A multi-cluster parallel coordination control method of a lithium iron phosphate energy storage system

Abstract

The application discloses a kind of multi-cluster parallel coordination control methods of lithium iron phosphate energy storage system, specifically related to lithium iron phosphate energy storage system parallel coordination control field, including obtaining by energy storage control device and write in industrial big data of control beat, industrial big data includes the end voltage of each battery cluster, branch current, branch impedance, branch power, contactor state, power instruction of grid load balance and other direct current power;With the control beat of target battery cluster contactor from closed state to open state as exit beat;By using historical battery cluster exit data to determine the branch power migration relationship after target battery cluster exits, and combining the remaining battery cluster power boundary to form pre-exit preset power and exit distribution increment in advance;Through historical exit sample mapping migration power after target battery cluster exits, and pre-setting the power of remaining battery cluster before exit, the initial current surge can be relatively inhibited, and the power mismatch caused by secondary current limiting, derating or exit is reduced.

Description

Technical Field

[0001] This invention relates to the field of parallel coordinated control technology for lithium iron phosphate energy storage systems, and more specifically, to a multi-cluster parallel coordinated control method for lithium iron phosphate energy storage systems. Background Technology

[0002] With the development of the new energy vehicle industry, lithium iron phosphate batteries are widely used in the power battery systems of passenger cars, commercial vehicles and other new energy vehicles due to their cycle life, thermal stability and cost characteristics. The cell consistency control, module integration, battery management and charge and discharge control technologies developed in the power battery manufacturing process are also gradually extending to the energy storage field. Some lithium iron phosphate cells and modules that use the same material system and manufacturing system as the power batteries of new energy vehicles are further formed into energy storage battery clusters, and multiple battery clusters are connected in parallel to the DC bus to form an energy storage power station. In such lithium iron phosphate energy storage systems, the energy storage management device typically allocates station-level power based on the state of charge, allowable charge and discharge power, terminal voltage, temperature, and operating status of each battery cluster. When a battery cluster needs to be taken out of operation due to abnormal single-cell voltage, abnormal temperature, branch fault, life status, or maintenance, the corresponding battery cluster is usually disconnected first, and then the power borne by each branch is re-determined based on the operating status of the remaining online battery clusters. Taking the continuous execution of grid discharge commands by the energy storage power station as an example, before the power is withdrawn, multiple battery clusters share the predetermined output power, and the target battery cluster already has the corresponding branch current. When the grid power demand remains unchanged, after the target battery cluster is disconnected, its original branch current needs to be taken over by the other parallel branches. The contactor action, current redistribution, and establishment of the new branch power allocation result are sequential processes. Due to differences in cell condition, terminal voltage, connection impedance, and online assembly among different battery clusters, the initial current migration after the target battery cluster exits is not entirely consistent with the subsequent power allocation result. Some remaining battery clusters will initially take on more branch current after the exit action occurs, and then switch to the redistributed operating state. For multi-cluster energy storage systems formed by extending the power battery technology system of new energy vehicles, different manufacturing batches, service conditions, and electrical connection conditions will also cause differences in the degree of current absorption of each battery cluster in the same exit event. When the remaining battery clusters bearing the initial migration current are close to the allowable charge and discharge range, the short-term increase in branch current can easily trigger current limiting, derating, or shutdown, causing the shutdown of a single battery cluster to further restrict other battery clusters and causing the actual output power of the energy storage power station to deviate from the power supplied by the grid. It is difficult to handle the branch current migration that has already formed in the early stage of the shutdown action if the power is only redistributed according to the current state of the remaining battery clusters after the target battery cluster is shut down. The technical problem to be solved by this application is: how to use the multi-cluster operation data of the lithium iron phosphate energy storage system before and after the withdrawal of each battery cluster to determine the branch current migration of each remaining battery cluster after the withdrawal of the target battery cluster, and adjust the power load of each remaining battery cluster in advance accordingly, so as to reduce the grid-side output power mismatch caused by the secondary current limiting, derating or withdrawal of other battery clusters during the withdrawal of the target battery cluster. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a multi-cluster parallel coordinated control method for lithium iron phosphate energy storage systems. This method utilizes historical battery cluster exit data to determine the branch power migration relationship after the target battery cluster exits, and combines the remaining battery cluster power boundaries to pre-form the pre-exit power and the post-exit allocation increment, thereby solving the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a multi-cluster parallel coordinated control method for a lithium iron phosphate energy storage system, comprising: S1. Obtain the industrial big data written by the energy storage control device according to the control step. The industrial big data includes the terminal voltage, branch current, branch impedance, branch power, contactor status, grid load balancing power command and other DC power of each battery cluster. Take the control step when the contactor of the target battery cluster changes from the closed state to the open state as the exit step. Extract the data from the step before exit to the time before the grid load balancing power command changes to form the exit sample. S2. For the exited sample, the equivalent voltage of each battery cluster is obtained by subtracting the product of the branch current and the branch impedance from the terminal voltage of the previous step before exiting. The equivalent voltages of any two battery clusters are subtracted and their absolute values ​​are taken to form a relation matrix. The branch current that increases along the current direction of the target battery cluster in the first step after exiting is taken as the receiving current. The migration ratio is obtained by dividing each receiving current by the sum of all receiving currents. S3. After determining the target battery cluster, generate the current relationship matrix as in S2. Use the current remaining battery cluster branch power ratio and the exited sample remaining battery cluster branch power ratio as the current transport quality and sample transport quality, respectively. Multiply each current transport quality by each sample transport quality to form an initial transport matrix, which serves as the front-wheel transport matrix. For each transport position determined by the current remaining battery cluster and the sample remaining battery cluster, subtract the relationship value between the current remaining battery cluster and another current remaining battery cluster in the current relationship matrix and the relationship value between the sample remaining battery cluster and another sample remaining battery cluster in the sample relationship matrix, square the result, and multiply it by the relationship value between the other current remaining battery cluster and another sample remaining battery cluster in the front-wheel transport matrix. The transportation volume at a fixed location is calculated, and the transportation cost is summed for all other current remaining battery clusters and other sample remaining battery clusters. The transportation matrix is ​​updated using a cost scaling push flow algorithm, and the current migration ratio is obtained by weighting the sample migration ratio with the transportation matrix and normalizing it according to the current transportation quality. The current power of the target battery cluster is allocated according to the current migration ratio, and the step size is adjusted and rounded according to the PCS power to obtain the migration power. When the migration power vector of this round is consistent with the migration power vector of a previously saved round, the round between the two rounds is determined as the repetition segment. The relationship distortion value is obtained by multiplying the transportation cost by the transportation volume at the same location and summing the results. The migration power corresponding to the first position of the relationship distortion value in the repetition segment and each exited sample in ascending order is read. S4. Using the current power of the target battery cluster as the power to be allocated, divide the allocable capacity between the current power and the corresponding BMS power boundary of each remaining battery cluster into power steps according to the PCS power adjustment step. For each power step of each remaining battery cluster, subtract the square of the difference between the cumulative allocated power after adding the corresponding power step and the corresponding migration power from the square of the difference between the cumulative allocated power before adding the corresponding power step to obtain the edge cost corresponding to the power step. Push the flow along the negatively reduced cost remaining edge and shift the cost scale to the right step by step through the cost scaling push flow algorithm until the power to be allocated is zero to obtain the allocation increment of each remaining battery cluster. S5. The preset power is obtained by adding the allocation increment to the current branch power of each remaining battery cluster and then subtracting the migration power. The preset power is executed while maintaining the current branch power of the target battery cluster, so that the algebraic sum of each preset power, the current branch power of the target battery cluster, and the current power of other DC power sources in the same power direction is equal to the grid load balancing power command in the same step. In the next control step after the preset power is executed, the target battery cluster is disconnected, so that the remaining battery clusters are transferred to the sum of the current branch power and the allocation increment, and the migration ratio is updated according to the difference in branch current before and after disconnection and then written back to the industrial big data. Among them, the control phase is a control cycle in which the energy storage control device completes branch data acquisition, power calculation and power command issuance in sequence according to a fixed control cycle; the first phase after exit is the first complete sampling control phase after the target battery cluster contactor is disconnected.

[0005] In a preferred embodiment, S1 includes: S11. The exit power is obtained by multiplying the bus voltage of the previous phase by the branch current of the target battery cluster, and the grid load balancing power command of the same phase is written into the exit sample as the power reference. The power reference indicates the total power that the energy storage system should maintain in a uniform power direction before the target battery cluster exits. Among them, the grid load balance power instruction refers to the charging or discharging power target value generated by the energy storage energy management device based on the active power deficit or surplus on the grid side and sent to the lithium iron phosphate energy storage system. It is used to determine the total power that the energy storage device needs to absorb from or output to the DC bus at the corresponding control step. A battery cluster refers to a battery branch consisting of multiple lithium iron phosphate battery cells connected in series and parallel, with an independent branch current acquisition circuit and contactor, and connected to the DC bus through the contactor to independently undertake the power distribution. S12. For each control phase after exiting the phase, the DC compensation power is obtained by subtracting the power of the previous phase from the current power of other DC power sources, and then the net exit power that the remaining battery cluster needs to take over is obtained by subtracting the DC compensation power from the exit power. Among them, the control step refers to a control cycle in which the energy storage control device completes branch data acquisition, power calculation and power command issuance in sequence according to a fixed control cycle. Adjacent control steps are numbered in the same control cycle sequence. The contactor state change and the corresponding branch power are all attributed to the control step that has taken the sampling or execution action. S13. For each remaining battery cluster, subtract the branch power corresponding to the previous step from the current branch power to obtain the branch increment; write zero when the branch increment is less than zero, sum all the positive values ​​after writing to obtain the total positive value, divide each positive value by the total positive value and multiply by the net exit power to obtain the migration reference amount of the corresponding remaining battery cluster, and write the migration reference amount into the exit sample with the corresponding control step until the grid load balance power command changes.

[0006] In a preferred embodiment, S2 includes: S21. For each exited sample, read the terminal voltage, current and branch impedance of the previous step according to the battery cluster number, subtract the product of current and branch impedance from the terminal voltage to obtain the cluster equivalent voltage, and form an equivalent voltage sequence according to the battery cluster number. S22. Based on the equivalent voltage sequence, calculate the absolute value of the cluster equivalent voltage difference between any two battery clusters, and write the corresponding values ​​according to the row and column positions determined by the two battery cluster numbers, and write zeros according to the positions of the same cluster, to obtain a relation matrix symmetric about the main diagonal. The row and column positions refer to the positions of matrix elements in the relation matrix determined by the row and column numbers corresponding to the two battery clusters, respectively. The row number corresponds to the first battery cluster, and the column number corresponds to the second battery cluster. The absolute value of the equivalent voltage difference between the two battery clusters is written at this position. The main diagonal refers to the set of positions formed by matrix elements with the same row and column numbers in the relation matrix. Each position corresponds to the equivalent voltage difference between the same battery cluster and itself, so the corresponding value is written as zero.

[0007] In a preferred embodiment, S2 further includes: S23. In the first step after exiting, take the current direction of the target battery cluster in the previous step as the migration direction, record the current increase of each remaining battery cluster relative to the previous step along the migration direction as the receiving current, and write the reverse change part as zero to obtain the receiving current of each remaining battery cluster. S24. The total capacity is obtained by summing the current of each remaining battery cluster. The migration ratio is obtained by dividing the current of each remaining battery cluster by the total capacity. The migration ratio is then bound to the corresponding exit sample in the relationship matrix. The migration power of the target battery cluster before exiting the power grid during the load balancing period is calculated.

[0008] In a preferred embodiment, S3 includes: S31. Determine the target battery cluster and generate the current relationship matrix according to S2. Use the current remaining battery cluster branch power ratio to form the current transport quality, use the remaining battery cluster branch power ratio of the exited sample to form the sample transport quality, and multiply the two transport qualities item by item to form the initial transport matrix. S32. For each transportation location, the transportation location is determined by the row position corresponding to a current remaining battery cluster and the column position corresponding to a sample remaining battery cluster. Traverse the relationship values ​​between the current remaining battery cluster and another current remaining battery cluster corresponding to the row position in the current relationship matrix, and the relationship values ​​between the sample remaining battery cluster and another sample remaining battery cluster corresponding to the column position in the sample relationship matrix. Subtract the two relationship values, square them, multiply them by the transportation volume determined by the other current remaining battery cluster and another sample remaining battery cluster in the previous round transportation matrix, and accumulate all other current remaining battery clusters and other sample remaining battery clusters to obtain the transportation cost of the corresponding transportation location. After completing the calculation of all transportation locations, a transportation cost matrix is ​​formed. The relation value refers to the matrix element corresponding to the two battery clusters in the current relation matrix or sample relation matrix. The value is the absolute value of the difference in the equivalent voltage of the two battery clusters, which is used to represent the difference in electrical relationship between the two battery clusters in the corresponding exit state. Transport volume refers to a matrix element in the transport matrix that is jointly determined by a current remaining battery cluster and a sample remaining battery cluster. It represents the mass share of the current remaining battery cluster's transport mass allocated to the corresponding sample remaining battery cluster. The sum of transport volumes in the same row equals the corresponding current transport mass, and the sum of transport volumes in the same column equals the corresponding sample transport mass.

[0009] In a preferred embodiment, S3 further includes: S33. Limit the row sum and column sum of the transportation matrix with the current transportation quality and the sample transportation quality respectively. Distribute the transportation volume along the negative cost remaining edge through the cost scaling push flow algorithm. When there is no corresponding remaining edge, adjust the node potential according to the cost reduction of the remaining edge and reduce the cost scale step by step until the row sum and column sum satisfy the corresponding transportation quality, and obtain the transportation matrix of this round. The process of using the cost scaling push flow algorithm is as follows: When using the cost scaling push flow algorithm, a starting side node is established based on the current remaining battery clusters, and an ending side node is established based on the sample remaining battery clusters. The amount to be pushed by the starting side node is taken as the corresponding current transportation quality, and the amount to be received by the ending side node is taken as the corresponding sample transportation quality. A transportation edge is established between the two sides of the node according to the transportation cost matrix. Each time, a transportation edge with negative reduced cost and both the starting side node and the ending side node still have an amount to be pushed and received is read. The transportation amount is determined by the value of the starting side's amount to be pushed and the ending side's amount to be received first, and it is deducted simultaneously. When there is no negative reduced cost transportation edge, the node potential is adjusted and the cost scale is reduced, and transportation continues until all the amounts to be pushed and received are reduced to zero, forming the transportation matrix for this round. The node potential refers to the cost correction value attached to the current remaining battery cluster node and the sample remaining battery cluster node, respectively, which is used to convert the original transportation cost of the transportation edge into the reduced cost. When there is no negative reduced cost transportation edge, the node potential is corrected according to the reduced cost of the remaining edge associated with the current node, so that the subsequent transportation edge can re-participate in the transportation volume allocation. The cost scale refers to the integer scale used in the cost scaling push flow algorithm to process transportation costs in stages. In the first round, the cost scale is taken as an integer power of two, which is no less than the first value of the absolute value of the transportation cost. After each round of transportation is completed, the cost scale is divided by two until it is reduced to one, thereby completing the solution of the transportation matrix with a cost resolution from coarse to fine. S34. The current migration ratio is obtained by weighting the sample migration ratio with the current transportation matrix and normalizing it according to the current transportation quality. Then, the current power of the target battery cluster is allocated according to the current migration ratio, and the PCS power adjustment step is rounded and the rounding margin is supplemented to obtain the migration power. The process of weighting the sample migration ratio and normalizing it according to the current transportation quality is as follows: For each current remaining battery cluster, read the transportation volume in the same row of the current round transportation matrix one by one, multiply each transportation volume by the sample migration ratio of the corresponding sample remaining battery cluster, sum them up, and then divide by the current transportation quality corresponding to the current remaining battery cluster to obtain the current migration ratio of the current remaining battery cluster. PCS power refers to the active power of the power conversion system when it performs bidirectional power conversion between the DC bus and the grid side in accordance with the grid load balancing power command. S35. Return the current round of transportation matrix to S32 to recalculate the transportation cost and execute S33 and S34 in sequence until the migration power is the same as the migration power of the previous round. Use the sum of the product of transportation cost and corresponding transportation volume as the relationship distortion value, and read the migration power of the first-ranked relationship distortion value in the repeated segment and each exit sample.

[0010] In a preferred embodiment, S4 includes: S41. Read the allocable capacity of each remaining battery cluster from the current power to the BMS power boundary in the current power direction of the target battery cluster. Divide the power steps step by step according to the power adjustment step of each power command change of the PCS, and retain the last segment that is less than one power adjustment step separately to form the power edge of each remaining battery cluster. The BMS power boundary refers to the branch power limit given by the battery management system based on the current allowable charging power or allowable discharging power of the corresponding battery cluster. When the current power direction of the target battery cluster is discharging, the allowable discharging power limit is taken; when the current power direction of the target battery cluster is charging, the allowable charging power limit is taken. Capacity refers to the amount of additional power that the remaining battery cluster can handle between the current branch power change and the corresponding BMS power boundary along the current power direction of the target battery cluster. Its value is the absolute value of the difference between the corresponding BMS power boundary and the current branch power along the same power direction. S42. For each power edge, the edge cost is obtained by subtracting the square of the difference between the cumulative allocated power after adding the corresponding power step and the migration power of the corresponding remaining battery cluster in S3 from the square of the difference between the cumulative allocated power before adding the corresponding power step and the corresponding migration power. The migration power is the power amount after the target battery cluster's current power is allocated according to the current migration ratio obtained in S3 and rounded according to the PCS power adjustment step. The corresponding power step is used as the edge capacity to form a push network from the target battery cluster's current power to each remaining battery cluster.

[0011] In a preferred embodiment, S4 further includes: S43. Determine the cost scale by reading the binary bit width of the first value of the absolute value of the edge cost in descending order through the cost scaling push algorithm. Obtain the reduced cost by adding the edge cost to the starting node potential and subtracting the ending node potential. Push the value that is exhausted first between the node remaining capacity and the edge remaining capacity along the remaining edge with negative reduced cost. When there is no remaining edge with negative reduced cost, read the first value of the edge reduced cost in ascending order and combine it with the cost scale to correct the node potential. After each round, shift the cost scale one bit to the right. After the cost scale drops to one, continue the calculation until the power to be allocated is zero and the reduced cost of all remaining edges is non-negative. S44. The allocation increment is obtained according to the push flow received by each remaining battery cluster, so that the sum of the current power of each remaining battery cluster and the allocation increment does not exceed the corresponding BMS power boundary, and the sum of the allocation increment and the compensation power of other DC power sources are equal to the current power of the target battery cluster, so as to provide the energy storage device with the power to maintain the grid load balance. The push flow refers to the amount of power actually allocated to the corresponding remaining battery clusters along the power edge in the push network by the cost scaling push algorithm. Each push flow takes the power that is not yet allocated at the starting node and the edge capacity that is not yet occupied by the corresponding power edge that is exhausted first. The push flows of multiple power edges are merged according to the remaining battery clusters to form the allocation increment.

[0012] In a preferred embodiment, S5 includes: S51. For each remaining battery cluster, the preset power is obtained by adding the current branch power to the allocation increment and then subtracting the migration power. The power margin generated by rounding each preset power according to the PCS power adjustment step is sequentially added to the remaining battery clusters that have not reached the BMS power boundary, so that each preset power, the current branch power of the target battery cluster and the current power of other DC power sources are kept as the grid load balance power command according to the algebraic sum of the unified power direction. S52. When the target battery cluster maintains the current branch power, a preset power is issued to each remaining battery cluster. The actual branch power is read step by step, and the first step in which the actual branch power of each remaining battery cluster is rounded according to the PCS power adjustment step and is consistent with the preset power is determined as the exit execution step.

[0013] In a preferred embodiment, S5 further includes: S53. In the next control step after exiting the execution step, disconnect the target battery cluster and simultaneously change the power command of each remaining battery cluster to the sum of the current branch power and the allocation increment, so that the algebraic sum of each preset power before disconnection, the current branch power of the target battery cluster and the current power of other DC power sources in the same power direction, and the algebraic sum of the current branch power and the allocation increment of each remaining battery cluster and the current power of other DC power sources in the same power direction after disconnection are all equal to the same grid load balancing power command. S54. Obtain the actual current received by the difference between the remaining battery cluster branch currents before and after the exit execution step. Sum the actual current received along the current direction before the target battery cluster exits to obtain the updated migration ratio. Write the updated migration ratio along with the exit execution step into the industrial big data.

[0014] The technical effects and advantages of this invention are as follows: 1. This solution maps the migration power of the target battery cluster after exiting by using historical exit samples, and pre-sets the remaining battery cluster power before exiting, which can relatively suppress the initial current surge and reduce the power mismatch caused by secondary current limiting, derating or exiting. 2. Based on the cluster equivalent voltage, a relation matrix is ​​constructed, and the Gromov-Wasestein optimal transmission matching is used to match the current and historical parallel structures, so that the historical migration pattern can be adapted to different online battery cluster combinations, thereby improving the pertinence of migration power determination. 3. By using cost scaling to push the flow, the power to be withdrawn is allocated under the constraints of BMS power boundary and PCS power adjustment step size, so that the allocation increment of each remaining battery cluster is limited by the actual carrying capacity, reducing the risk of power redistribution exceeding the allowable range of the branch. 4. The current branch power plus the distribution increment and then the migration power are used to form the preset power, so that the battery clusters expected to receive more migrations can form the bearing space in advance, which can relatively alleviate the transient impact between contactor disconnection and steady-state power redistribution. 5. The actual current received in the first step after exiting the system is recalculated as the updated migration ratio and written back to the industrial big data. This enables subsequent exit control to continuously use real operating results to correct the historical sample basis, thereby relatively improving the consistency between migration power calculation and field conditions. Attached Figure Description

[0015] Figure 1 This is a flowchart of the multi-cluster parallel coordination control method 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 The present invention provides a multi-cluster parallel coordinated control method for a lithium iron phosphate energy storage system, comprising: S1. Obtain the industrial big data written by the energy storage control device according to the control step. The industrial big data includes the terminal voltage, branch current, branch impedance, branch power, contactor status, grid load balancing power command and other DC power of each battery cluster. Take the control step when the contactor of the target battery cluster changes from the closed state to the open state as the exit step. Extract the data from the step before exit to the time before the grid load balancing power command changes to form the exit sample. To extract the actual power changes of a target battery cluster before and after its exit from industrial big data, it is necessary to first unify the power direction, control the timing of the shots, and define the effective range of the exit samples. Industrial big data is generated by energy storage control devices according to control batches. Each record includes at least the battery cluster number, terminal voltage, branch current, branch impedance, branch power, contactor status, grid load balancing power command and other DC power. Records within the same control batch are written according to the same power direction and control batch number, which is used to determine the branch electrical status, power changes and other DC power compensation results before and after the target battery cluster is removed. When the lithium iron phosphate battery cluster outputs power to the DC bus, it is recorded as a positive value; when it absorbs power from the DC bus, it is recorded as a negative value. The sampling records of each device are uniformly reset according to the control cycle of the energy storage control device. Each exit process is associated with only one target battery cluster, and the grid load balancing power command must remain unchanged from the cycle before exit to the first cycle after exit. This implementation process includes the following steps: The branch power actually borne by the target battery cluster before disconnection is used as the power starting point of the exit process; the bus voltage and the branch current of the target battery cluster are read in the control step before the contactor executes the disconnection action, and the two are multiplied to obtain the exit power. At the same time, the grid load balancing power command of the same step is saved as the power reference, thus completing S11. Exit the control step when the contactor changes from closed to open state, and exit the last complete sampling step when the contactor was in the closed state. If two battery clusters exit in the same control step, the target battery cluster exits with zero power, the contactor opens and then closes again, the power load balancing command changes before and after exit, or any record of bus voltage, branch current, or contactor status is missing, then the corresponding exit process will not be written into the exit sample. The power changes generated by other DC power sources on the DC bus during the shutdown period need to be deducted from the power output of the target battery cluster. For each control step after exiting the control step, first calculate the algebraic sum of the power of other DC power supplies connected to the same DC bus in the same positive and negative directions, then subtract the sum of the power of the previous control step from the sum of the power of the current control step to obtain the DC compensation power, and then subtract the DC compensation power from the exit power to obtain the net exit power, thus completing S12, and writing the net exit power into the exit sample according to the control step. If other battery clusters join or leave during the sample recording period, the current exiting sample will be terminated; if the net exit power direction is opposite to the exit power direction, the corresponding control unit will not participate in the subsequent migration calculation. The change in branch power of the remaining battery cluster after its withdrawal is used to determine the actual share of net withdrawn power among the branches. After unifying the power direction of each remaining battery cluster branch with the power direction before the target battery cluster exits, calculate the branch increment obtained by subtracting the power of the corresponding branch from the power of the branch before exiting the current control step, write the reverse increment to zero, and then sum the positive branch increments. Then, the net exit power is allocated according to the proportion of each forward branch increment to the total forward branch increment, and the migration reference amount of each remaining battery cluster is obtained, thus completing S13; No migration baseline is generated when the sum of positive branch increments is zero; the current exit sample is terminated when the net exit power direction is abnormal. The migration reference quantity is written into the industrial big data with each control cycle until the power grid load balance command changes. After the migration ratio is formed, the net exit power of the same cycle is allocated according to the migration ratio and rounded according to the PCS power adjustment step. Then it is checked against the migration reference quantity cluster by cluster. Exit samples with consistent rounding results continue to be read for migration power calculation, while exit samples with inconsistent results stop participating in subsequent calculations. When the energy storage device executes a 200kW discharge power command, the bus voltage of the target battery cluster before it exits is 750V and the branch current is 80A, so a 60kW exit power can be obtained; when the power of other DC power sources changes from 20kW to 28kW, the DC compensation power is 8kW, and the net exit power is 52kW. The branch increments of the three remaining battery clusters along the discharge direction are 20kW, 15kW and 17kW respectively. Therefore, the net power withdrawal of 52kW is allocated in the ratio of 20:15:17 to form the migration base quantities of 20kW, 15kW and 17kW. After the above processing, the exited samples can save the exit power of the target battery cluster, the compensation results of other DC power supplies, and the actual acceptance results of the remaining battery clusters, providing a unified data foundation for subsequent calculation of the relationship matrix and migration ratio.

[0018] S2. For the exited sample, the equivalent voltage of each battery cluster is obtained by subtracting the product of the branch current and the branch impedance from the terminal voltage of the previous step before exiting. The equivalent voltages of any two battery clusters are subtracted and their absolute values ​​are taken to form a relation matrix. The branch current that increases along the current direction of the target battery cluster in the first step after exiting is taken as the receiving current. The migration ratio is obtained by dividing each receiving current by the sum of all receiving currents. Before exiting the sample and proceeding to the subsequent structural calculations, the branch electrical state before exiting needs to be converted into unified data that can reflect the relative relationship of the battery clusters, and the actual receiving direction needs to be extracted from the first complete sampling result after exiting. Following the control step and power direction determined by S1, the branch current output by the battery cluster to the DC bus is recorded as positive, and the branch current absorbed from the DC bus is recorded as negative; after exiting, the first step is the first complete sampling step after the target battery cluster contactor is disconnected; The branch impedance is uniformly taken as the DC equivalent impedance between the battery cluster terminal voltage sampling point and the DC bus junction point, and the calibration value that was already in effect before exit is read and remains unchanged during the same exit sample calculation; the implementation process includes the following steps: Before exiting, each battery cluster needs to be converted to the same branch voltage reference. For a valid exit sample, the terminal voltage, branch current and branch impedance of the previous step are read according to the battery cluster number. The equivalent voltage of the cluster is obtained by subtracting the product of the branch current and the branch impedance from the terminal voltage. The equivalent voltage sequence is written according to the battery cluster number order, thus completing S21. The branch current participates in the multiplication along the aforementioned positive and negative directions, so that the discharge branch deducts the line voltage drop and the charging branch participates in the calculation as a negative current. If a branch impedance record is missing, the record time is later than the exit time, or the calibration endpoint is inconsistent with the specified endpoint, the corresponding exit sample will stop entering the relation matrix calculation. The equivalent voltage sequence can further fix the relative electrical relationship between each battery cluster; when executing S22, the row number and column number of the relationship matrix are determined simultaneously by the battery cluster number in the equivalent voltage sequence, the equivalent voltage of the two clusters are read one by one and the absolute value of the difference is calculated, and the result is written into the corresponding row and column positions. If two numbers are the same, write zero and write the calculation result to the position symmetrical about the main diagonal until all battery clusters are processed and the relation matrix is ​​obtained. The relation matrix retains all rows and columns of the online battery clusters before exiting and binds them to the exited samples; when entering the S3 transport calculation, the rows and columns corresponding to the historically exited battery clusters are deleted to form the historical remaining battery cluster relation matrix, thus allowing different exited samples to have different numbers of remaining battery clusters; The first complete sampling result after the contactor is disconnected is used to identify the actual branch current of the remaining battery clusters; the direction of the branch current before the target battery cluster exits is read, and the difference in branch current before and after the exit of each remaining battery cluster is unified in S23 according to this direction: when the branch current before the target battery cluster exits is positive, the branch current before the exit is subtracted from the branch current of the first exit after exiting. Positive values ​​are recorded as the current being received, and negative values ​​are written as zero. When the branch current before the target battery cluster exits is negative, the branch current before exiting is subtracted from the branch current after exiting. Positive values ​​are recorded as the receiving current, and negative values ​​are written as zero. If the branch current is zero before the target battery cluster exits, the migration direction cannot be determined, and no current is generated for the corresponding exit sample. If the branch current is missing, repeated and cannot be determined according to the data version in the first shot after exiting, or if the set of parallel battery clusters has changed, the corresponding exit sample will also be stopped. Once the current carrying capacity is determined, the current carrying capacity of each branch needs to be converted into a proportion that can be directly read in subsequent structural mapping. According to S24, the total current of all remaining battery clusters is summed to obtain the total current. Then, the migration ratio is obtained by dividing the current of each remaining battery cluster by the total current. The migration ratio and the relationship matrix are written into the same exit sample according to the battery cluster number. When the total current is zero, the division is not performed, and the corresponding exit sample is marked as unusable for S3. After the migration ratio is formed, the net exit power corresponding to the first step after exiting in S1 is read. The net exit power is allocated according to the migration ratio and the step size is adjusted and rounded according to the PCS power. Then, it is checked against the migration reference amount of the same step cluster by cluster. When the two sets of results are consistent, the exit sample is retained for S3 to read. When there is a difference, the exit sample is stopped from participating in the migration power calculation. For example, if the branch current of the target battery cluster is positive before it exits a certain exit sample, and the branch currents of the three remaining battery clusters are 60A, 55A and 50A respectively before they exit, and become 80A, 70A and 67A respectively after they exit, then the currents they receive are 20A, 15A and 17A respectively, and the total current they receive is 52A. The corresponding migration ratios are 20 / 52, 15 / 52 and 17 / 52 respectively. If the net exit power of S1 is 52kW, then 20kW, 15kW and 17kW are obtained according to the migration ratio. After the PCS power adjustment step size is processed, it is consistent with the migration reference amount of the same shot. The exit sample continues to enter S3. After this processing, the relation matrix preserves the relative electrical structure of the multiple clusters before exit, and the migration ratio preserves the actual acceptance distribution of the first batch after exit. The two parts of data are associated with the same exit sample, providing corresponding structural inputs and migration results for subsequent Gromov-Wasestein optimal transmission calculations.

[0019] S3. After determining the target battery cluster, generate the current relationship matrix as in S2. Use the current remaining battery cluster branch power ratio and the exited sample remaining battery cluster branch power ratio as the current transport quality and sample transport quality, respectively. Multiply each current transport quality by each sample transport quality to form an initial transport matrix, which serves as the front-wheel transport matrix. For each transport position determined by the current remaining battery cluster and the sample remaining battery cluster, subtract the relationship value between the current remaining battery cluster and another current remaining battery cluster in the current relationship matrix and the relationship value between the sample remaining battery cluster and another sample remaining battery cluster in the sample relationship matrix, square the result, and multiply it by the relationship value between the other current remaining battery cluster and another sample remaining battery cluster in the front-wheel transport matrix. The transportation volume at a fixed location is calculated, and the transportation cost is summed for all other current remaining battery clusters and other sample remaining battery clusters. The transportation matrix is ​​updated using a cost scaling push flow algorithm, and the current migration ratio is obtained by weighting the sample migration ratio with the transportation matrix and normalizing it according to the current transportation quality. The current power of the target battery cluster is allocated according to the current migration ratio, and the step size is adjusted and rounded according to the PCS power to obtain the migration power. When the migration power vector of this round is consistent with the migration power vector of a previously saved round, the round between the two rounds is determined as the repetition segment. The relationship distortion value is obtained by multiplying the transportation cost by the transportation volume at the same location and summing the results. The migration power corresponding to the first position of the relationship distortion value in the repetition segment and each exited sample in ascending order is read. After the target battery cluster enters the planned exit phase, the number of remaining battery clusters and the power bearing relationship in the current parallel structure are usually different from those in the historical exit samples. It is necessary to first convert the relationship matrix on both sides to a unified transport quality scale, and then use the inter-cluster electrical relationship to repeatedly correct the transport matrix so that the historical migration ratio can fall into the current remaining battery cluster. The current relation matrix is ​​generated using the calculation method of S2, and the rows and columns corresponding to the target battery cluster are deleted before entering the transportation calculation; the relation matrix of the exited sample is simultaneously deleted, and the rows and columns corresponding to the historical exited battery cluster are deleted. Therefore, the relation matrices on both sides are allowed to have different orders. During an S3 calculation, the current branch power, relation matrix, exit sample version, and PCS power adjustment step size are frozen. Newly generated industrial big data is saved for the next exit calculation. The specific processing is as follows: When entering the corresponding operation of the two sides structure, the branch power is first converted into non-negative transport mass; the current remaining battery cluster forms the current transport mass according to the ratio of the absolute value of the branch power to the sum of the absolute values ​​of the branch power of the current remaining battery cluster; the remaining battery cluster that exits the sample forms the sample transport mass according to the ratio of the absolute value of the branch power in the previous step to the corresponding sum. Subsequently, in S31, each current transport mass is multiplied by each sample transport mass, and the product is written into the row of the current battery cluster and the column of the sample battery cluster to form the initial transport matrix. The sum of the transport quantities in each row equals the corresponding current transport quality, and the sum of the transport quantities in each column equals the corresponding sample transport quality. For battery clusters with zero branch power, the corresponding transport mass is written to zero and the corresponding transport matrix row or column is fixed to zero; when the sum of the absolute values ​​of the branch power on either side is zero, the current exit calculation stops, or the corresponding exit sample does not participate in the subsequent structure correspondence. Transportation costs are recalculated as the transportation matrix changes to maintain the Gromov-Wasestein optimal transmission algorithm's comparison of the internal electrical relationships on both sides; Let a transportation location be determined by the current remaining battery cluster i and the sample remaining battery cluster j. When executing S32, iterate through the current remaining battery cluster k and the sample remaining battery cluster l, read the relationship value between i and k in the current relationship matrix and the relationship value between j and l in the sample relationship matrix, multiply the square of the difference between the two relationship values ​​by the transportation volume corresponding to k and l in the previous round transportation matrix, and sum over all combinations of k and l. The result is written as the transportation cost of location i and j. After the same operation is performed on all transportation locations, a transportation cost matrix is ​​formed. The initial transportation matrix formed by S31 is used as the transportation matrix of the previous round in the first round. In subsequent rounds, the transportation matrix output by the previous round is directly read and recalculated so that the difference in the relation matrix and the already formed inter-cluster correspondence jointly determine the transportation cost of the next round. Once the transportation cost is determined, the current round's transportation matrix is ​​calculated by cost scaling under row and column constraints. When implementing S33, the current transportation quality and sample transportation quality are first converted into integer power units according to their respective branch power recording units, and then multiplied by the common multiple of the sum of the power units on both sides, so that the total amount to be pushed by each node on the current side is equal to the total amount to be received by each node on the sample side. The transportation cost is uniformly fixed in decimal places according to the relation value field and shifted as an integer cost. Establish a starting side node for the current remaining battery cluster, establish an ending side node for the sample remaining battery cluster, establish a transport edge between the two sides and write the initial value of the node potential to zero, and at the same time establish a reverse remaining edge with opposite cost sign for each pushed transport edge. The cost metric is taken as the binary bit width corresponding to the absolute value of all integer costs. The reduced cost is obtained by adding the edge cost to the starting node's potential and then subtracting the ending node's potential. When there are remaining edges with negative cost, read the first remaining edge that meets the conditions in the order of the starting point node number and the ending point node number, and use the value of the first to be exhausted among the starting point push quantity, the ending point collect quantity and the remaining edge capacity as the current transportation quantity and deduct it at the same time. If a starting node that still needs to be pushed does not have any remaining edges with negative reduced costs, read the first value of the reduced costs of all remaining outgoing edges of that node in ascending order, subtract the sum of this value and the current cost scale from the node potential, and make the corresponding remaining edge re-enter the push range. After all the quantities to be pushed under the current cost scale are reduced to zero, the cost scale is divided by two and the existing transportation quantities are further corrected using the reverse residual edge until the cost scale is one, the quantities to be pushed and the quantities to be received on both sides are reduced to zero, and the reduced costs of the residual edge are all non-negative, thus obtaining the transportation matrix for this round. The current transport matrix is ​​then used to transfer the historical migration results to the current remaining battery clusters. For each current remaining battery cluster, the transport volume of the same row is read column by column in S34. The transport volume is multiplied by the migration ratio of the corresponding sample remaining battery cluster and summed. Then, it is divided by the current transport quality of the row to obtain the row migration value. When the current transport quality is zero, the row migration value is written as zero. After summing all the migration values, normalize them again to make the sum of the current migration ratios equal to one. Then, allocate the current power of the target battery cluster to each of the remaining battery clusters according to the current migration ratio. The current power of the target battery cluster is taken from the power command of the PCS branch that has been activated, and converted into an integer power step according to the PCS power adjustment step. Each allocation result first takes the complete power step, and the remaining power steps are filled in descending order according to the decimal part obtained by dividing the allocation result by the PCS power adjustment step. If the decimal parts are the same, they are processed according to the battery cluster number order until all power steps are allocated and the migration power is obtained. Gromov-Wasestein transport relations achieve stable selection through repeated back-substitution; after completing S34, the transport matrix of this round is returned to S32 to regenerate the transport cost matrix, and then S33 and S34 are executed, and the migration power vector is saved according to the round. When the newly generated migration power vector is consistent with any of the saved migration power vectors, the first occurrence of the same migration power vector up to the current round will be identified as a repeating segment and the iteration will stop, thus covering the loop state of more than two rounds. For each candidate transport matrix in the overlapping segment, S32 is re-executed according to the candidate transport matrix itself to obtain the corresponding transport cost. Then, the transport cost is multiplied by the transport volume at the same location item by item and summed to obtain the relationship distortion value. S35 reads the migration power of the first ascending value of the relation distortion value in the repeated segment, and repeats the above operation for all valid exit samples. Then, it reads the result of the first ascending value of the relation distortion value. When the relation distortion values ​​are the same, the result is determined according to the exit sample number and round order to obtain the migration power for S4 to read. When the current remaining battery cluster branch power ratios are 0.40, 0.35, and 0.25, and the corresponding ratios of historical exit samples are 0.30, 0.45, and 0.25, S31 forms a three-row, three-column initial transportation matrix. For example, the three transportation quantities in the first row are 0.12, 0.18, and 0.10, respectively. Subsequently, the transportation cost recalculation and transportation matrix correction are repeatedly performed according to the relationship matrix on both sides. If the final current migration ratio is 0.40, 0.35 and 0.25, the current power of the target battery cluster is 60kW and the PCS power adjustment step is 5kW, then the three initial allocation results correspond to 4.8, 4.2 and 3 power steps respectively. First, take 4, 4 and 3 complete power steps, and fill the remaining power step into the first battery cluster with a decimal part of 0.8. The final migration power is 25kW, 20kW and 15kW. Thus, the inter-cluster electrical structure and migration ratio in the historical exit samples are associated with the current parallel structure through the same transport matrix, and the migration power is formed by discrete PCS power steps that can be directly read for subsequent power allocation.

[0020] S4. Using the current power of the target battery cluster as the power to be allocated, divide the allocable capacity between the current power and the corresponding BMS power boundary of each remaining battery cluster into power steps according to the PCS power adjustment step. For each power step of each remaining battery cluster, subtract the square of the difference between the cumulative allocated power after adding the corresponding power step and the corresponding migration power from the square of the difference between the cumulative allocated power before adding the corresponding power step to obtain the edge cost corresponding to the power step. Push the flow along the negatively reduced cost remaining edge and shift the cost scale to the right step by step through the cost scaling push flow algorithm until the power to be allocated is zero to obtain the allocation increment of each remaining battery cluster. The power after the target battery cluster exits needs to be redistributed within the allowable range of the remaining battery clusters. At the same time, the redistribution result must correspond to the migration power obtained in S3, so that the power reserve formed before exit can offset the natural power migration after the contactor is disconnected. When entering S4, the current branch power of each remaining battery cluster, BMS power boundary, PCS power adjustment step, and other DC power compensation power are frozen; the current power of the target battery cluster is used as the power to be allocated, of which the other DC power compensation power is used as a fixed amount, and the remaining part is taken over by the remaining battery cluster. If frozen data changes during calculation, the current allocation result is discarded, and the calculation is recalculated based on the updated control frame; the specific process is as follows: The power allocation range is first determined by the branch power that each remaining battery cluster can continue to bear; the current branch power and BMS power boundary are read along the current power direction of the target battery cluster. When the current branch power has not yet reached the BMS power boundary, the power difference between the two along the same direction is taken as the allocable capacity; when it has reached or crossed the BMS power boundary, the allocable capacity is written to zero. Subsequently, in S41, the allocable capacity is divided step by step from zero according to the PCS power adjustment step. Each complete step forms a power step, and the remaining power at the end forms a boundary power step separately. The corresponding PCS command is directly locked to the BMS power boundary, thus forming power edges arranged in ascending order. The BMS power boundary remains the same version during this calculation. If the boundary changes, the current power boundary is stopped and regenerated. The cost borne by each power side is directly determined by the S3 migration power, so that the manual allocation amount is prioritized to the power expected to migrate naturally after the contactor is disconnected; For a remaining battery cluster, read the power steps sequentially, calculate the square of the difference between the cumulative allocated power after adding the current power step and the corresponding migration power, then subtract the square of the difference before adding, and write the resulting increment into the current power edge as the edge cost, thus completing S42; the edge capacity is taken as the power amount of the current power step itself; Since the cumulative power allocated to the same battery cluster increases with the power step order, the edge cost formed by the squared difference also increases with the order. Therefore, the preceding power edge enters the push flow in terms of cost before the following power edge. The power edges of each remaining battery cluster are connected to the power source node of the target battery cluster, and the compensation power of other DC power sources is written as a fixed flow branch, so that the current power of the target battery cluster entering the network is simultaneously borne by the fixed compensation branch and the power edges of the remaining battery clusters. The power destination is corrected step by step according to the discrete cost in the push flow operation; before executing S43, all edge costs are shifted to integers according to the decimal places of the PCS power record field, the binary bit width of the first value of the absolute value of the edge cost in descending order is read to determine the first round cost scale, and the initial value of the potential of each node is written to zero. For each forward power edge, a reverse residual edge is simultaneously established. The capacity of the reverse residual edge is taken as the power that has already been pushed, and the edge cost is the opposite of the forward edge cost. When calculating each remaining edge, the reduced cost is obtained by adding the edge cost to the starting node potential and then subtracting the ending node potential. The remaining edges with negative reduced costs are read according to the remaining battery cluster number and power step order. The power that has not yet been allocated at the starting point and the value of the first depleted value in the remaining edge capacity are used as the current push flow. When the current node still has power to be allocated and there are no negative reduced cost outgoing edges, read the first value of the reduced cost in ascending order of the outgoing edge, subtract the sum of this value and the current cost scale from the current node's potential, and then recalculate the reduced cost. If there are no remaining edges that can further reduce costs under the current cost scale, shift the cost scale one position to the right and readjust the existing push flow using the forward and reverse remaining edges; After the cost scale is reduced to one, the process continues until the power to be allocated to the target battery cluster is fully taken up by the fixed compensation branch and the remaining battery cluster power edge, and the reduced cost of all remaining edges is non-negative. If the sum of the allocable capacity of all remaining battery clusters and the compensation power of other DC power supplies is less than the current power of the target battery cluster, the planned exit will not be executed, and the insufficient capacity result will be returned to the current control unit to redetermine the exit arrangement. The push flow result ultimately needs to be restored to the branch power increment actually used by each remaining battery cluster; after completing the cost scaling, in S44, the push flow of all power edges of the same battery cluster is summarized by battery cluster number to obtain the corresponding allocation increment, and the BMS power boundary is checked with the current branch power plus the allocation increment; if any result exceeds the frozen BMS power boundary, the current push flow result is discarded and the corresponding power edge is regenerated. The fixed compensation branch and all battery cluster power edges together satisfy the source node power conservation. Therefore, the sum of each allocation increment plus the compensation power of other DC power sources equals the current power of the target battery cluster. The resulting allocation increment, together with the S3 migration power, is written into the current exit calculation record for S5 to calculate the preset power. Taking the planned withdrawal of a 60kW target battery cluster as an example, other DC power supplies undertake 10kW of fixed compensation power in the same withdrawal arrangement. The allocable capacities of the three remaining battery clusters are 30kW, 25kW and 20kW respectively. The PCS power adjustment step is 5kW. The migration power obtained by S3 is 28kW, 17kW and 15kW respectively. After the power edge is expanded in 5kW increments, the square difference increment makes each battery cluster have a prior cost when it is close to the power step of the corresponding migration power. The cost scaling push flow gets 25kW, 15kW and 10kW allocation increments when it takes on the remaining 50kW. The three of them, together with the 10kW compensation power of other DC power sources, total 60kW, and none of them exceed the corresponding BMS power boundary. The resulting allocation increment simultaneously satisfies the correspondence between the branch carrying capacity, the target battery cluster power transfer, and the S3 migration power, and can be directly used for the pre-set power calculation before the next stage exit.

[0021] S5. The preset power is obtained by adding the allocation increment to the current branch power of each remaining battery cluster and then subtracting the migration power. The preset power is executed while maintaining the current branch power of the target battery cluster, so that the algebraic sum of each preset power, the current branch power of the target battery cluster, and the current power of other DC power sources in the same power direction is equal to the grid load balancing power command in the same step. In the next control step after the preset power is executed, the target battery cluster is disconnected, so that the remaining battery clusters are transferred to the sum of the current branch power and the allocation increment, and the migration ratio is updated according to the difference in branch current before and after disconnection and then written back to the industrial big data. When entering the target battery cluster exit execution phase, the migration power obtained in S3 is used to determine the power expected to be taken by each remaining battery cluster at the moment of exit, and the allocation increment obtained in S4 is used to determine the amount of stable load that each remaining battery cluster needs to increase after exit. The branch power of each remaining battery cluster at the start of S4 is fixed and recorded as the reference branch power. The power is calculated based on the same reference branch power after subsequent presets and exits. Each battery cluster is connected to the DC bus through an independent bidirectional power adjustment channel, so that the branch power can be sent and adjusted separately. When other DC power supplies participate in S4 compensation, their compensation power is executed synchronously during the preset phase and maintains the corresponding power until the exit is complete; the specific execution is as follows: In the pre-setting stage, the power position of each remaining battery cluster before exit is first determined based on the power of the same reference branch; for each remaining battery cluster, the initial pre-setting power is obtained by adding the allocation increment to the reference branch power and then subtracting the migration power. Subsequently, in S51, the initial preset power is converted into a complete power step according to the PCS power adjustment step, and the power margin generated by rounding is saved. The remaining battery clusters that are still within the BMS power boundary are added to the power step according to the descending order of the absolute value of the margin. When the margins are the same, they are processed according to the battery cluster number order. When the end corresponds to the boundary power step reserved in S41, the boundary power step is directly used. When other DC power sources have compensation power, first confirm that the actual power has increased by the corresponding compensation amount relative to the start of S4, and then check that the sum of all preset power, the current branch power of the target battery cluster and the current power of other DC power sources is consistent with the grid load balancing power command. When the compensation power is zero, the total preset power is directly compared with the current branch power of the target battery cluster; if the rounding margin cannot be eliminated within the BMS power boundary, the current preset result is discarded and returned to S4 for reallocation. After each branch receives the preset power, the power execution status is confirmed sequentially according to the control step; during this period, the target battery cluster maintains the current branch power when entering S5, and each remaining battery cluster executes the preset power respectively; S52 reads the actual branch power step by step and discretizes it according to the same PCS power adjustment step and boundary power step rules as S51. When the discretization result of all remaining battery clusters is consistent with the preset power for the first time, the corresponding control step is determined as the exit execution step. The next control step after exiting the execution step is the control step for the target battery cluster to perform the disconnection action; in this control step, the power command corresponding to the exit of each remaining battery cluster is first latched, then the target battery cluster contactor is disconnected and the latched power command is released, so that the station-level power before and after the exit of the target battery cluster is connected according to the same grid load balancing power command. The number of waiting steps is obtained by dividing the PCS rated power command response time by the control cycle and rounding up; if no consistent result is formed after the last waiting step is reached, the plan is canceled and exited. If the power load balancing command, online battery cluster set, BMS power boundary or other DC power compensation power changes during the waiting period, the current results will also be discarded and the relevant calculations will be re-executed from the updated control step. After the exit execution phase is reached, the power command and contactor action are executed in a fixed order within the same control phase; first, the power of each remaining battery cluster after exit is latched, and the latched value is uniformly taken as the sum of the corresponding reference branch power and the allocation increment; The target battery cluster power command is then reduced to the zero power step. After the branch power readback enters the zero power step, the target battery cluster contactor is disconnected, and the remaining battery cluster power command that has been latched is released, thus completing S53. Before the target battery cluster is removed, the preset power and other DC power compensations maintain the station-level power; after the target battery cluster is removed, the remaining battery clusters bear the power according to the sum of the base branch power and the allocated increment, so that the station-level power continues to correspond to the same grid load balancing power command. If the BMS issues a protection trip command during the planned exit process, the protection action will be executed before the above exit sequence. This event will be recorded as a protection exit and will not be used as the result of the planned exit control. After the contactor is disconnected, the migration results in the industrial big data need to be corrected by using the actual branch changes. S54 reads the last complete closed sampling point before the target battery cluster is disconnected and the first complete sampling point after disconnection. It calculates the actual current of each remaining battery cluster along the migration direction determined by S23, writes it to zero in the reverse direction, and then divides each actual current by the sum of the actual currents to obtain the updated migration ratio. When the actual total current received is zero, the first complete sampling period is missing, the power load balancing command changes between two sampling periods, or the online battery cluster set changes again, no update migration ratio is generated; Valid results are written to the new industrial big data version according to the exit event number, and the exit execution time, branch sampling records before and after exit, and updated migration ratio are saved. The data versions used in S3 to S5 remain unchanged, and the new version is read from the next target battery cluster exit calculation. For example, the base branch power of the three remaining battery clusters is 40kW, 35kW and 25kW respectively. S3 obtains the migration power of 25kW, 20kW and 15kW, S4 obtains the allocation increment of 20kW, 15kW and 15kW, and the compensation power of other DC power sources is 10kW. Then the three initial preset power are 35kW, 30kW and 25kW respectively. After other DC power sources are added by 10kW, the target battery cluster remains at 60kW, and the total station power maintains the original grid load balance power. After the three remaining battery clusters reach the preset power, they enter the exit execution phase. The target battery cluster exits, and the three remaining battery clusters switch to 60kW, 50kW, and 40kW respectively. After disconnection, the actual current received by the first complete sampling phase is normalized according to the same migration direction and written back to the industrial big data, so that the actual exit result of this exit becomes a new sample for subsequent exit calculations.

[0022] exist Figure 1 It should be noted that: This diagram represents the overall control flow when the target battery cluster exits; the solid arrows indicate the normal execution sequence of the scheme, that is, firstly, exit samples are formed from industrial big data, then the relationship matrix and migration ratio of battery clusters are calculated, then the migration power is obtained according to the current operating status of the battery cluster, and the allocation increment of each remaining battery cluster is calculated in combination with the BMS power boundary and PCS power adjustment step. Finally, the power of the remaining battery clusters is adjusted before the target battery cluster exits, and the target battery cluster exits after the adjustment is completed; the dashed arrows indicate the data write-back process, that is, after the target battery cluster exits, the migration ratio is obtained again according to the actual current received by the remaining battery clusters, and the exit result is written back to the industrial big data for subsequent reading and calculation when other battery clusters exit.

[0023] Overall Implementation and Explanation: This solution first extracts the branch changes before and after the battery clusters are withdrawn from the industrial big data generated by the long-term operation of the lithium iron phosphate energy storage system. After deducting the power compensation of other DC power sources, the migration relationship of the actual power received by each remaining battery cluster is obtained. Then, the equivalent voltage of the cluster is calculated based on the terminal voltage, current and branch impedance of each battery cluster before withdrawal, and a relationship matrix is ​​constructed accordingly. When a target battery cluster needs to be decommissioned, the Gromov-Wasestein optimal transfer algorithm is used to compare the inter-cluster electrical relationship between the current parallel structure and the historical decommissioning samples, and the historical migration ratio is mapped to the current remaining battery cluster to obtain the migration power. Then, the cost scaling push flow algorithm is used to allocate the power after the target battery cluster decommissions under the BMS power boundary and PCS power adjustment step limit of each remaining battery cluster to obtain the allocation increment. The pre-set power before exit is formed by adding the current branch power to the allocation increment and then subtracting the migration power. This allows the remaining battery clusters to free up the power space required for the expected migration in advance. After the target battery cluster exits, the allocated branch power is then transferred, and the actual acceptance results are rewritten into the industrial big data for continued use in subsequent exit control. For example, when an energy storage power station is performing grid discharge, one of the battery clusters needs to be decommissioned due to operational limitations. The system first finds the decommissioning process corresponding to the current electrical relationship of the multiple clusters from the historical decommissioning records, and calculates that after the target battery cluster is disconnected, the three remaining battery clusters are expected to naturally take over 25 kW, 20 kW, and 15 kW, respectively. Then, based on the current remaining power capacity of each battery cluster, it calculates the additional stable load of 20 kW, 15 kW, and 15 kW that need to be added after the target battery cluster is decommissioned. Before exiting, the system adjusts the three remaining battery clusters according to the current branch power plus the stable load and then the expected migration power, so that the branch that can easily take on more power at the moment of disconnection can make room in advance; after the preset power is executed, the target battery cluster is disconnected and the remaining battery clusters are put into a new power allocation state. In this way, the natural power migration that occurs when the target battery cluster exits can fall into the pre-reserved branch space, and the energy storage device continues to operate at the original grid load balance power. At the same time, the actual migration results continue to be added to the industrial big data.

[0024] 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. A multi-cluster parallel coordinated control method for a lithium iron phosphate energy storage system, characterized in that, include: S1. Obtain the industrial big data written by the energy storage control device according to the control step. The industrial big data includes the terminal voltage, branch current, branch impedance, branch power, contactor status, grid load balancing power command and other DC power of each battery cluster. Take the control step when the contactor of the target battery cluster changes from the closed state to the open state as the exit step. Extract the data from the step before exit to the time before the grid load balancing power command changes to form the exit sample. S2. For the exited sample, the equivalent voltage of each battery cluster is obtained by subtracting the product of the branch current and the branch impedance from the terminal voltage of the previous step before exiting. The equivalent voltages of any two battery clusters are subtracted and their absolute values ​​are taken to form a relation matrix. The branch current that increases along the current direction of the target battery cluster in the first step after exiting is taken as the receiving current. The migration ratio is obtained by dividing each receiving current by the sum of all receiving currents. S3. After determining the target battery cluster, generate the current relationship matrix according to S2. Use the current remaining battery cluster branch power ratio and the exited sample remaining battery cluster branch power ratio as the current transport quality and sample transport quality, respectively. Multiply each current transport quality by each sample transport quality to form an initial transport matrix and use it as the front wheel transport matrix. For each transport location determined by the current remaining battery cluster and the sample remaining battery cluster, the square of the difference between the relationship value between the current remaining battery cluster and another current remaining battery cluster in the current relationship matrix and the relationship value between the sample remaining battery cluster and another sample remaining battery cluster in the sample relationship matrix is ​​multiplied by the transport volume of the location determined by another current remaining battery cluster and another sample remaining battery cluster in the previous transport matrix. The transport cost is then summed over all other current remaining battery clusters and other sample remaining battery clusters. The transport matrix is ​​updated using a cost scaling push algorithm, and the current migration ratio is obtained by weighting the sample migration ratio with the transport matrix and normalizing it according to the current transport quality. The target battery cluster's current power is allocated according to the current migration ratio, and the step size is adjusted and rounded according to the PCS power to obtain the migration power. When the migration power vector of this round is consistent with the migration power vector of the previous round saved in each cluster, the round between the two rounds is determined as the repeated segment. The relationship distortion value is obtained by multiplying the transportation cost by the transportation volume at the same location and summing the results. The migration power corresponding to the first position of the relationship distortion value in the repeated segment and each exited sample in ascending order is read. S4. Using the current power of the target battery cluster as the power to be allocated, divide the allocable capacity between the current power and the corresponding BMS power boundary of each remaining battery cluster into power steps according to the PCS power adjustment step. For each power step of each remaining battery cluster, subtract the square of the difference between the cumulative allocated power after adding the corresponding power step and the corresponding migration power from the square of the difference between the cumulative allocated power before adding the corresponding power step to obtain the edge cost corresponding to the power step. Push the flow along the negatively reduced cost remaining edge and shift the cost scale to the right step by step through the cost scaling push flow algorithm until the power to be allocated is zero to obtain the allocation increment of each remaining battery cluster. S5. The preset power is obtained by adding the allocation increment to the current branch power of each remaining battery cluster and then subtracting the migration power. The preset power is executed while maintaining the current branch power of the target battery cluster, so that the algebraic sum of each preset power, the current branch power of the target battery cluster, and the current power of other DC power sources in the same power direction is equal to the grid load balancing power command in the same step. In the next control step after the preset power is executed, the target battery cluster is disconnected, so that the remaining battery clusters are transferred to the sum of the current branch power and the allocation increment, and the migration ratio is updated according to the difference in branch current before and after disconnection and then written back to the industrial big data. Among them, the control phase is a control cycle in which the energy storage control device completes branch data acquisition, power calculation and power command issuance in sequence according to a fixed control cycle; the first phase after exit is the first complete sampling control phase after the target battery cluster contactor is disconnected.

2. The multi-cluster parallel coordinated control method for a lithium iron phosphate energy storage system according to claim 1, characterized in that: S1 includes: S11. The exit power is obtained by multiplying the bus voltage of the previous phase by the branch current of the target battery cluster, and the grid load balancing power command of the same phase is written into the exit sample as the power reference. The power reference indicates the total power that the energy storage system should maintain in a uniform power direction before the target battery cluster exits. S12. For each control phase after exiting the phase, the DC compensation power is obtained by subtracting the power of the previous phase from the current power of other DC power sources, and then the net exit power that the remaining battery cluster needs to take over is obtained by subtracting the DC compensation power from the exit power. S13. For each remaining battery cluster, subtract the branch power corresponding to the previous step from the current branch power to obtain the branch increment; write zero when the branch increment is less than zero, sum all the positive values ​​after writing to obtain the total positive value, divide each positive value by the total positive value and multiply by the net exit power to obtain the migration reference amount of the corresponding remaining battery cluster, and write the migration reference amount into the exit sample with the corresponding control step until the grid load balance power command changes.

3. The multi-cluster parallel coordinated control method for a lithium iron phosphate energy storage system according to claim 2, characterized in that: S2 includes: S21. For each exited sample, read the terminal voltage, current and branch impedance of the previous step according to the battery cluster number, subtract the product of current and branch impedance from the terminal voltage to obtain the cluster equivalent voltage, and form an equivalent voltage sequence according to the battery cluster number. S22. Based on the equivalent voltage sequence, calculate the absolute value of the cluster equivalent voltage difference between any two battery clusters, and write the corresponding values ​​according to the row and column positions determined by the two battery cluster numbers, and write zeros according to the positions of the same cluster, to obtain a relation matrix symmetric about the main diagonal.

4. The multi-cluster parallel coordinated control method for a lithium iron phosphate energy storage system according to claim 3, characterized in that: S2 also includes: S23. In the first step after exiting, take the current direction of the target battery cluster in the previous step as the migration direction, record the current increase of each remaining battery cluster relative to the previous step along the migration direction as the receiving current, and write the reverse change part as zero to obtain the receiving current of each remaining battery cluster. S24. The total capacity is obtained by summing the current of each remaining battery cluster. The migration ratio is obtained by dividing the current of each remaining battery cluster by the total capacity. The migration ratio is then bound to the corresponding exit sample by the relationship matrix.

5. The multi-cluster parallel coordinated control method for a lithium iron phosphate energy storage system according to claim 4, characterized in that: S3 includes: S31. Determine the target battery cluster and generate the current relationship matrix according to S2. Use the current remaining battery cluster branch power ratio to form the current transport quality, use the remaining battery cluster branch power ratio of the exited sample to form the sample transport quality, and multiply the two transport qualities item by item to form the initial transport matrix. S32. For each transportation location, the transportation location is determined by the row position corresponding to a current remaining battery cluster and the column position corresponding to a sample remaining battery cluster. Traverse the relationship values ​​between the current remaining battery cluster and another current remaining battery cluster corresponding to the row position in the current relationship matrix, and the relationship values ​​between the sample remaining battery cluster and another sample remaining battery cluster corresponding to the column position in the sample relationship matrix. Subtract the two relationship values, square them, multiply them by the transportation volume determined by the other current remaining battery cluster and another sample remaining battery cluster in the previous round transportation matrix, and accumulate all other current remaining battery clusters and other sample remaining battery clusters to obtain the transportation cost of the corresponding transportation location. After completing the calculation of all transportation locations, a transportation cost matrix is ​​formed.

6. The multi-cluster parallel coordinated control method for a lithium iron phosphate energy storage system according to claim 5, characterized in that: S3 also includes: S33. Limit the row sum and column sum of the transportation matrix with the current transportation quality and the sample transportation quality respectively. Distribute the transportation volume along the negative cost remaining edge through the cost scaling push flow algorithm. When there is no corresponding remaining edge, adjust the node potential according to the cost reduction of the remaining edge and reduce the cost scale step by step until the row sum and column sum satisfy the corresponding transportation quality, and obtain the transportation matrix of this round. S34. The current migration ratio is obtained by weighting the sample migration ratio with the current transportation matrix and normalizing it according to the current transportation quality. Then, the current power of the target battery cluster is allocated according to the current migration ratio, and the PCS power adjustment step is rounded and the rounding margin is supplemented to obtain the migration power. S35. Return the current round of transportation matrix to S32 to recalculate the transportation cost and execute S33 and S34 in sequence until the migration power is the same as the migration power of the previous round. Use the sum of the product of transportation cost and corresponding transportation volume as the relationship distortion value, and read the migration power of the first-ranked relationship distortion value in the repeated segment and each exit sample.

7. The multi-cluster parallel coordinated control method for a lithium iron phosphate energy storage system according to claim 6, characterized in that: S4 includes: S41. Read the allocable capacity of each remaining battery cluster from the current power to the BMS power boundary in the current power direction of the target battery cluster. Divide the power steps step by step according to the power adjustment step of each power command change of the PCS, and retain the last segment that is less than one power adjustment step separately to form the power edge of each remaining battery cluster. S42. For each power edge, the edge cost is obtained by subtracting the square of the difference between the cumulative allocated power after adding the corresponding power step and the migration power of the corresponding remaining battery cluster in S3 from the square of the difference between the cumulative allocated power before adding the corresponding power step and the corresponding migration power. The migration power is the power amount after the target battery cluster's current power is allocated according to the current migration ratio obtained in S3 and rounded according to the PCS power adjustment step. The corresponding power step is used as the edge capacity to form a push network from the target battery cluster's current power to each remaining battery cluster.

8. The multi-cluster parallel coordinated control method for a lithium iron phosphate energy storage system according to claim 7, characterized in that: S4 also includes: S43. Determine the cost scale by reading the binary bit width of the first value of the absolute value of the edge cost in descending order through the cost scaling push algorithm. Obtain the reduced cost by adding the edge cost to the starting node potential and subtracting the ending node potential. Push the value that is exhausted first between the node remaining capacity and the edge remaining capacity along the remaining edge with negative reduced cost. When there is no remaining edge with negative reduced cost, read the first value of the edge reduced cost in ascending order and combine it with the cost scale to correct the node potential. After each round, shift the cost scale one bit to the right. After the cost scale drops to one, continue the calculation until the power to be allocated is zero and the reduced cost of all remaining edges is non-negative. S44. The allocation increment is obtained according to the push flow received by each remaining battery cluster, so that the sum of the current power of each remaining battery cluster and the allocation increment does not exceed the corresponding BMS power boundary, and the sum of the allocation increment and the compensation power of other DC power sources is equal to the current power of the target battery cluster.

9. The multi-cluster parallel coordinated control method for a lithium iron phosphate energy storage system according to claim 8, characterized in that: S5 includes: S51. For each remaining battery cluster, the preset power is obtained by adding the current branch power to the allocation increment and then subtracting the migration power. The power margin generated by rounding each preset power according to the PCS power adjustment step is sequentially added to the remaining battery clusters that have not reached the BMS power boundary, so that each preset power, the current branch power of the target battery cluster and the current power of other DC power sources are kept as the grid load balance power command according to the algebraic sum of the unified power direction. S52. When the target battery cluster maintains the current branch power, a preset power is issued to each remaining battery cluster. The actual branch power is read step by step, and the first step in which the actual branch power of each remaining battery cluster is rounded according to the PCS power adjustment step and is consistent with the preset power is determined as the exit execution step.

10. The multi-cluster parallel coordinated control method for a lithium iron phosphate energy storage system according to claim 9, characterized in that: S5 also includes: S53. In the next control step after exiting the execution step, disconnect the target battery cluster and simultaneously change the power command of each remaining battery cluster to the sum of the current branch power and the allocation increment, so that the algebraic sum of each preset power before disconnection, the current branch power of the target battery cluster and the current power of other DC power sources in the same power direction, and the algebraic sum of the current branch power and the allocation increment of each remaining battery cluster and the current power of other DC power sources in the same power direction after disconnection are all equal to the same grid load balancing power command. S54. Obtain the actual current received by the difference between the remaining battery cluster branch currents before and after the exit execution step. Sum the actual current received along the current direction before the target battery cluster exits to obtain the updated migration ratio. Write the updated migration ratio along with the exit execution step into the industrial big data.

Images