reverse current change, the duty cycle step is halved and recalculated, and after generating a same-direction current change, it continues to complete the remaining number of steps. When the assisted start is completed, the start task number and action position form a record key, and the action sequence, duty cycle sequence, start record and assisted start result are written to the industrial big data record database at once. If the write fails, it is saved locally on the electronic control device. After communication is restored, it is rewritten in the order of record number, and the rewritten record is included in the subsequent data version.

[0022] exist Figure 1 It should be noted that: the flowchart shows the health status assessment, duty cycle adjustment, historical transport weighting, candidate charging action decision and current action position update in sequence from S1 to S5; the diamond box represents the judgment node of successful vehicle start or reaching the termination state, the solid arrow indicates the actual execution order, the negative loop indicates to continue to S2 to execute the next round of adjustment, and the positive arrow indicates to output the auxiliary start result and write it into the industrial big data record library. exist Figure 2 It should be noted that: the system diagram shows the status acquisition layer, electronic control device, power converter, vehicle DC power supply network and parameter and industrial big data record library from top to bottom; solid arrows indicate the actual transmission direction of sampled values, duty cycle, charging output and status feedback; the bidirectional arrows between the database and the electronic control device indicate the reading of control parameters and historical start records and the writing back of start records and auxiliary start results, respectively.

[0023] Furthermore, the present invention also includes an intelligent auxiliary power supply start-up control system based on battery health status, the system comprising a health assessment module, a charging adjustment module, a weight generation module, an action decision module, and a start-up execution module: The health assessment module is used to obtain the measured internal resistance, measured cold start current and static voltage of the rescued battery, convert and weight them to obtain the health status value, and read the corresponding first charging action from the charging action table. The charging adjustment module is used to adjust the duty cycle of the power converter step by step according to the current charging action after the auxiliary power supply and the rescued battery are connected in parallel to the vehicle's DC power supply network, control the charging current or charging voltage, form a start record, and add the current charging action to the action sequence. The weight generation module is used to read historical start-up records from the industrial big data record library according to battery type and action sequence, group them according to the original charging action and another charging action, generate transportation costs according to the state difference before the action, and use the unbalanced Sinkhorn algorithm to iteratively obtain the transportation matrix, and form transportation weights by row sums. The action decision module uses a causal forest algorithm based on transportation weight to divide split records and estimation records. It recursively divides the records according to the midpoint of the status field. In the leaf node corresponding to the start record, it calculates the protection trigger difference, start success difference, and response burden difference of another charging action relative to the original charging action to determine the next charging action. The start-up execution module is used to update the current action position according to the next charging action and keep other action positions unchanged. The electronic control device adjusts the duty cycle of the power converter and repeats S2 to S4 until the vehicle starts successfully or reaches the nominal termination state. The auxiliary start-up result is output and written to the industrial big data record library.

[0024] Overall execution and implementation instructions: This plan first collects the internal resistance, cold start current and static voltage of the battery being rescued, and then verifies the internal resistance through two different currents for verification charging. Based on this, the health status value is calculated and the first charging action is determined. After the auxiliary power supply is connected in parallel with the battery being rescued to the vehicle's DC power supply network, the electronic control device adjusts the duty cycle of the power converter step by step according to the feedback current or feedback voltage to form a start-up record and action sequence. Subsequently, historical start-up records with consistent battery type and action sequence are selected from the industrial big data record library. The unbalanced Sinkhorn algorithm is used to allocate transportation weights according to the state before the action. Then, the causal forest algorithm is used to compare the actual impact of different charging actions on protection triggering, start-up success, and response burden, and the next charging action is determined from this. The electronic control device only updates the current action position and continues to adjust the charging current or charging voltage until the vehicle starts successfully or reaches the termination state. At the same time, the complete start-up data is written back to the industrial big data record library. Taking a vehicle with a depleted battery in a low-temperature environment as an example, after the auxiliary power is connected, the static voltage of the battery is read and two-stage verification charging is performed. The internal resistance is corrected by the terminal voltage response. If the calculated health status value indicates that the battery still has the ability to accept charging, the electronic control device will increase the duty cycle of the power converter step by step according to the corresponding charging action, so that the charging current is stable and close to the given value. After each charging action is completed, the system compares the current state with the previous actions in the history. First, it uses the unbalanced Sinkhorn algorithm to eliminate the difference in the number of samples and the state. Then, it uses the causal forest algorithm to determine the impact of continuing to increase the current, maintaining the output, or reducing the output on the success of the start-up and the battery load, and selects the next charging action accordingly. If the terminal voltage and battery temperature remain within the allowable range, the auxiliary power supply continues to output and allows the vehicle to start; after the vehicle starts, the output stops and the sequence of actions and the start-up result are saved for subsequent charging control of similar batteries.

[0025] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent start-up control method for auxiliary power supply based on battery health status, characterized in that, include: S1. Obtain the measured internal resistance, measured cold start current and static voltage of the battery being rescued, convert and weight them to obtain the health status value, and read the corresponding first charging action from the charging action table. S2. After the auxiliary power supply and the rescued battery are connected in parallel to the vehicle's DC power supply network, the electronic control device adjusts the duty cycle of the power converter step by step according to the current charging action, controls the charging current or charging voltage, forms a start record, and adds the current charging action to the action sequence. S3. Read historical startup records from the industrial big data record library according to battery type and action sequence, group them according to the original charging action and another charging action, generate transportation costs according to the state difference before the action, and use the unbalanced Sinkhorn algorithm to iteratively obtain the transportation matrix, and form transportation weights by row sums. S4. Based on the transport weight, the causal forest algorithm is used to divide the split record and the estimated record. The record is recursively split according to the midpoint of the state field. In the leaf node corresponding to the start record, the protection trigger difference, start success difference and response burden difference of the other charging action relative to the original charging action are calculated to determine the next charging action. S5. Update the current action position according to the next charging action and keep other action positions unchanged. The electronic control device adjusts the duty cycle of the power converter and repeats S2 to S4 until the vehicle starts successfully or reaches the nominal termination state. Output the auxiliary start result and write it into the industrial big data record library.

2. The intelligent start-up control method for auxiliary power supply based on battery health status according to claim 1, characterized in that: S1 includes: S11. After the static voltage sampling is completed, the electronic control device controls the power converter to output the first verification charging segment and the second verification charging segment in sequence, so that the charging current of the second verification charging segment is twice that of the first verification charging segment and the number of execution steps is half that of the first verification charging segment, and stops outputting the same number of steps after each verification charging segment to form a voltage response sequence. S12. Based on the voltage response sequence, divide the voltage difference between the first current-on beat and the last current-off beat of each verification charging segment by the corresponding charging current to obtain two dynamic internal resistance values. Correct the measured internal resistance with the arithmetic mean of the two dynamic internal resistance values. Then normalize the absolute values ​​of the internal resistance health value, cold start health value and voltage health value deviating from one to weights respectively, and sum them according to the weights to obtain the health status value. S13. Locate the value segment in the charging action table according to the health status value, read the current setpoint or voltage setpoint, duty cycle step and execution number recorded in the value segment, form the first charging action and write it into the electronic control device.

3. The intelligent start-up control method for auxiliary power supply based on battery health status according to claim 2, characterized in that: S2 includes: S21. After the auxiliary power supply and the rescued battery are connected in parallel to the vehicle's DC power supply network, the set value, starting duty cycle, duty cycle step distance and execution number of the current charging action are read. In the first two steps, the electronic control device drives the power converter with the starting duty cycle and the starting duty cycle plus the duty cycle step distance respectively. The difference between the feedback quantities of the two steps is divided by the duty cycle step distance to obtain the initial response slope. The feedback quantity and the set value are both charging current or charging voltage. S22. Starting from the third beat, subtract the current beat feedback value from the given value to obtain the control difference. Divide the control difference by the current response slope to obtain the duty cycle correction value. Add the duty cycle correction value to the current beat duty cycle. Write zero when the sum is less than zero, write one when the sum is greater than one, and retain the sum when the sum is between zero and one to form the next beat duty cycle. Then, divide the difference between the feedback values ​​of two adjacent beats by the difference between the duty cycles of two adjacent beats to update the current response slope. Execute beat by beat until the number of beats is reached, and output the duty cycle sequence. S23. Drive the power converter to adjust the charging current or charging voltage of the auxiliary power supply according to the duty cycle sequence. Calculate the voltage drop and battery temperature rise based on the start and end of the action to form a start record and add the current charging action to the action sequence.

4. The intelligent start-up control method for auxiliary power supply based on battery health status according to claim 3, characterized in that: S3 includes: S31. Freeze the same data version from the industrial big data record library according to the battery type and action sequence. Form a source group by taking the historical start record of the original charging action for the next action, and form a target group by taking the historical start record of each candidate charging action for the next action. Then, splice the health status value and ambient temperature to the terminal voltage and battery temperature at the start of the action to form a status column, and output the source status matrix and the target status matrix. S32. Calculate the median and median absolute deviation of each state field based on the source state matrix and the target state matrix. Subtract the median from the field value and divide by the median absolute deviation to form a dimensionless state matrix. Write zero for fields with a median absolute deviation of zero. Then perform singular value decomposition on the covariance matrix of the dimensionless state matrix and take the reciprocal of the non-zero singular values ​​to form a generalized inverse matrix. Multiply the difference vector between the source state column and the target state column on both sides of the generalized inverse matrix to obtain the transportation cost matrix.

5. The intelligent start-up control method for auxiliary power supply based on battery health status according to claim 4, characterized in that: S3 also includes: S33. The source marginal quantity and target marginal quantity are formed by the reciprocal of the number of records in the source group and the reciprocal of the number of records in the target group. The median of the positive values ​​in the transportation cost matrix is ​​used as the entropy regularization quantity, and the transportation kernel is formed by dividing the negative values ​​of the transportation cost by the exponent of the entropy regularization quantity. When the transportation cost matrix is ​​all zero, the transportation kernel is written as a matrix of all ones. Then, the unbalanced Sinkhorn algorithm is used to alternately update the source scaling quantity by dividing the source marginal quantity by the weighted row of the transportation kernel and the square root of the result, and update the target scaling quantity by dividing the target marginal quantity by the weighted column of the transportation kernel and the square root of the result, until the transportation matrices of two adjacent rounds are consistent bit by bit under the fixed-point format of the electronic control device, and the initial transportation matrix is ​​output. S34. The Dixtra projection algorithm is used to alternately perform row quality projection and column quality projection on the initial transportation matrix, so that the quality of each row does not exceed the corresponding source margin and the quality of each column does not exceed the corresponding target margin. The projection stops when the projection matrices of two adjacent rounds are consistent bit by bit under the fixed-point format of the electronic control device, and the transportation weights of the historical start record are formed by the sum of the rows of the projection matrix.

6. The intelligent start-up control method for auxiliary power supply based on battery health status according to claim 5, characterized in that: S4 includes: S41. For each candidate charging action, read the positive transport element in the transport matrix, form a pairing record by connecting the original charging action record and the candidate charging action record with the positive transport element, form a pairing state by the arithmetic mean of the state fields before the action of the two records, form an action difference vector by subtracting the protection trigger value, start success value and response burden value of the original charging action record from the candidate charging action record, and form a pairing weight with the positive transport element. S42. Circularly shift the pairing record numbers to form the record order. In each record order, treat odd-numbered pairing records as split records and even-numbered pairing records as estimated records. Use the honest causal forest algorithm to split the records sequentially by the midpoint of the adjacent values ​​of each field in the pairing state. Calculate the mean of the pairing weights of the action difference vectors in the two child nodes respectively. Divide the product of the sum of the pairing weights of the two child nodes by the square of the sum of the pairing weights of the parent node, and then multiply by the sum of the squares of the differences between the two mean vectors to obtain the splitting value. Read the first and second midpoints in descending order of the splitting value to form tree nodes until there are no midpoints in the node. Output the action difference tree.

7. The intelligent start-up control method for auxiliary power supply based on battery health status according to claim 6, characterized in that: S4 also includes: S43. Send the current start-up record's pre-action state into each action difference tree. Calculate the mean of the action difference vector in the estimated record falling into the leaf node according to the pairing weight. Then calculate the median and median absolute deviation for the same component output by each action difference tree. Use the median minus the median absolute deviation to form the lower bound of the action difference, and use the median plus the median absolute deviation to form the upper bound of the action difference. Output the action difference interval of the candidate charging action. S44. First, arrange the candidate charging actions in ascending order according to the upper limit of the protection trigger difference. Then, arrange the candidate charging actions in the same position in descending order according to the lower limit of the start success difference. Subsequently, arrange the candidate charging actions that are still in the same position in ascending order according to the upper limit of the response burden difference. Read the first candidate charging action as the next charging action. When there are fewer than two positive transport elements, retain the original charging action and have the electronic control device read the current setpoint or voltage setpoint of the next charging action to update the charging current or charging voltage at the current action position.

8. The intelligent start-up control method for auxiliary power supply based on battery health status according to claim 7, characterized in that: S5 includes: S51. Read the next charging action and the current action position, copy the charging action table to the execution action table, replace the original charging action at the current action position in the execution action table with the next charging action, restore the other action positions to the corresponding values ​​in the charging action table, and output the updated execution action table. S52. Read the given value and duty cycle step of the current action position according to the action table. Divide the difference between the given value and the current feedback amount by the current response slope to obtain the duty cycle correction amount. Limit the duty cycle correction amount to between positive and negative duty cycle step sizes and add it to the current duty cycle. Write zero when the sum is less than zero, write one when the sum is greater than one, and retain the sum when the sum is between zero and one to form a candidate duty cycle.

9. The intelligent start-up control method for auxiliary power supply based on battery health status according to claim 8, characterized in that: S5 also includes: S53. The electronic control device drives the power converter to run one control step with the candidate duty cycle. It calculates the predicted feedback quantity corresponding to the candidate duty cycle based on the current response slope. When the direction of change of the predicted feedback quantity is consistent with the direction of change of the measured feedback quantity, the candidate duty cycle is retained. When the two change directions are opposite, the current duty cycle is restored and the duty cycle step size is halved before S52 is executed again until the two change directions are consistent or the duty cycle step size is zero after fixed-point encoding. The duty cycle sequence is then output. S54. Adjust the charging current or charging voltage output by the auxiliary power supply to the battery being rescued according to the duty cycle sequence, add the next charging action to the action sequence and repeat S2 to S4 until the vehicle start signal appears, or either the terminal voltage or the battery temperature reaches the corresponding termination value in the nominal parameter table, output the auxiliary start result, and associate the start record with the auxiliary start result and write it into the industrial big data record database.

10. An intelligent auxiliary power supply start-up control system based on battery health status, used to implement the intelligent auxiliary power supply start-up control method based on battery health status as described in any one of claims 1-9, the system comprising a health assessment module, a charging adjustment module, a weight generation module, an action decision module, and a start-up execution module, characterized in that: The health assessment module is used to obtain the measured internal resistance, measured cold start current and static voltage of the rescued battery, convert and weight them to obtain the health status value, and read the corresponding first charging action from the charging action table. The charging adjustment module is used to adjust the duty cycle of the power converter step by step according to the current charging action after the auxiliary power supply and the rescued battery are connected in parallel to the vehicle's DC power supply network, control the charging current or charging voltage, form a start record, and add the current charging action to the action sequence. The weight generation module is used to read historical start-up records from the industrial big data record library according to battery type and action sequence, group them according to the original charging action and another charging action, generate transportation costs according to the state difference before the action, and use the unbalanced Sinkhorn algorithm to iteratively obtain the transportation matrix, and form transportation weights by row sums. The action decision module uses a causal forest algorithm based on transportation weight to divide split records and estimation records. It recursively divides the records according to the midpoint of the status field. In the leaf node corresponding to the start record, it calculates the protection trigger difference, start success difference, and response burden difference of another charging action relative to the original charging action to determine the next charging action. The start-up execution module is used to update the current action position according to the next charging action and keep other action positions unchanged. The electronic control device adjusts the duty cycle of the power converter and repeats S2 to S4 until the vehicle starts successfully or reaches the nominal termination state. The auxiliary start-up result is output and written to the industrial big data record library.

Images