Deep learning-based method for determining compensation volume of air bag of marine power battery pack

Abstract

The application discloses a marine power battery pack air bag compensation volume discrimination method based on deep learning, relates to the technical field of marine power battery pack air bag capacity discrimination, and aims to solve the problem that it is difficult to combine the space connection relationship to discriminate the cold shrinkage negative pressure compensation volume in the temperature cycle breathing process at sea. According to the pressure time sequence, the module temperature time sequence and the compressible gas space distribution, the method determines a cold shrinkage negative pressure compensation volume reference value, divides gas sub-chambers, and constructs a gas sub-chamber connection hypergraph. The temperature change, the pressure change, the sub-chamber volume and the connection relationship are written into nodes and hyperedges, the local compensation volume contribution value is generated by inputting the hypergraph neural network, the one-sided capacity shortage loss is embedded in the readout training link, and the air bag compensation volume discrimination value is obtained. Through comparison with the effective compensation volume of the air bag to be configured, the capacity satisfaction or deficiency result is output, and the method can be used for air bag capacity configuration discrimination of a marine power battery pack.

Description

Technical Field

[0001] This invention relates to the field of marine power battery pack airbag capacity discrimination technology, and in particular to a deep learning-based method for discriminating the compensation volume of marine power battery pack airbags. Background Technology

[0002] When marine power battery packs operate in a marine environment, the external temperature fluctuates periodically depending on the day and night, the navigation area, and the internal thermal conditions. The battery pack housing typically requires sealing, and the space inside the housing not occupied by battery modules, structural components, connectors, and seals forms a compressible gas space. During the marine temperature cycle, the gas inside the housing contracts as the temperature drops, causing the internal pressure of the battery pack housing to decrease with each data collection point, creating a need for cold contraction negative pressure compensation. To counteract this cold contraction negative pressure, airbags can be configured in engineering. These airbags utilize the volume change generated within their effective travel range to provide compensating volume to the inside of the battery pack housing. Therefore, it is necessary to determine whether the effective compensating volume of the airbag to be configured covers the cold contraction negative pressure compensation requirement.

[0003] Existing technical solutions typically collect internal pressure and module temperature values ​​of the battery pack housing through pressure and temperature measuring points. These values ​​are then used to create pressure and temperature time sequences based on the collection time, establishing a correspondence between the temperature decrease and pressure reduction processes. The capacity estimation stage generally combines the total volume of compressible gas inside the housing, temperature changes, pressure changes, and preset volume conversion records to obtain a reference value for the cold contraction negative pressure compensation volume. The structural processing stage identifies the spatial location, boundaries, and volume distribution of compressible gas based on the inner wall of the housing, the outer contours of the battery modules, structural components, connectors, and seals, and records gas connectivity based on the locations of openings, gaps, and vents. The judgment stage typically uses the volume change of the airbag to be configured between preset compression and expansion limits as the effective compensation volume, comparing it with the calculated compensation volume requirement to arrive at an airbag capacity configuration conclusion. In engineering implementation, local spatial volumes are often obtained from geometric models, volume calibration data, or structural records. Sensor installation locations are used to determine the source of pressure and temperature data, providing a traceable data basis for capacity judgment.

[0004] The existing solutions described above mostly treat the compressible gas space inside the battery pack housing as a whole, making it difficult to express the ventilation openings, isolation positions, and pressure transmission ranges between multiple gas sub-cavities after structural separation. The attribution relationships between temperature and pressure measuring points and gas sub-cavities are also difficult to incorporate into the capacity determination process. When capacity requirements are formed by multiple local spaces, existing solutions struggle to distinguish the local contributions of different connected gas sub-cavities to the compensation volume. When reading the compensation volume, compensation requirements below and above the cold contraction negative pressure are often treated the same, lacking stronger constraints in the direction of capacity shortage. This results in the airbag compensation volume determination value failing to reflect the joint participation of locally connected gas regions in cold contraction negative pressure changes during marine temperature cycling.

[0005] Therefore, a method for determining the airbag compensation volume of marine power battery packs that can overcome the shortcomings of the existing technology is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a deep learning-based method for determining the airbag compensation volume of marine power battery packs. The core technical problem to be solved by this application is: during the marine power battery pack's temperature-cycle breathing process at sea, how to incorporate the division of gas sub-cavities, sub-cavity connectivity, temperature changes, pressure changes, and the effective compensation volume of the airbag to be configured into the same determination process, forming an airbag compensation volume determination value that can be used to output the determination result of whether the airbag capacity meets the determination result or the determination result of whether the airbag capacity is insufficient.

[0007] The deep learning-based method for determining the airbag compensation volume of a marine power battery pack according to embodiments of the present invention includes:

[0008] S1. Obtain marine power battery pack operation data, including the internal pressure sequence of the battery pack housing during the marine temperature cycle breathing process, the module temperature sequence, the spatial distribution of compressible gas inside the battery pack housing, and the effective compensation volume of the airbag to be configured.

[0009] S2. Determine the reference value of the cold shrinkage negative pressure compensation volume based on the internal pressure sequence and module temperature sequence of the battery pack housing, and form the gas sub-cavity division result based on the spatial distribution of compressible gas inside the battery pack housing.

[0010] S3. Construct a gas sub-cavity connected hypergraph based on the gas sub-cavity partitioning results. The nodes of the gas sub-cavity connected hypergraph include gas sub-cavities, temperature measuring points, and pressure measuring points. The hyperedges of the gas sub-cavity connected hypergraph are formed by the nodes corresponding to the connected gas sub-cavities.

[0011] S4. Write the temperature change, pressure change, and sub-cavity volume into the nodes of the gas sub-cavity connected hypergraph, and write the sub-cavity connectivity relationship into the hyperedge of the gas sub-cavity connected hypergraph to obtain the sub-cavity connected hypergraph features.

[0012] S5. Input the sub-cavity connected hypergraph features into the hypergraph neural network to generate local compensation volume contribution values. In the readout training stage where the local compensation volume contribution values ​​converge into the airbag compensation volume output value, embed the unilateral capacity shortage loss. The difference between the airbag compensation volume output value and the reference value of the cold shrinkage negative pressure compensation volume is determined as the capacity shortage amount. The difference between the airbag compensation volume output value and the reference value of the cold shrinkage negative pressure compensation volume is determined as the capacity surplus amount. According to the constraint that the loss corresponding to the capacity shortage amount is greater than the loss corresponding to the capacity surplus amount, the airbag compensation volume discrimination value is obtained.

[0013] S6. Compare the effective compensation volume of the airbag to be configured with the airbag compensation volume discrimination value. When the effective compensation volume of the airbag to be configured is not less than the airbag compensation volume discrimination value, the output airbag capacity meets the discrimination result. When the effective compensation volume of the airbag to be configured is less than the airbag compensation volume discrimination value, the output airbag capacity is insufficient discrimination result.

[0014] Optionally, S1 includes:

[0015] When the marine power battery pack is in the process of temperature cycling at sea, the pressure values ​​of the pressure measuring points inside the battery pack are collected and arranged according to the time of collection to form the time sequence of the internal pressure of the battery pack.

[0016] During the same marine temperature cycle breathing process that forms the internal pressure sequence of the battery pack housing, the temperature values ​​of the module temperature measurement points are collected and arranged according to the collection time to form the module temperature sequence.

[0017] Based on the structural boundaries, gas communication locations, and sub-cavity volume sources of the compressible gas space inside the battery pack housing, the distribution of the compressible gas space inside the battery pack housing is obtained.

[0018] The effective compensation volume of the airbag to be configured is obtained based on the volume change of the airbag within its effective stroke. The internal pressure timing of the battery pack housing, the module temperature timing, the spatial distribution of compressible gas inside the battery pack housing, and the effective compensation volume of the airbag to be configured are combined to form the operating data of the marine power battery pack.

[0019] Terminology Explanation:

[0020] The marine temperature cycle breathing process refers to the process by which the gas inside the battery pack expands and contracts with temperature changes under the periodic rise and fall of the marine ambient temperature, and the internal pressure of the battery pack changes with the time of data collection.

[0021] The compressible gas space distribution inside the battery pack housing refers to the location, boundary, and volume distribution of the space inside the battery pack housing that is not occupied by battery modules, structural components, connectors, and seals and can accommodate gas compression or expansion.

[0022] The source of sub-cavity volume is determined by the structural boundaries of the compressible gas space inside the battery pack housing, the gas communication location, and the local space formed by the separation of structural components to determine the corresponding volume of each gas sub-cavity.

[0023] Effective stroke is the stroke within which the airbag to be configured can undergo volume change within the working range not exceeding the preset compression limit and preset expansion limit;

[0024] The effective compensation volume of the airbag to be configured is the amount of volume change that the airbag to be configured can provide to the inside of the battery pack housing to offset the negative pressure caused by cold contraction within its effective stroke.

[0025] Optionally, S2 includes:

[0026] The temperature drop process is extracted from the module temperature time series, and the internal pressure time series of the battery pack is matched with the temperature drop process to form a cold contraction negative pressure correspondence.

[0027] Based on the correspondence between cold shrinkage and negative pressure, the volume compensation requirement corresponding to the pressure change inside the battery pack housing and the temperature change of the module is calculated, and the volume compensation requirement is determined as the reference value for cold shrinkage and negative pressure compensation volume.

[0028] Based on the spatial distribution of compressible gas inside the battery pack housing, structural boundaries and gas connection locations are extracted to form sub-cavity boundary data;

[0029] The gas sub-cavities are divided based on the sub-cavity boundary data, and the sub-cavity volume of each gas sub-cavity is determined to form the gas sub-cavity division result.

[0030] Terminology Explanation:

[0031] The cold shrinkage negative pressure correspondence is the correspondence established between the temperature drop process in the module temperature time series and the pressure drop process in the battery pack internal pressure time series according to the acquisition time.

[0032] The volume compensation requirement corresponding to the change in internal pressure of the battery pack housing and the change in module temperature is the gas volume requirement that needs to be compensated by the airbag to be configured, which is determined based on the gas contraction trend caused by the change in module temperature and the change in internal pressure of the battery pack housing during the temperature drop process.

[0033] The reference value for the cold shrinkage negative pressure compensation volume is a reference volume determined based on the cold shrinkage negative pressure correspondence, used to represent the cold shrinkage negative pressure that needs to be compensated during the temperature cycle breathing process of marine power battery packs at sea.

[0034] Sub-cavity boundary data is structural boundary data, gas connection locations, and boundary data of each local space extracted from the spatial distribution of compressible gas inside the battery pack housing, used to divide the gas sub-cavities.

[0035] The gas sub-cavity division result is the result of dividing the compressible gas space inside the battery pack housing into multiple gas sub-cavities based on the sub-cavity boundary data. The gas sub-cavity division result includes the spatial boundary and sub-cavity volume of each gas sub-cavity.

[0036] Optionally, S3 includes:

[0037] Based on the gas sub-cavity division results, the spatial boundary, adjacent interfaces, ventilation openings, and partition positions of each gas sub-cavity are extracted to form sub-cavity boundary connectivity data. Based on this data, the gas connectivity status between any gas sub-cavity and its adjacent sub-cavities is determined. Gas sub-cavities interconnected through ventilation openings and capable of jointly transmitting pressure changes are identified as connected gas sub-cavities, forming connected gas sub-cavity groups used to constitute hyperedges. Based on these groups and the sub-cavity boundary connectivity data, the pressure transmission range corresponding to each connected gas sub-cavity group is determined, and the pressure transmission range is associated with the corresponding connected gas sub-cavity group to form connected pressure range data. Based on the connected pressure range data and the installation positions of temperature and pressure measuring points, the temperature and pressure measuring points are matched to their respective gas sub-cavities. The connected gas sub-cavities are grouped to form a measurement point affiliation relationship. Based on the gas sub-cavity division results and the measurement point affiliation relationship, a node set is established. The nodes in the node set include gas sub-cavities, temperature measurement points, and pressure measurement points, and the temperature and pressure measurement points are assigned to their corresponding gas sub-cavities. Based on the connected gas sub-cavity grouping, connected pressure range data, and node set, the nodes corresponding to the gas sub-cavities within the same connected gas sub-cavity group are organized into a hyperedge, and the temperature and pressure measurement points belonging to the same pressure transmission range are associated with the hyperedge to form a hyperedge set. Based on the node set and the hyperedge set, a gas sub-cavity connectivity hypergraph is constructed, so that each hyperedge in the gas sub-cavity connectivity hypergraph is composed of the nodes corresponding to the connected gas sub-cavities, and the gas sub-cavity connectivity hypergraph retains the spatial boundaries and sub-cavity connectivity relationships in the gas sub-cavity division results.

[0038] Terminology Explanation:

[0039] Sub-cavity boundary connectivity data consists of the spatial boundary, adjacent interface, ventilation opening and partition position of each gas sub-cavity extracted based on the gas sub-cavity division results, used to determine whether the gas sub-cavities are connected.

[0040] It can transmit pressure changes together, and multiple gas sub-cavities are connected through ventilation openings without pressure blockage caused by partitions, so that pressure changes in one gas sub-cavity can be transmitted to adjacent gas sub-cavities.

[0041] A connected gas sub-cavities are gas sub-cavities that are interconnected through ventilation openings and can jointly transmit pressure changes.

[0042] Connecting gas sub-cavities are grouped together by grouping multiple gas sub-cavities that can jointly transmit pressure changes. Each connecting gas sub-cavities group corresponds to a connecting gas region that can jointly bear the changes in cold contraction negative pressure.

[0043] Pressure transmission range is the spatial range within which pressure changes in the same connected gas sub-cavities can propagate along the gas connection points and be characterized by both temperature and pressure measuring points.

[0044] The connected pressure range data is the data formed by associating each connected gas sub-cavity group with its corresponding pressure transmission range.

[0045] The attribution of measuring points is determined by analyzing the connected pressure range data and the installation locations of temperature and pressure measuring points to establish the correspondence between the gas sub-cavities where the temperature and pressure measuring points are located and the corresponding connected gas sub-cavities.

[0046] A hyperedge is a connection relationship formed by nodes corresponding to gas sub-cavities within the same connected gas sub-cavities group, used to indicate that multiple gas sub-cavities within the same pressure transmission range jointly participate in the cold contraction negative pressure change.

[0047] The gas sub-cavity connected hypergraph is a hypergraph structure constructed with gas sub-cavities, temperature measuring points, and pressure measuring points as nodes, and nodes corresponding to the connected gas sub-cavities as hyperedges. It is used to transform the spatial distribution of compressible gas inside the battery pack housing into a structural input that can be processed by the hypergraph neural network.

[0048] Optionally, S4 includes:

[0049] The temperature measurement point is calculated from the module temperature time series to determine the temperature change direction between adjacent acquisition times, and the acquisition period with the temperature change direction decreasing is determined as the temperature decrease interval, thus forming the temperature decrease interval and the temperature change used for writing to the node.

[0050] The pressure response range is determined based on the temperature drop range in the internal pressure timing of the battery pack housing, and the pressure difference between the pressure measurement points within the pressure response range is calculated to form the pressure change used for writing nodes.

[0051] Based on the node composition of the gas sub-cavity connected hypergraph, the connection relationship between the temperature measuring point and the corresponding node of the gas sub-cavity, as well as the connection relationship between the pressure measuring point and the corresponding node of the gas sub-cavity, are determined to form the measurement point node correspondence relationship.

[0052] Based on the correspondence between measurement point nodes, temperature changes are assigned to the nodes corresponding to the corresponding gas sub-cavities, and pressure changes are assigned to the nodes corresponding to the corresponding gas sub-cavities, forming node temperature changes and node pressure changes; based on the gas sub-cavity division results, the sub-cavity volume of each gas sub-cavity is determined, and the node temperature changes, node pressure changes, and sub-cavity volumes are written into the nodes of the gas sub-cavity connectivity hypergraph to form node compensation volume data.

[0053] Based on the nodes corresponding to the connected gas sub-cavities contained in each hyperedge of the gas sub-cavity connectivity hypergraph, the sub-cavity connectivity relationships between gas sub-cavities within the same hyperedge are determined, forming hyperedge connectivity data;

[0054] Write the hyperedge connectivity data into the hyperedge of the gas sub-cavity connectivity hypergraph, and update the gas sub-cavity connectivity hypergraph according to the node compensation volume data and the hyperedge connectivity data, so that temperature change, pressure change, sub-cavity volume and sub-cavity connectivity relationship correspond to the same connected gas sub-cavity, and obtain the sub-cavity connectivity hypergraph features.

[0055] Terminology Explanation:

[0056] The temperature drop interval is the sampling period in the module temperature time series where the temperature change of the temperature measurement point between adjacent sampling times is in the direction of decrease;

[0057] The pressure response interval is the data acquisition period determined based on the internal pressure time series of the battery pack housing according to the temperature drop interval, used to characterize the response of the internal pressure of the battery pack housing to the temperature drop process.

[0058] The correspondence between measurement points and nodes is determined based on the node structure of the gas sub-cavity connected hypergraph, specifically the correspondence between temperature measurement points and the corresponding nodes of the gas sub-cavities, as well as the correspondence between pressure measurement points and the corresponding nodes of the gas sub-cavities.

[0059] The node compensation volume data consists of node temperature changes, node pressure changes, and sub-cavity volumes in the nodes written into the gas sub-cavity connectivity hypergraph, which are used to represent the local physical quantities of the corresponding gas sub-cavities in the cold contraction negative pressure compensation.

[0060] Hyperedge connectivity data refers to the connectivity relationships of sub-cavities written into the hyperedge of the gas sub-cavity connectivity hypergraph, used to represent the pressure transmission path between gas sub-cavities within the same hyperedge;

[0061] The sub-cavity connectivity hypergraph feature is formed by writing the node compensation volume data into the nodes of the gas sub-cavity connectivity hypergraph and writing the hyperedge connectivity data into the hyperedge of the gas sub-cavity connectivity hypergraph. The sub-cavity connectivity hypergraph feature makes temperature changes, pressure changes, sub-cavity volume and sub-cavity connectivity relationships correspond to the same connected gas sub-cavity.

[0062] Optionally, S5 includes:

[0063] The sub-cavity connectivity hypergraph features are input into the hypergraph neural network. Based on the node and hyperedge connectivity relationships of the gas sub-cavity connectivity hypergraph, the temperature changes, pressure changes, and sub-cavity volumes of the written nodes are aggregated within the hyperedges to form a hyperedge compensation state.

[0064] Based on the sub-cavity connectivity, the super-edge compensation state is written back to the node corresponding to the gas sub-cavity contained in the super-edge, and the temperature change, pressure change and sub-cavity volume within the node are updated to form the node compensation state.

[0065] Based on the node compensation state and the hyperedges of the gas sub-cavity connectivity hypergraph, calculate the volume contribution of the connected gas sub-cavity corresponding to each hyperedge during the cooling and shrinking stage, and generate a local compensation volume contribution value.

[0066] During the readout training phase, the local compensation volume contribution values ​​corresponding to each hyperedge are aggregated according to the sub-cavity connectivity to generate the airbag compensation volume output value.

[0067] The airbag compensation volume output value is compared with the cold shrink negative pressure compensation volume reference value. When the airbag compensation volume output value is lower than the cold shrink negative pressure compensation volume reference value, the volume shortage is obtained. When the airbag compensation volume output value is higher than the cold shrink negative pressure compensation volume reference value, the volume surplus is obtained.

[0068] During the readout training phase, the unilateral capacity shortage loss is calculated based on the capacity shortage and capacity surplus. In accordance with the constraint that the loss corresponding to the capacity shortage is greater than the loss corresponding to the capacity surplus, the convergence relationship between the local compensation volume contribution value and the airbag compensation volume output value is updated to form a capacity constraint convergence relationship. Based on the capacity constraint convergence relationship, the local compensation volume contribution value is converged to obtain the airbag compensation volume discrimination value.

[0069] Terminology Explanation:

[0070] Hypergraph neural network is a neural network that receives the features of the sub-cavity connected hypergraph and transmits information according to the node and hyperedge connection relationship of the gas sub-cavity connected hypergraph. The hypergraph neural network aggregates the temperature change, pressure change and sub-cavity volume of the written node within the same hyperedge, and writes the aggregated information back to the node corresponding to the gas sub-cavity contained in the hyperedge, so that the connected gas sub-cavity corresponding to each hyperedge forms a local compensation volume contribution value during the cooling and shrinkage stage.

[0071] Hyperedge aggregation refers to the aggregation of temperature changes, pressure changes, and sub-cavity volumes in the nodes corresponding to the gas sub-cavities contained in the same hyperedge of the gas sub-cavity connected hypergraph, in order to form a processing that characterizes the cold contraction negative pressure compensation state of the same connected gas region.

[0072] The super-edge compensation state is the state formed after the temperature change, pressure change and sub-cavity volume of the write node are aggregated within the super-edge. It is used to represent the compensation volume related state of the connected gas sub-cavities corresponding to the same super-edge during the cooling and shrinking stage.

[0073] Write-back is the process of transmitting the superedge compensation state to the node corresponding to the gas sub-cavity contained in the superedge according to the sub-cavity connectivity, and using it to update the temperature change, pressure change and sub-cavity volume within the node.

[0074] The node compensation state is the state formed after updating the temperature change, pressure change, and sub-cavity volume within the corresponding node of the gas sub-cavity based on the hyperedge compensation state;

[0075] The local compensation volume contribution value is the volume calculated based on the node compensation state and the hyperedge of the gas sub-cavity connectivity hypergraph, representing the contribution of each hyperedge-corresponding connected gas sub-cavity to the airbag compensation volume output value during the cooling and shrinking stage.

[0076] The readout training phase is the phase in which the local compensation volume contribution values ​​corresponding to each hyperedge are aggregated into the airbag compensation volume output value according to the sub-cavity connectivity relationship, and the aggregation relationship is trained and updated based on the cold shrinkage negative pressure compensation volume reference value.

[0077] The airbag compensation volume output value is an output value formed by the aggregation of the local compensation volume contribution values ​​corresponding to each hyperedge during the readout training phase, which represents the compensation volume that the airbag to be configured needs to provide.

[0078] The capacity shortage is the difference that occurs when the airbag compensation volume output value is lower than the reference value of the cold shrinkage negative pressure compensation volume. It is used to represent the portion of the cold shrinkage negative pressure compensation requirement that the airbag compensation volume output value does not cover.

[0079] The capacity margin is the difference that occurs when the airbag compensation volume output value is higher than the reference value of the cold shrinkage negative pressure compensation volume. It is used to represent the portion of the airbag compensation volume output value that exceeds the cold shrinkage negative pressure compensation requirement.

[0080] The one-sided capacity shortage loss is a training constraint embedded in the readout training process. The one-sided capacity shortage loss is formed based on the capacity shortage and capacity surplus, and the loss corresponding to the capacity shortage is greater than the loss corresponding to the capacity surplus. Thus, in the process of the local compensation volume contribution value converging into the airbag compensation volume output value, a greater constraint is applied to the direction where the airbag compensation volume output value is lower than the cold shrink negative pressure compensation volume reference value.

[0081] The convergence relationship between the local compensation volume contribution value and the airbag compensation volume output value is the corresponding relationship for converting the local compensation volume contribution value corresponding to each hyperedge into the airbag compensation volume output value during the readout training phase.

[0082] The capacity-constrained convergence relationship is the convergence relationship updated after calculating the one-sided capacity shortage loss based on the capacity shortage and capacity surplus, and according to the constraint that the loss corresponding to the capacity shortage is greater than the loss corresponding to the capacity surplus.

[0083] The airbag compensation volume discrimination value is a compensation volume discrimination value obtained by aggregating the local compensation volume contribution values ​​according to the capacity constraint aggregation relationship, and is used to compare with the effective compensation volume of the airbag to be configured.

[0084] Optionally, S6 includes:

[0085] Receive the effective compensation volume of the airbag to be configured and the airbag compensation volume discrimination value, and form a capacity comparison difference based on the volume difference between the effective compensation volume of the airbag to be configured and the airbag compensation volume discrimination value;

[0086] Based on the capacity comparison difference, it is determined whether the effective compensation volume of the airbag to be configured can cover the airbag compensation volume discrimination value, thus forming a capacity coverage relationship;

[0087] When the capacity coverage relationship indicates that the effective compensation volume of the airbag to be configured is not less than the airbag compensation volume discrimination value, the output airbag capacity satisfies the discrimination result.

[0088] When the capacity coverage relationship indicates that the effective compensation volume of the airbag to be configured is less than the airbag compensation volume discrimination value, the output airbag capacity insufficient discrimination result is obtained.

[0089] Terminology Explanation:

[0090] The capacity comparison difference is the volume difference between the effective compensation volume of the airbag to be configured and the airbag compensation volume discrimination value.

[0091] The capacity coverage relationship is a relationship formed by determining whether the effective compensation volume of the airbag to be configured is not less than the airbag compensation volume discrimination value based on the capacity comparison difference.

[0092] The airbag capacity meets the discrimination result, which is the discrimination result formed when the effective compensation volume of the airbag to be configured is not less than the discrimination value of the airbag compensation volume;

[0093] The airbag capacity insufficiency judgment result is the judgment result generated when the effective compensation volume of the airbag to be configured is less than the airbag compensation volume judgment value.

[0094] Optionally, the formation of connected gas sub-cavities groups and hyperedge sets includes: reading the ventilation openings and isolation positions between any gas sub-cavity and its adjacent gas sub-cavities based on the sub-cavity boundary connectivity data; when there is a ventilation opening between any gas sub-cavity and its adjacent gas sub-cavities, and there is no isolation position that blocks pressure changes, then any gas sub-cavity and its adjacent gas sub-cavities are assigned to the same connected gas sub-cavity group; when there is no ventilation opening between any gas sub-cavities and its adjacent gas sub-cavities, or there is an isolation position that blocks pressure changes, then any gas sub-cavities and their adjacent gas sub-cavities are not assigned to the same connected gas sub-cavities group; based on the pressure transmission range corresponding to the same connected gas sub-cavities group, the nodes corresponding to the gas sub-cavities within the same connected gas sub-cavities group and belonging to the same pressure transmission range are organized into the same hyperedge, forming a hyperedge set.

[0095] Optionally, the formation of node compensation volume data includes: writing temperature changes into the node corresponding to the gas sub-cavity of the temperature measuring point according to the correspondence between the temperature drop range and the measuring point node; writing pressure changes into the node corresponding to the gas sub-cavity of the pressure measuring point according to the correspondence between the pressure response range and the measuring point node; when the temperature measuring point corresponds to multiple gas sub-cavities, distributing the temperature changes to the nodes corresponding to the multiple gas sub-cavities according to the correspondence between the measuring point node; when the pressure measuring point corresponds to multiple gas sub-cavities, distributing the pressure changes to the nodes corresponding to the multiple gas sub-cavities according to the correspondence between the measuring point node; and writing the sub-cavity volume of each gas sub-cavity into the node corresponding to the same gas sub-cavity, so that the node corresponding to the same gas sub-cavity forms node compensation volume data containing node temperature changes, node pressure changes, and sub-cavity volumes.

[0096] Optionally, the formation of the capacity constraint convergence relationship includes: when the airbag compensation volume output value is lower than the cold shrink negative pressure compensation volume reference value, updating the convergence relationship of the local compensation volume contribution value based on the capacity shortage to the airbag compensation volume output value; when the airbag compensation volume output value is higher than the cold shrink negative pressure compensation volume reference value, updating the convergence relationship based on the capacity surplus; setting the update effect of the loss corresponding to the capacity shortage on the convergence relationship to be greater than the update effect of the loss corresponding to the capacity surplus on the convergence relationship; determining the updated convergence relationship as the capacity constraint convergence relationship; and converging the local compensation volume contribution value based on the capacity constraint convergence relationship to obtain the airbag compensation volume discrimination value.

[0097] The beneficial effects of this invention are:

[0098] (1) This invention proposes an improved method for determining the airbag compensation volume of a marine power battery pack. The method changes the treatment of the compressible gas space inside the battery pack from a single volume to a process of dividing the gas sub-cavities and processing the gas sub-cavities into connected hypergraphs. Connected gas sub-cavities are grouped based on structural boundaries, gas connection locations, ventilation openings, and partition locations. Gas sub-cavities, temperature measuring points, and pressure measuring points are used as nodes, and connected gas sub-cavities capable of jointly transmitting pressure changes are used as hyperedges. Compared to existing algorithms that only use total volume or a single connectivity relationship, this method incorporates spatial boundaries, pressure transmission range, and measuring point affiliation into the same structural input. This ensures that the determination of the cold shrinkage negative pressure compensation volume no longer deviates from the local spatial division state inside the battery pack, and the obtained airbag compensation volume determination value corresponds to the sub-cavity connectivity relationship.

[0099] (2) This invention proposes a hypergraph neural network processing method based on the characteristics of sub-cavity connected hypergraphs. Within the temperature decrease interval and pressure response interval, temperature changes, pressure changes, and sub-cavity volumes are written into the corresponding nodes, and the sub-cavity connectivity relationships are written into the hyperedges. Through hyperedge aggregation, hyperedge compensation state write-back, and node compensation state update, the local compensation volume contribution value of each hyperedge is generated. Compared with the scheme that directly maps the temperature change, pressure change, and total compressible gas volume into a single capacity result, this method enables the contributions of different connected gas sub-cavities to the compensation volume during the cooling and contraction stage to be formed and aggregated separately. It can express the process of multiple local spaces jointly bearing the pressure change, and it is also convenient to retain the source relationship between temperature measurement points, pressure measurement points, and gas sub-cavities.

[0100] (3) This invention proposes a readout training step that embeds a unilateral capacity shortage loss. The difference between the airbag compensation volume output value and the reference value of the cold shrinkage negative pressure compensation volume is taken as the capacity shortage, and the difference above the reference value of the cold shrinkage negative pressure compensation volume is taken as the capacity surplus. The loss corresponding to the capacity shortage is set to be greater than the loss corresponding to the capacity surplus, so as to form a capacity constraint convergence relationship. Compared with the readout algorithm that symmetrically processes positive and negative deviations, this method applies a stronger constraint on the capacity shortage direction when training the convergence relationship, making the final airbag compensation volume discrimination value more suitable for capacity coverage determination in the marine temperature cycle breathing process. After directly comparing the effective compensation volume of the airbag to be configured with the airbag compensation volume discrimination value, the airbag capacity satisfaction discrimination result or airbag capacity insufficiency discrimination result can be output, so that the airbag configuration discrimination process is connected with the cold shrinkage negative pressure compensation scenario. Attached Figure Description

[0101] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0102] Figure 1This is a flowchart of the deep learning-based method for determining the airbag compensation volume of marine power battery packs proposed in this invention.

[0103] Figure 2 This is a flowchart of the marine power battery pack operation data acquisition method based on deep learning proposed in this invention.

[0104] Figure 3 This is a flowchart illustrating the determination of reference values ​​for cold shrinkage negative pressure compensation volume and the division of gas sub-cavities in the deep learning-based marine power battery pack airbag compensation volume discrimination method proposed in this invention.

[0105] Figure 4 This is a flowchart of the gas sub-cavity connected hypergraph construction process for the deep learning-based marine power battery pack airbag compensation volume discrimination method proposed in this invention.

[0106] Figure 5 This is a flowchart of the sub-cavity connected hypergraph feature formation for the deep learning-based marine power battery pack airbag compensation volume discrimination method proposed in this invention.

[0107] Figure 6 This is a flowchart illustrating the generation of airbag compensation volume discrimination values ​​in the deep learning-based marine power battery pack airbag compensation volume discrimination method proposed in this invention.

[0108] Figure 7 This is a flowchart illustrating the airbag capacity determination process of the deep learning-based marine power battery pack airbag compensation volume determination method proposed in this invention. Detailed Implementation

[0109] In Example 1, reference Figures 1 to 7 A deep learning-based method for determining the airbag compensation volume of marine power battery packs includes:

[0110] S1. Obtain marine power battery pack operation data, including the internal pressure sequence of the battery pack housing during the marine temperature cycle breathing process, the module temperature sequence, the spatial distribution of compressible gas inside the battery pack housing, and the effective compensation volume of the airbag to be configured.

[0111] S2. Determine the reference value of the cold shrinkage negative pressure compensation volume based on the internal pressure sequence and module temperature sequence of the battery pack housing, and form the gas sub-cavity division result based on the spatial distribution of compressible gas inside the battery pack housing.

[0112] S3. Construct a gas sub-cavity connected hypergraph based on the gas sub-cavity partitioning results. The nodes of the gas sub-cavity connected hypergraph include gas sub-cavities, temperature measuring points, and pressure measuring points. The hyperedges of the gas sub-cavity connected hypergraph are formed by the nodes corresponding to the connected gas sub-cavities.

[0113] S4. Write the temperature change, pressure change, and sub-cavity volume into the nodes of the gas sub-cavity connected hypergraph, and write the sub-cavity connectivity relationship into the hyperedge of the gas sub-cavity connected hypergraph to obtain the sub-cavity connected hypergraph features.

[0114] S5. Input the sub-cavity connected hypergraph features into the hypergraph neural network to generate local compensation volume contribution values. In the readout training stage where the local compensation volume contribution values ​​converge into the airbag compensation volume output value, embed the unilateral capacity shortage loss. The difference between the airbag compensation volume output value and the reference value of the cold shrinkage negative pressure compensation volume is determined as the capacity shortage amount. The difference between the airbag compensation volume output value and the reference value of the cold shrinkage negative pressure compensation volume is determined as the capacity surplus amount. According to the constraint that the loss corresponding to the capacity shortage amount is greater than the loss corresponding to the capacity surplus amount, the airbag compensation volume discrimination value is obtained.

[0115] S6. Compare the effective compensation volume of the airbag to be configured with the airbag compensation volume discrimination value. When the effective compensation volume of the airbag to be configured is not less than the airbag compensation volume discrimination value, the output airbag capacity meets the discrimination result. When the effective compensation volume of the airbag to be configured is less than the airbag compensation volume discrimination value, the output airbag capacity is insufficient discrimination result.

[0116] In this embodiment, step S1 includes:

[0117] When the marine power battery pack is undergoing temperature cycling at sea, the start and end intervals for data collection are recorded as follows: , The start time of data collection. At the end of the data collection, the marine temperature cycle breathing process refers to the process by which the gas inside the marine power battery pack expands and contracts with temperature changes under the periodic rise and fall of the marine ambient temperature, causing the internal pressure of the battery pack to change with the time of data collection. Let the set of pressure measurement points be denoted as . , For the first Pressure measurement points inside the battery pack housing. The number of pressure measurement points inside the battery pack housing is denoted as , and the original pressure sampling sequence corresponding to each pressure measurement point inside the battery pack housing is denoted as . , For the first The first pressure measurement point inside the battery pack housing At each pressure value acquisition time, For the corresponding pressure value, For the first The number of pressure values ​​sampled at each pressure measuring point inside the battery pack housing, for according to Arranged from smallest to largest, forming a sequence of pressure values ​​organized according to the time of collection;

[0118] During the same marine temperature cycle breathing process that forms the internal pressure sequence of the battery pack housing, let the set of module temperature measurement points be... , For the first Temperature measurement points for each module. The number of temperature measurement points in the module is denoted as , and the original temperature sampling sequence corresponding to each module temperature measurement point is denoted as . , For the first The first module temperature measurement point At which temperature values ​​were collected, For the corresponding temperature value, For the first The number of temperature value samples taken from each module's temperature measurement points, for according to Arranged from smallest to largest, forming a temperature value sequence organized according to the time of collection;

[0119] To ensure that the internal pressure timing of the battery pack housing and the module temperature timing are at the same time index, a data acquisition control clock is used to form candidate acquisition times, and in... A unified timeline is obtained by filtering by time sequence. , To standardize the number of data acquisition points in the timeline, the pressure matching window is set as follows: Temperature matching window is For each ,exist Selecting from those that meet the requirements And the pressure value with the smallest absolute time difference is taken as ,exist Selecting from those that meet the requirements And the temperature value with the smallest absolute time difference is taken as If a candidate acquisition time does not simultaneously meet the matching conditions of all pressure measurement points and all module temperature measurement points, the corresponding candidate acquisition time will not be included. This forms the internal pressure timing of the battery pack housing. ,in , dimension And form the module temperature timing. ,in , dimension ;

[0120] When obtaining the compressible gas space distribution inside the battery pack housing, a structural coordinate reference consistent with the installation positions of the pressure measuring points and module temperature measuring points inside the battery pack housing is used. The inner wall of the battery pack housing is taken as the outer structural boundary, and the outlines of the battery modules, structural components, connectors, and seals are taken as the occupied boundaries. The space inside the battery pack housing that is not occupied by battery modules, structural components, connectors, and seals and can accommodate gas compression or expansion is defined as the compressible gas space. Local spaces formed by structural components are identified based on the structural boundaries. Gas communication positions are identified based on the positions of openings, gaps, and vents. The corresponding volume is determined based on the geometric boundaries of the local spaces. The compressible gas space distribution inside the battery pack housing is denoted as... , For the number of local spaces, For the first A five-field record in a local space, This refers to the location of a local space inside the battery pack housing. For the structural boundary of the local space, For the volume distribution of local space, This represents the gas connectivity location corresponding to a local space. The source of subcavity volume is determined based on structural boundaries, gas connection locations, and local spaces formed by structural components;

[0121] When obtaining the effective compensation volume of the airbag to be configured, let the internal volume of the airbag at the lower limit of its effective stroke be... The internal volume of the airbag to be configured at the upper limit of its effective stroke is: The lower limit of the effective stroke is determined by a preset compression limit, and the upper limit of the effective stroke is determined by a preset expansion limit. These limits are obtained based on the volume calibration data of the airbag to be configured or the geometric boundaries of the cavity within the effective stroke. and ,Will and The difference is determined as the effective compensation volume of the airbag to be configured. The input organization used for processing gas sub-cavity partitioning results, constructing gas sub-cavity connected hypergraphs, and forming sub-cavity connected hypergraph features includes dimensions of... The internal pressure sequence and dimensions of the battery pack housing are as follows: Module temperature timing, including The five fields record the distribution of compressible gas space inside the battery pack housing, as well as the one-dimensional effective compensation volume of the airbag to be configured.

[0122] In this embodiment, step S2 includes:

[0123] Using the internal pressure timing of the battery pack housing Module temperature timing and the distribution of compressible gas space inside the battery pack housing As input, , To standardize the number of data collection points in the timeline. Indicates the first Pressure values ​​at various points inside the battery pack housing at specific acquisition times. This refers to the number of pressure measurement points inside the battery pack housing. Indicates the first Temperature values ​​at module temperature measurement points at each acquisition time. The number of temperature measurement points in the module, and the spatial distribution of compressible gas inside the battery pack housing are represented as follows: , For the number of local spaces, For the first A five-field record in a local space, This refers to the location of a local space inside the battery pack housing. For the structural boundary of the local space, For the volume distribution of local space, This represents the gas connectivity location corresponding to a local space. This serves as the source of sub-cavity volume for the local space;

[0124] When extracting the temperature decrease process from the module temperature time series, the temperature values ​​of all module temperature measurement points within the same acquisition time are converted into the module representative temperature. The module represents temperature. Depend on The average of all temperature values ​​is used to obtain the value. Let the preset temperature drop threshold be... The module representative temperature is compared between adjacent acquisition times. Below And the temperature drop is not less than At that time, the index To Index The corresponding data collection period is determined as the decreasing data collection period. Consecutive decreasing data collection periods are merged to form a set of temperature decreasing processes. , The number of temperature drop processes. Indicates the first The start and end indices of a temperature decrease process;

[0125] When mapping the internal pressure sequence of the battery pack enclosure to the temperature drop process, the pressure values ​​of all internal pressure measurement points of the battery pack enclosure within the same acquisition time are converted into the representative pressure of the enclosure. The box represents pressure. Depend on The average of all pressure values ​​is used to obtain the value. Let the preset pressure drop threshold be... Let the preset pressure hysteresis window length be... For each temperature drop process In the index range The search function identifies the period during which the pressure represented by the chamber decreases, and the chamber's pressure falls within the index range. The memory pressure drop is no less than During the data collection period, the temperature drop process will be considered. Binding to the pressure reduction process that meets the conditions, when the index interval When memory experiences multiple pressure reduction processes that meet certain conditions, the starting index for the pressure reduction is selected. The process of reducing pressure by minimizing the index distance is used for binding, and after binding, a cold contraction negative pressure correspondence is formed. , To complete the binding of the corresponding relationships, This represents the starting index of the temperature decrease process, the ending index of the temperature decrease process, the starting index of the pressure decrease process, and the ending index of the pressure decrease process.

[0126] When calculating the volume compensation requirement based on the relationship between cold contraction and negative pressure, the compressible gas space distribution inside the battery pack housing is considered. The volume distribution of each local space Convert the contents to their corresponding local volumes, and sum all the local volumes to obtain the total compressible gas volume. Corresponding relationship for each cold contraction negative pressure ,according to and Determine the module temperature change based on and Determine the pressure change inside the battery pack housing, and set the preset volume conversion record as follows: , To preset the number of volume conversion records, , To calibrate the temperature drop rate, To calibrate the pressure drop, To calibrate the total volume of compressible gas, To calibrate the compensation volume, a preset volume conversion record was obtained by collecting temperature drop data from a marine power battery pack of the same model under known compensation volume conditions. The preset volume conversion record covers the range of values ​​for module temperature change, battery pack internal pressure change, and total compressible gas volume. The current cooling and contraction negative pressure correspondence is then used to determine the module temperature change, battery pack internal pressure change, and... respectively with In , and The matching process is performed based on the three fields meeting their respective preset matching tolerances. If multiple matching records exist, the selected record is chosen. The largest matching record, and the selected The volume compensation requirement corresponding to the current cold contraction negative pressure relationship is determined, and the maximum value among all volume compensation requirements corresponding to the cold contraction negative pressure relationships is determined as the reference value for cold contraction negative pressure compensation volume. ;

[0127] When extracting structural boundaries and gas connectivity locations based on the spatial distribution of compressible gas inside the battery pack housing, each local space record is read. ,Will , , , and Write subcavity boundary data Subcavity boundary data , , used to indicate the first The location, structural boundaries, gas connectivity, sub-cavity volume sources, and volume distribution of each local space; when dividing the gas sub-cavities based on the sub-cavity boundary data, the structural merging relationship is assumed to be... Structural merging relationship Used to indicate whether two local spaces belong to the same closed or semi-closed compressible gas space, for any two local spaces corresponding to and ,when Structural boundaries and The structural boundaries belong to the same continuous spatial boundary, and and When there is no separating boundary formed by structural components, Write structure merge relation ,when and When there are separation boundaries formed by structural components, do not Write structure merge relation According to the structural merging relationship By connecting and merging local spaces, local spaces that can be reached from each other through structural merging relationships are grouped into the same gas sub-cavity, forming a set of gas sub-cavities. , This refers to the number of gas sub-cavities. For the first Each gas sub-cavity contains a set of local spaces, and for each gas sub-cavity... ,according to The structural boundaries of all local spaces within the gas sub-cavity form the spatial boundaries of the gas sub-cavity. ,according to The sub-cavity volume of the gas sub-cavity is obtained by summing the volume distribution of all local spaces within the sub-cavity. The gas sub-cavity partitioning result is each , is used to represent the set of local spaces contained in a gas subcavity, the spatial boundary of a gas subcavity, and the subcavity volume of a gas subcavity.

[0128] In this embodiment, step S3 includes:

[0129] Results of gas sub-cavity partitioning Temperature measuring point installation location set Collection of pressure measuring point installation locations As input, This refers to the number of gas sub-cavities. Indicates the first A gas sub-cavity, For the first The set of local spaces contained within a gas sub-cavity For the first The spatial boundary of each gas sub-cavity For the first The sub-cavity volume of each gas sub-cavity and the set of temperature measuring point installation locations are as follows: , For the first The installation positions of the temperature measuring points under the coordinate reference inside the battery pack housing. The set of temperature measuring points and pressure measuring point installation locations is: , For the first The installation positions of the pressure measuring points under the coordinate reference inside the battery pack housing. This represents the number of pressure measurement points.

[0130] When extracting the spatial boundary of each gas sub-cavity based on the gas sub-cavity partitioning results, read... The inner wall boundaries, structural component boundaries, connector boundaries, and sealing component boundaries of any two gas sub-cavities and ,when and Contact tolerance at preset boundaries The memory exists on a common interface, and the area of ​​the common interface is not less than a preset threshold for the area of ​​adjacent interfaces. When common interfaces are defined as adjacent interfaces, and obtain adjacent interfaces. area ,when and When the adjacent interface area condition is not met, the gas sub-cavity is not... and gas sub-cavity Generate adjacent interface records;

[0131] Determining adjacent interfaces Then, based on the gas connectivity positions from the compressible gas spatial distribution inside the battery pack housing in the gas sub-cavity division results, from the adjacent interface... Extract the openings, gaps, and ventilation slots to form ventilation openings. When adjacent interfaces When there are multiple openings, gaps, or vents, the total ventilable contour area of ​​each opening, gap, or vent is used to obtain the ventilation opening. Effective opening area According to the ventilation opening Penetrating adjacent interfaces along the direction of pressure propagation The geometric distance is used to obtain the length of the ventilation opening. According to adjacent interfaces Extract the partition location from the solid blocking area formed by structural components, connectors, or seals. and the partition position Projected onto the ventilation opening The location of the partition is obtained from the ventilated outline. Cover ventilation opening Effective opening projection area The spatial boundary, adjacent interface, adjacent interface area, ventilation opening, effective opening area, ventilation opening length, partition position, and effective opening projected area of ​​each pair of adjacent gas sub-cavities are written into the sub-cavity boundary connectivity data. ,in Each record in the database is ;

[0132] Based on the subcavity boundary connectivity data When determining the gas connection state, the gas connection state is used as the standard. Indicates the first The gas sub-cavity and the first Whether the pressure changes can be transmitted between the individual gas sub-cavities is determined by the following formula:

[0133] ;

[0134] in, It is in a gas-connected state. The value of 1 indicates the first The gas sub-cavity and the first The individual gas chambers are interconnected through ventilation openings and can collectively transmit pressure changes. A value of zero indicates the first The gas sub-cavity and the first The gas sub-cavities are not identified as connected gas sub-cavities. and All are gas sub-cavity indices. This is a conditional indicator; it takes the value of one if all conditions within the square brackets are true, and takes the value of zero if any condition within the square brackets is false. For adjacent interfaces area, To preset the adjacent interface area threshold, and , For ventilation openings The effective opening area To preset the ventilation opening area threshold, and , Partition location Cover ventilation opening The effective opening projection area To preset the threshold for the proportion of the blocking area, It is a dimensionless numerical value and satisfies , For ventilation openings Penetrating adjacent interfaces along the direction of pressure propagation Length, The preset reference length for the ventilation opening is, and , To preset the gas connectivity threshold, It is a dimensionless numerical value and satisfies , To process the smaller value, To obtain a larger value, the constants in the formula For dimensionless boundary values, in the formula , , , and All are area measurements. and All are length quantities, which are used in the same gas connectivity determination after being processed by ratio;

[0135] According to the gas connection state Establish a pressure-transmitting connection ,when When the value is one, Write compressible connection relationship ,when When the value is zero, do not Write compressible connection relationship According to the pressure-transmitting connection relationship Connect and merge the gas sub-cavities, starting from any gas sub-cavity and along... All gas sub-cavities reachable by the connection pairs in the diagram are identified as belonging to the same group of connected gas sub-cavities, thus forming connected gas sub-cavity groups. , The number of connected gas sub-cavities grouped together. For the first The set of gas sub-cavity indices contained in a connected gas sub-cavity group;

[0136] When determining the pressure transmission range based on the interconnected gas sub-cavities and sub-cavity boundary connectivity data, for each interconnected gas sub-cavity group... Read The spatial boundaries of all gas sub-cavities within the chamber were determined and read. The internal adjacent gas sub-cavities satisfy the following conditions The ventilation opening will The compressible gas space defined by the spatial boundaries of all internal gas sub-cavities, and satisfying The pressure transmission channel formed by the ventilation opening is defined as the pressure transmission range. Pressure transmission range This represents the spatial extent in which pressure changes can propagate along the gas connection points and are shared by groups of gas sub-cavities connected in the same gas connection. Each group of gas sub-cavities is then divided into... With pressure transmission range Correlation, forming connected pressure range data ;

[0137] Based on the connected pressure range data and the installation locations of temperature and pressure measuring points, a measuring point attribution relationship is established, assigning each temperature measuring point to a specific location. and the installation location of each pressure measuring point The installation location is relative to the spatial boundary of each gas sub-cavity. Perform an inclusion determination; when the installation location falls within... When the time is right, the measuring point will be assigned to the first... Each gas sub-cavity, when the installation position does not fall into any of the gas sub-cavities, During this time, calculate the installation location to each The boundary distance, and the minimum boundary distance is not greater than the preset measurement point belonging distance threshold. In the case of a minimum boundary distance, the measuring point is assigned to the gas sub-cavity corresponding to the minimum boundary distance. When multiple gas sub-cavities correspond to the same minimum boundary distance, the gas sub-cavity with the smaller index is selected as the measuring point's assignment object. Based on the connected gas sub-cavity group to which the assigned gas sub-cavity belongs, the connected gas sub-cavity group to which the measuring point belongs is determined, thus forming the measuring point assignment relationship. Measurement point attribution relationship Each record includes a measurement point category, a measurement point index, a gas sub-cavity index, and a connected gas sub-cavity grouping index;

[0138] A node set is established based on the gas sub-cavity division results and the measurement point attribution relationships. Node set Includes a set of gas sub-cavity nodes Temperature measurement point node set and pressure measurement point node set , For the first The nodes corresponding to each gas sub-cavity For the first The nodes corresponding to each temperature measuring point For the first The nodes corresponding to each pressure measuring point are determined according to the measuring point affiliation relationship. The temperature and pressure measuring points are assigned to the corresponding gas sub-cavity nodes to ensure that the installation positions of the measuring points are consistent with the spatial boundaries of the gas sub-cavity.

[0139] A hyperedge set is formed based on the connected gas sub-cavities grouping, connected pressure range data, and node set. Grouping each connected gas sub-cavity ,Will The gas sub-cavity node organization corresponding to the internal gas sub-cavity is hyperedge. Member set, hyperedge The member set is According to the attribution of measuring points and pressure transmission range It will belong to Temperature and pressure measurement nodes are associated with the hyperedge. and maintain super-edge The gas subcavity member matrix is ​​obtained by constructing the gas subcavity nodes corresponding to the connected gas subcavities. Temperature measurement point correlation matrix Correlation matrix with pressure measurement points The matrix value of 1 indicates that the corresponding node or measurement point belongs to the corresponding hyperedge, and the matrix value of 0 indicates that there is no belonging relationship.

[0140] Based on the node set Hyperedge set Gas subcavity member matrix Temperature measurement point correlation matrix Correlation matrix with pressure measurement points Constructing a gas sub-cavity connected hypergraph Gas sub-cavity connectivity hypergraph Include Each node and Each hyperedge corresponds to a connected gas sub-cavity group that can collectively transmit pressure changes. Gas sub-cavity nodes retain the spatial boundaries from the sub-cavity partitioning results. Hyperedges retain the sub-cavity connectivity relationships determined by the positions of ventilation openings and partitions. Temperature and pressure measurement point nodes are bound to their respective gas sub-cavities and connected gas sub-cavity groups through measurement point affiliation relationships. The gas sub-cavity connectivity hypergraph... The compressible gas space distribution inside the battery pack housing is represented by a structure with physically connected boundaries, so that temperature changes, pressure changes and sub-cavity volumes can be written into nodes and superedges with the same pressure transmission range.

[0141] In this embodiment, step S4 includes:

[0142] Using module temperature timing Internal pressure timing of the battery pack housing Gas sub-cavity partitioning results and gas sub-cavity connected hypergraph As input, To standardize the number of data collection points in the timeline. , Indicates the first Temperature values ​​at module temperature measurement points at each acquisition time. The number of temperature measuring points. Indicates the first Pressure values ​​at various points inside the battery pack housing at specific acquisition times. The number of pressure measurement points and the result of gas sub-cavity division. , Indicates the first A gas sub-cavity, For the first The volume of each gas sub-cavity, and the connectivity of the gas sub-cavities in the hypergraph. Includes a set of gas sub-cavity nodes Temperature measurement point node set Pressure measurement point node set Hyperedge set Records of the attribution of temperature measuring point nodes and gas sub-cavity nodes, records of the attribution of pressure measuring point nodes and gas sub-cavity nodes, and records of gas sub-cavity nodes contained in the super-edge.

[0143] When calculating the direction of temperature change from the module temperature time series, for each temperature measurement point Compare temperature values ​​at adjacent acquisition times and Let the preset temperature drop threshold be... , The temperature threshold used to determine whether the temperature drop is effective, when Below And the temperature drop is not less than At that time, the index To Index The corresponding data collection period was determined as the temperature measurement point. During the temperature drop sampling period, consecutive temperature drop sampling periods at the same temperature measuring point are merged to obtain a set of temperature drop intervals. , This represents the number of temperature drop intervals. , This is the starting index for the temperature decrease range. The index ends at the end of the temperature decrease range. To trigger the temperature measurement point index within the temperature drop range;

[0144] Let the write window length be... , Not greater than And greater than Integers, in implementable configurations For each temperature drop interval ,by As the starting index for the temperature writing window, for each temperature measurement point Extracting the index range from the module temperature time series The temperature value within the window, and the temperature value at each acquisition time within the window relative to... The differences are arranged in chronological order to form temperature measurement points. In the temperature drop range Temperature change sequence , Indicates length is The space of real number sequences, if End index beyond the unified timeline ,use The corresponding temperature difference values ​​are used to fill in the missing locations with boundary preservation, and the temperature change sequence is shown. Used to write temperature measurement point nodes and the corresponding gas sub-cavity nodes of temperature measurement points;

[0145] When determining the pressure response range based on the temperature drop range, the preset minimum pressure response hysteresis length is set as follows: Let the preset maximum pressure response hysteresis length be... ,in For each temperature drop interval Candidate starting index set For all conditions not less than Not greater than and not greater than Integer index, when the candidate starting index set When empty, Write to candidate starting index set Let the preset pressure drop threshold be... Let the preset lag penalty coefficient be... Based on the timing of the internal pressure of the battery pack housing, in the candidate starting index set The degree of continuous decrease within the range determines the start and end indices of the pressure response interval:

[0146] ;

[0147] in, For the first The starting index of the pressure response interval corresponding to each temperature decrease interval. For the first The index ends at the pressure response interval corresponding to each temperature decrease interval. For the candidate starting index set The starting index of the selected pressure response interval. Index for the temperature decrease range, As a candidate starting index, For the first The set of candidate starting indices corresponding to each temperature decrease interval. This indicates the selection operation for the candidate starting index that maximizes the value within the square brackets. The number of pressure measuring points. For the length of the write window, To preset the pressure drop threshold, Having pressure dimensions and , For pressure measurement point index, This is used to write the offset index of adjacent acquisition times within the window. For pressure measurement points In the index The pressure value at that location, For pressure measurement points In the index The pressure value at that location, To preset the lag penalty coefficient, It is a dimensionless numerical value and satisfies , For the first Starting index for each temperature decrease interval To preset the minimum hysteresis length of the pressure response, To preset the maximum hysteresis length of the pressure response, To express summation, This indicates that the smaller value will be used. This indicates that the larger value is taken, and the constant in the formula is used for this purpose. Zero pressure difference is represented in the pressure difference term, and it is a constant in the formula. For dimensionless boundary values, the cumulative term of pressure drop in the formula and The ratio is a dimensionless value, the hysteresis penalty term is a dimensionless value, and the whole thing in square brackets is used for selecting the pressure response range.

[0148] Start Index and end index Composition of pressure response range For each pressure measuring point Extracting the index interval from the internal pressure timing sequence of the battery pack housing The pressure value within the window is compared with the pressure value at each acquisition time within the window. The differences are arranged in chronological order to form pressure measurement points. Within the pressure response range Pressure change sequence Pressure change sequence Used to write pressure measurement point nodes and corresponding gas sub-cavity nodes for pressure measurement points.

[0149] Based on the node composition of the gas sub-cavity connected hypergraph, the correspondence between measurement point nodes is formed. Let the set of nodes corresponding to temperature measurement points and gas sub-cavities be denoted as . , Indicates the first Each temperature measurement node is connected in the gas sub-cavity hypergraph. The set of gas sub-cavity node indices to which the pressure measurement point belongs, let be the set of nodes corresponding to the gas sub-cavities. , Indicates the first Each pressure measurement node is connected in the gas sub-cavity hypergraph. The set of gas sub-cavity node indices to which the node belongs is determined by reading the gas sub-cavity connected hypergraph. The attribution records of the intermediate temperature measurement point nodes and the gas sub-cavity nodes are obtained. By reading the gas sub-cavity connectivity hypergraph The attribution records of the pressure measurement point nodes and gas sub-cavity nodes are obtained. When a temperature measuring point node or a pressure measuring point node belongs to only one gas sub-cavity node, the corresponding set contains one gas sub-cavity node index. When a temperature measuring point node or a pressure measuring point node belongs to multiple gas sub-cavity nodes, the corresponding set contains multiple gas sub-cavity node indices.

[0150] Temperature and pressure changes are allocated based on the correspondence between measuring points and nodes, and for each temperature drop interval... Each temperature measuring point Corresponding temperature change sequence Write For each gas sub-cavity, a node is assigned a candidate node for temperature change. When a gas sub-cavity node receives temperature change sequences from multiple temperature measurement points, the average of these sequences is taken at the same sampling location to form a sequence of length [length missing]. The node temperature change, when a gas sub-cavity node does not receive a temperature change sequence, will send a sequence of length 1000. The zero sequence is used as the node temperature change for each pressure measurement point. , pressure change sequence Write For each gas sub-cavity, a node is selected to form a candidate node pressure change for that gas sub-cavity node. When a gas sub-cavity node receives pressure change sequences from multiple pressure measurement points, the average of these multiple pressure change sequences is taken at the same sampling location to form a sequence of length [length missing]. The node pressure change, when a gas sub-cavity node does not receive the pressure change sequence, will send a sequence of length 1000. The zero sequence is used as the node pressure change;

[0151] The volume of each gas sub-cavity is determined based on the gas sub-cavity division results. The node temperature change, node pressure change, and sub-cavity volume are written into the gas sub-cavity nodes of the gas sub-cavity connectivity hypergraph to form node compensation volume data. Each gas sub-cavity node, the node compensation volume data includes a length of... The node temperature change and length are Nodal pressure changes and one-dimensional subcavity volume , for the Each temperature measurement point node is written to the temperature measurement point. The length is Temperature change sequence , length is The zero-pressure change sequence and one-dimensional zero-volume value, for the first There are 1 pressure measurement point node, and the write length is 1. Zero temperature change sequence, pressure measurement point The length is Pressure change sequence To maintain consistency with the node category distinction at the input of the hypergraph neural network, and with the one-dimensional zero-volume value, the node category value is written into the node features. The node category value of the gas sub-cavity node is... The node category value of the temperature measuring point node is The node category value of the pressure measuring point node is This forms the node feature matrix. ,exist At that time, the node feature matrix Each node is Dimensional node features, including Temperature changes Pressure changes The volume of the visceral cavity and Dimensional node category value;

[0152] Based on the nodes corresponding to the connected gas sub-cavities contained in each hyperedge of the gas sub-cavity connectivity hypergraph, the sub-cavity connectivity relationships between gas sub-cavities within the same hyperedge are determined. For the ... super edge Read the super edge The included gas sub-cavity node index forms a length of Subcavity connectivity vector , of which The value at position 1 indicates that the first position has a value of 1. Each gas subcavity node belongs to the hyperedge. , No. The value of zero at the position indicates that the [number]th position is [value]. Each gas subcavity node does not belong to the hyperedge. Connect the subcavities to the vector The number of connected gas sub-cavities and the total volume of the corresponding sub-cavities are written into the hyperedge. This forms hyperedge-connected data, where the number of connected gas sub-cavities is determined by the sub-cavity connectivity vector. The number of positions with a value of one is determined, and the total volume of the sub-cavities corresponding to the connected gas sub-cavities is determined by the volume of the sub-cavities corresponding to the positions with a value of one. The feature matrix of the hyperedge is determined by accumulation and formed from all hyperedge connected data. , To determine the number of sides, the compressible gas space distribution inside the battery pack housing is formed. In the feasible configuration of each gas sub-cavity, each hyperedge is 3D hyperedge features, including Viaduct cavity connectivity Number of interconnected gas sub-cavities and Total volume of sub-cavities corresponding to the interconnected gas sub-cavities;

[0153] Node feature matrix Write gas subcavity connected hypergraph The nodes will have hyperedge feature matrices Write gas subcavity connected hypergraph The hyperedges are used to obtain the features of the subcavity connected hypergraph. To form a pattern for the entire temperature drop range Arranged subcavity connected hypergraph feature sequence Features of each sub-cavity connected hypergraph Including node feature matrix Hyperedge feature matrix Gas sub-cavity connectivity hypergraph The node connectivity relationships and gas sub-cavity connectivity hypergraph Hyperedge connectivity, node feature matrix The dimension is Hyperedge feature matrix The dimension is ,exist , , , , In the feasible configuration, the sub-cavity connected hypergraph features include Each node Each superedge, each node Dimensional node features and each hyperedge The dimensional hyperedge feature makes temperature changes, pressure changes, sub-cavity volume, and sub-cavity connectivity correspond to the same connected gas sub-cavity.

[0154] In this embodiment, step S5 includes:

[0155] Employing subcavity connected hypergraph features Reference value for cold contraction negative pressure compensation volume As input, Index for the temperature decrease range, The number of temperature decrease intervals, and the connected hypergraph features of each sub-cavity. Including node feature matrix Hyperedge feature matrix Gas sub-cavity connectivity hypergraph The node connectivity relationships and gas sub-cavity connectivity hypergraph Hyperedge connectivity, node feature matrix The dimension is Each node feature includes a length of Temperature change, length Pressure changes, one-dimensional subcavity volume and one-dimensional node category values, hyperedge feature matrix The dimension is Each hyperedge feature includes The connectivity relationships of the one-dimensional connected gas sub-cavities, the number of one-dimensional connected gas sub-cavities, and the total volume of the corresponding one-dimensional connected gas sub-cavities are all available for configuration. , , , , Features of each sub-cavity connected hypergraph include Each node Each superedge, each node Dimensional node features and each hyperedge 3D hyperedge features;

[0156] The hypergraph neural network consists of a gas sub-cavity connected hypergraph feature input, a capacity-conserving lag hypergraph propagation, and a readout training stage embedding a one-sided capacity shortage loss. The node input mapping layer includes a cold-shrinkage driving branch and a negative pressure response branch. The cold-shrinkage driving branch receives node features from the hypergraph. 3D temperature change, 1D subcavity volume, and 1D node category value, and through Each output unit forms In the cold-shrink driving state, the characteristics of the receiving node in the negative pressure response branch are... Dimensional pressure changes and one-dimensional node category values, and through Each output unit forms The negative pressure response state will Cooling and shrinkage drive state and The negative pressure response state is spliced ​​as follows Dimensional nodes hide features, and the hyperedge input mapping layer receives the hyperedge feature matrix. Each super edge in 3D hyperedge features, and through Each output unit forms The dimensional hyperedge capacity state, the output units of both the node input mapping layer and the hyperedge input mapping layer contain connection weights, biases and linear rectification units;

[0157] The capacity-conserving hysteresis hypergraph propagation layer is set to two layers, each layer being configured as a gas sub-cavity connectivity hypergraph. The hyperedge connection relationship is propagated to the first... super edge Read the hyperedge features The connectivity of the subcavities is selected by choosing those belonging to the superedge. The gas subcavity node, according to its superedge The proportion of the sub-cavity volume of a gas sub-cavity node to the total sub-cavity volume of the connected gas sub-cavities is used to determine the selected gas sub-cavity nodes. The weighted aggregation of the cooling-contraction driven states forms a... In the ultra-extended shrinkage state, the same gas sub-cavity node selection method is used to select the gas sub-cavity nodes. The negative pressure response states are aggregated to form Dimensional super-edge negative pressure response state;

[0158] Within each capacity-conserving hysteresis hypergraph propagation layer, based on the temperature and pressure changes within the same hyperedge... The sequential relationship of each data collection moment will Each data acquisition time point is divided into four consecutive sub-windows. Within each sub-window, the index position where the temperature change first reaches the preset temperature drop condition is read, and the index position where the pressure change first reaches the preset pressure drop condition is read. If the index position where the pressure change first reaches the preset pressure drop condition is later than the index position where the temperature change first reaches the preset temperature drop condition, the pressure hysteresis weight of the corresponding sub-window is reset to a positive value. If no index position where the pressure change reaches the preset pressure drop condition occurs within the sub-window, the pressure hysteresis weight of the corresponding sub-window is reset to zero. The pressure hysteresis weights corresponding to the four consecutive sub-windows are arranged in chronological order to form... The pressure lag weighting and capacity conservation lag supergraph propagation layer adopts... Dimensional pressure hysteresis weight control Ultra-extreme cold contraction state and The fusion sequence of the super-edge negative pressure response states allows pressure changes to propagate along the sub-cavity connectivity defined by the same super-edge;

[0159] Fusion layer reception Ultra-extended cold contraction state, Dimensional super-edge negative pressure response state and The capacity state of the superedge is determined by... Each output unit forms The super-edge compensation state, based on the sub-cavity connectivity, will... Write back the state of the superedge compensation to the superedge For nodes corresponding to gas subcavities that belong to multiple hyperedges, the multiple hyperedges will be written back. The compensation state of the superedge is averaged according to the number of superedges to which the gas sub-cavity node belongs, and the averaged state is then... The dimensional state serves as the node compensation state for the gas sub-cavity nodes. The two capacity-conserving hysteresis hypergraph propagation layers sequentially perform hyperedge aggregation, pressure hysteresis weight constraint, fusion layer output, and write-back update, so that the node compensation state carries the temperature change, pressure change, and sub-cavity volume within the same connected gas sub-cavity group.

[0160] After the two-layer capacity-conserving lag hypergraph propagation layer is completed, for each hyperedge... of The contribution readout layer performs contribution readout on the hyperedge compensation state, and sets the contribution readout layer for each hyperedge. There are 1 fully connected neuron, and each fully connected neuron receives 16 fully connected neurons. Compensate for the state of the hyperedge and output a one-dimensional intermediate value. Composed of intermediate values The hidden features are contributed to a non-negative output neuron, which truncates its outputs that are less than zero to zero, generating the 1st generation. Local compensation volume contribution value corresponding to each superedge ,all The local compensation volume contribution values ​​corresponding to each hyperedge are arranged by hyperedge index as follows: Each local compensation volume contribution value is a one-dimensional volume.

[0161] The readout training process employs a global readout layer to aggregate the local compensation volume contribution values ​​into the airbag compensation volume output value. The global readout layer includes a non-negative linear neuron, which contains... Each non-negative connection weight and a non-negative bias is assigned a weight, with each non-negative connection weight corresponding to a hyperedge index. These weights represent the convergence relationship between the local compensation volume contribution values ​​corresponding to each hyperedge and the airbag compensation volume output value. The global readout layer multiplies each local compensation volume contribution value by its corresponding non-negative connection weight and adds it to the non-negative bias to obtain the result. The airbag compensation volume output value corresponding to each temperature drop range ,when Below At that time, and The difference is determined as the capacity shortage, when Higher than At that time, and The difference is determined as the capacity surplus, when equal At that time, both the capacity shortage and the capacity surplus are zero volume.

[0162] During the readout training phase, the total volume of the sub-cavities corresponding to the one-dimensional connected gas sub-cavities within each hyperedge feature is denoted as... And based on the airbag compensation volume output value Reference value of cold contraction negative pressure compensation volume Local compensation volumetric contribution value Total volume of the cavity Calculate the one-sided capacity shortage loss:

[0163] ;

[0164] In the formula, For the first The one-sided capacity shortage loss corresponding to each temperature drop interval Index for the temperature decrease range, The loss weight corresponds to the capacity shortage amount. The loss weight corresponds to the capacity surplus. This is a reference value for the volume of the cold contraction negative pressure compensation. For the first The airbag compensation volume output value corresponding to each temperature drop range. To obtain a larger value, the constants in the formula For zero volume, the constant in the formula This is a dimensionless reference value. The subcavity volume distribution constraint coefficient. It is a dimensionless value and is not less than zero. To achieve a summation, For superedge index, The number of superedges. For the first The total volume of the connected gas sub-cavities corresponding to each hyperedge. For the superedge index used for summation, To preset the volumetric stability constant, It is a positive volume quantity, used to avoid the denominator being zero. Treat it as an absolute value. For the first In the temperature drop interval, the first The local compensation volume contribution value corresponding to each hyperedge. , , , and All figures represent volume, and the ratios in the formulas are dimensionless values. One-sided capacity shortage loss... Its dimension is the square of volume;

[0165] During the readout training phase, the gradients of the non-negative connection weights, non-negative biases, contributing readout layer parameters, capacity-conserving hysteresis hypergraph propagation layer parameters, node input mapping layer parameters, and hyperedge input mapping layer parameters of the global readout layer are calculated based on the unilateral capacity shortage loss. The parameters are then updated in the direction of reducing the unilateral capacity shortage loss according to the preset learning rate. After each parameter update, non-negative constraints are applied to the connection weights, biases, and non-negative output neuron parameters of the global readout layer. Parameters less than zero are set to zero, and the loss weights corresponding to capacity shortage are greater than the loss weights corresponding to capacity surplus. When the output value of the airbag compensation volume is lower than the reference value of the cold shrinkage negative pressure compensation volume, a greater update effect is applied to the convergence relationship between the local compensation volume contribution value and the airbag compensation volume output value. After training, the non-negative connection weights and non-negative biases of the global readout layer form a capacity constraint convergence relationship.

[0166] Based on the capacity-constrained convergence relationship, the connected hypergraph features of each sub-cavity are analyzed. Re-execute node input mapping, hyperedge input mapping, two-layer capacity-conserving hysteresis hypergraph propagation, contribution readout, and global readout to obtain the corresponding airbag compensation volume discriminant value. When there are multiple temperature drop ranges, all of them will be The maximum value in is determined as the airbag compensation volume discrimination value. The airbag compensation volume discrimination value is a one-dimensional volume value, and the correspondence between the local compensation volume contribution value of each hyperedge, the sub-cavity connectivity relationship, the cold shrinkage negative pressure compensation volume reference value and the capacity constraint convergence relationship is maintained.

[0167] In this embodiment, step S6 includes:

[0168] The volume is effectively compensated by the airbag to be configured. and airbag compensation volume discriminant value As input, This refers to the one-dimensional volume that the airbag to be configured can provide to the inside of the battery pack housing during its effective stroke to counteract the negative pressure caused by cold contraction. This is the one-dimensional volume value obtained by aggregating the local compensation volume contribution values ​​according to the capacity constraint aggregation relationship. The constraints on the capacity shortage direction imposed by retaining the sub-cavity connectivity relationships, local compensation volume contribution values, and cold contraction negative pressure compensation volume reference values ​​in the gas sub-cavity connectivity hypergraph are as follows: and Before performing capacity comparison, Volume units and All volume units should be standardized to the same volume unit, and this should be confirmed. and All are non-negative volume values;

[0169] When comparing the difference in capacity between the effective compensation volume of the airbag to be configured and the airbag compensation volume discrimination value, As the volume to be reduced, As the volume value to be covered, The value minus The values ​​are used to obtain the capacity comparison difference. , This is a signed one-dimensional volume difference value, used to represent the capacity difference between the effective compensation volume of the airbag to be configured and the airbag compensation volume discrimination value. A positive volume value indicates that the effective compensation volume of the airbag to be configured is higher than the airbag compensation volume discrimination value. A zero volume value indicates that the effective compensation volume of the airbag to be configured is equal to the airbag compensation volume discrimination value. A negative volume value indicates that the effective compensation volume of the airbag to be configured is lower than the airbag compensation volume discrimination value;

[0170] When determining capacity coverage relationships based on capacity comparison differences, a capacity coverage relationship identifier is set. , It is a binary state identifier. A value of 1 indicates that the effective compensation volume of the airbag to be configured can cover the airbag compensation volume discrimination value. A value of zero indicates that the effective compensation volume of the airbag to be configured cannot cover the airbag compensation volume discrimination value. When the volume is greater than or equal to zero, Set to 1, and determine the capacity coverage relationship as follows: the effective compensation volume of the airbag to be configured is not less than the airbag compensation volume discrimination value. When the volume is less than zero, Set to zero, and determine the capacity coverage relationship as the effective compensation volume of the airbag to be configured is less than the airbag compensation volume discrimination value. The capacity coverage relationship uses the zero volume value as a fixed judgment boundary, and does not externally enlarge or reduce the airbag compensation volume discrimination value.

[0171] Capacity coverage relationship identification When the value is one, the generated airbag capacity meets the discrimination result. , Indicates the effective compensation volume of the airbag to be configured. Capable of covering airbag compensation volume discrimination value ,when When the volume is positive, the airbag capacity meets the discrimination result. Record As the remaining volume, when When the volume is zero, the airbag capacity meets the discrimination result. Record the zero volume value as the remaining volume; the airbag capacity meets the discrimination result. Relationship with capacity coverage correspond;

[0172] Capacity coverage relationship identification When the value is zero, an insufficient airbag capacity is generated as a judgment result. , Indicates the effective compensation volume of the airbag to be configured. The airbag compensation volume discriminant value cannot be covered. Insufficient airbag capacity determination results Will The absolute volume is recorded as insufficient volume. Insufficient volume corresponds to the portion of the airbag compensation volume judgment value that exceeds the effective compensation volume of the airbag to be configured. The result of insufficient airbag capacity judgment is as follows. Relationship with capacity coverage Correspondingly, the capacity difference Capacity coverage relationship identifier The airbag capacity meets the discrimination results. Or the result of insufficient airbag capacity. All by and It forms directly.

[0173] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A deep learning-based method for determining the airbag compensation volume of marine power battery packs, characterized in that, include: S1. Obtain marine power battery pack operation data, including the internal pressure sequence of the battery pack housing during the marine temperature cycle breathing process, the module temperature sequence, the spatial distribution of compressible gas inside the battery pack housing, and the effective compensation volume of the airbag to be configured. S2. Determine the reference value of the cold shrinkage negative pressure compensation volume based on the internal pressure sequence and module temperature sequence of the battery pack housing, and form the gas sub-cavity division result based on the spatial distribution of compressible gas inside the battery pack housing. S3. Construct a gas sub-cavity connected hypergraph based on the gas sub-cavity partitioning results. The nodes of the gas sub-cavity connected hypergraph include gas sub-cavities, temperature measuring points, and pressure measuring points. The hyperedges of the gas sub-cavity connected hypergraph are formed by the nodes corresponding to the connected gas sub-cavities. S4. Write the temperature change, pressure change, and sub-cavity volume into the nodes of the gas sub-cavity connected hypergraph, and write the sub-cavity connectivity relationship into the hyperedge of the gas sub-cavity connected hypergraph to obtain the sub-cavity connected hypergraph features. S5. Input the sub-cavity connected hypergraph features into the hypergraph neural network to generate local compensation volume contribution values. In the readout training stage where the local compensation volume contribution values ​​converge into the airbag compensation volume output value, embed the unilateral capacity shortage loss. The difference between the airbag compensation volume output value and the reference value of the cold shrinkage negative pressure compensation volume is determined as the capacity shortage amount. The difference between the airbag compensation volume output value and the reference value of the cold shrinkage negative pressure compensation volume is determined as the capacity surplus amount. According to the constraint that the loss corresponding to the capacity shortage amount is greater than the loss corresponding to the capacity surplus amount, the airbag compensation volume discrimination value is obtained. S6. Compare the effective compensation volume of the airbag to be configured with the airbag compensation volume discrimination value. When the effective compensation volume of the airbag to be configured is not less than the airbag compensation volume discrimination value, the output airbag capacity meets the discrimination result. When the effective compensation volume of the airbag to be configured is less than the airbag compensation volume discrimination value, the output airbag capacity is insufficient discrimination result.

2. The deep learning-based method for determining the airbag compensation volume of marine power battery packs according to claim 1, characterized in that, S1 includes: When the marine power battery pack is in the process of temperature cycling at sea, the pressure values ​​of the pressure measuring points inside the battery pack are collected and arranged according to the time of collection to form the time sequence of the internal pressure of the battery pack. During the same marine temperature cycle breathing process that forms the internal pressure sequence of the battery pack housing, the temperature values ​​of the module temperature measurement points are collected and arranged according to the collection time to form the module temperature sequence. Based on the structural boundaries, gas communication locations, and sub-cavity volume sources of the compressible gas space inside the battery pack housing, the distribution of the compressible gas space inside the battery pack housing is obtained. The effective compensation volume of the airbag to be configured is obtained based on the volume change of the airbag within its effective stroke. The internal pressure timing of the battery pack housing, the module temperature timing, the spatial distribution of compressible gas inside the battery pack housing, and the effective compensation volume of the airbag to be configured are combined to form the operating data of the marine power battery pack.

3. The deep learning-based method for determining the airbag compensation volume of marine power battery packs according to claim 1, characterized in that, S2 includes: The temperature drop process is extracted from the module temperature time series, and the internal pressure time series of the battery pack is matched with the temperature drop process to form a cold contraction negative pressure correspondence. Based on the correspondence between cold shrinkage and negative pressure, the volume compensation requirement corresponding to the pressure change inside the battery pack housing and the temperature change of the module is calculated, and the volume compensation requirement is determined as the reference value for cold shrinkage and negative pressure compensation volume. Based on the spatial distribution of compressible gas inside the battery pack housing, structural boundaries and gas connection locations are extracted to form sub-cavity boundary data; The gas sub-cavities are divided based on the sub-cavity boundary data, and the sub-cavity volume of each gas sub-cavity is determined to form the gas sub-cavity division result.

4. The deep learning-based method for determining the airbag compensation volume of a marine power battery pack according to claim 1, characterized in that, S3 include: Based on the gas sub-cavity division results, the spatial boundary, adjacent interface, ventilation opening and partition position of each gas sub-cavity are extracted to form sub-cavity boundary connectivity data; Based on the sub-cavity boundary connectivity data, the gas connectivity status between any gas sub-cavity and its adjacent gas sub-cavities is determined. Gas sub-cavities that are interconnected through ventilation openings and can jointly transmit pressure changes are identified as connected gas sub-cavities, forming connected gas sub-cavity groups used to constitute hyperedges. Based on the connected gas sub-cavity groups and the sub-cavity boundary connectivity data, the pressure transmission range corresponding to each connected gas sub-cavity group is determined, and the pressure transmission range is associated with the corresponding connected gas sub-cavity group to form connected pressure range data. Based on the connected pressure range data and the installation positions of temperature and pressure measuring points, the temperature and pressure measuring points are matched to their respective gas sub-cavities and connected gas sub-cavity groups to form measuring point affiliation relationships. Based on the gas... The sub-cavity partitioning results and the measurement point affiliation relationships are used to establish a node set. The nodes in the node set include gas sub-cavities, temperature measurement points, and pressure measurement points. Temperature measurement points and pressure measurement points are assigned to their corresponding gas sub-cavities. Based on the connected gas sub-cavity grouping, connected pressure range data, and node set, the nodes corresponding to the gas sub-cavities within the same connected gas sub-cavity group are organized into a hyperedge. Temperature measurement points and pressure measurement points belonging to the same pressure transmission range are associated with the hyperedge to form a hyperedge set. Based on the node set and hyperedge set, a gas sub-cavity connectivity hypergraph is constructed. Each hyperedge in the gas sub-cavity connectivity hypergraph is composed of nodes corresponding to the connected gas sub-cavities. The gas sub-cavity connectivity hypergraph retains the spatial boundaries and sub-cavity connectivity relationships in the gas sub-cavity partitioning results.

5. The deep learning-based method for determining the airbag compensation volume of marine power battery packs according to claim 1, characterized in that, S4 includes: The temperature measurement point is calculated from the module temperature time series to determine the temperature change direction between adjacent acquisition times, and the acquisition period with the temperature change direction decreasing is determined as the temperature decrease interval, thus forming the temperature decrease interval and the temperature change used for writing to the node. The pressure response range is determined based on the temperature drop range in the internal pressure timing of the battery pack housing, and the pressure difference between the pressure measurement points within the pressure response range is calculated to form the pressure change used for writing nodes. Based on the node composition of the gas sub-cavity connected hypergraph, the connection relationship between the temperature measuring point and the corresponding node of the gas sub-cavity, as well as the connection relationship between the pressure measuring point and the corresponding node of the gas sub-cavity, are determined to form the measurement point node correspondence relationship. Based on the correspondence between measurement point nodes, temperature changes are assigned to the nodes corresponding to the corresponding gas sub-cavities, and pressure changes are assigned to the nodes corresponding to the corresponding gas sub-cavities, forming node temperature changes and node pressure changes; based on the gas sub-cavity division results, the sub-cavity volume of each gas sub-cavity is determined, and the node temperature changes, node pressure changes, and sub-cavity volumes are written into the nodes of the gas sub-cavity connectivity hypergraph to form node compensation volume data. Based on the nodes corresponding to the connected gas sub-cavities contained in each hyperedge of the gas sub-cavity connectivity hypergraph, the sub-cavity connectivity relationships between gas sub-cavities within the same hyperedge are determined, forming hyperedge connectivity data; Write the hyperedge connectivity data into the hyperedge of the gas sub-cavity connectivity hypergraph, and update the gas sub-cavity connectivity hypergraph according to the node compensation volume data and the hyperedge connectivity data, so that temperature change, pressure change, sub-cavity volume and sub-cavity connectivity relationship correspond to the same connected gas sub-cavity, and obtain the sub-cavity connectivity hypergraph features.

6. The deep learning-based method for determining the airbag compensation volume of a marine power battery pack according to claim 1, characterized in that, S5 include: The sub-cavity connectivity hypergraph features are input into the hypergraph neural network. Based on the node and hyperedge connectivity relationships of the gas sub-cavity connectivity hypergraph, the temperature changes, pressure changes, and sub-cavity volumes of the written nodes are aggregated within the hyperedges to form a hyperedge compensation state. Based on the sub-cavity connectivity, the super-edge compensation state is written back to the node corresponding to the gas sub-cavity contained in the super-edge, and the temperature change, pressure change and sub-cavity volume within the node are updated to form the node compensation state. Based on the node compensation state and the hyperedges of the gas sub-cavity connectivity hypergraph, calculate the volume contribution of the connected gas sub-cavity corresponding to each hyperedge during the cooling and shrinking stage, and generate a local compensation volume contribution value. During the readout training phase, the local compensation volume contribution values ​​corresponding to each hyperedge are aggregated according to the sub-cavity connectivity to generate the airbag compensation volume output value. The airbag compensation volume output value is compared with the cold shrink negative pressure compensation volume reference value. When the airbag compensation volume output value is lower than the cold shrink negative pressure compensation volume reference value, the volume shortage is obtained. When the airbag compensation volume output value is higher than the cold shrink negative pressure compensation volume reference value, the volume surplus is obtained. During the readout training phase, the unilateral capacity shortage loss is calculated based on the capacity shortage and capacity surplus. In accordance with the constraint that the loss corresponding to the capacity shortage is greater than the loss corresponding to the capacity surplus, the convergence relationship between the local compensation volume contribution value and the airbag compensation volume output value is updated to form a capacity constraint convergence relationship. Based on the capacity constraint convergence relationship, the local compensation volume contribution value is converged to obtain the airbag compensation volume discrimination value.

7. The deep learning-based method for determining the airbag compensation volume of a marine power battery pack according to claim 1, characterized in that, S6 include: Receive the effective compensation volume of the airbag to be configured and the airbag compensation volume discrimination value, and form a capacity comparison difference based on the volume difference between the effective compensation volume of the airbag to be configured and the airbag compensation volume discrimination value; Based on the capacity comparison difference, it is determined whether the effective compensation volume of the airbag to be configured can cover the airbag compensation volume discrimination value, thus forming a capacity coverage relationship; When the capacity coverage relationship indicates that the effective compensation volume of the airbag to be configured is not less than the airbag compensation volume discrimination value, the output airbag capacity satisfies the discrimination result. When the capacity coverage relationship indicates that the effective compensation volume of the airbag to be configured is less than the airbag compensation volume discrimination value, the output airbag capacity insufficient discrimination result is obtained.

8. The deep learning-based method for determining the airbag compensation volume of a marine power battery pack according to claim 4, characterized in that, The formation of connected gas sub-cavities grouping and hyperedge sets includes: reading the ventilation opening and isolation positions between any gas sub-cavity and its adjacent gas sub-cavities based on the sub-cavity boundary connectivity data; when there is a ventilation opening between any gas sub-cavity and its adjacent gas sub-cavities and there is no isolation position that blocks pressure changes, any gas sub-cavity and its adjacent gas sub-cavities are assigned to the same connected gas sub-cavity group; when there is no ventilation opening between any gas sub-cavities and its adjacent gas sub-cavities, or there is an isolation position that blocks pressure changes, any gas sub-cavities and their adjacent gas sub-cavities are not assigned to the same connected gas sub-cavities group; and based on the pressure transmission range corresponding to the same connected gas sub-cavities grouping, organizing the nodes corresponding to the gas sub-cavities within the same connected gas sub-cavities grouping and belonging to the same pressure transmission range into the same hyperedge, forming a hyperedge set.

9. The deep learning-based method for determining the airbag compensation volume of a marine power battery pack according to claim 5, characterized in that, The formation of node compensation volume data includes: writing temperature changes into the node corresponding to the gas sub-cavity of the temperature measuring point according to the correspondence between the temperature drop range and the measuring point node; writing pressure changes into the node corresponding to the gas sub-cavity of the pressure measuring point according to the correspondence between the pressure response range and the measuring point node; when the temperature measuring point corresponds to multiple gas sub-cavities, distributing the temperature changes to the nodes corresponding to the multiple gas sub-cavities according to the correspondence between the measuring point node; when the pressure measuring point corresponds to multiple gas sub-cavities, distributing the pressure changes to the nodes corresponding to the multiple gas sub-cavities according to the correspondence between the measuring point node; and writing the sub-cavity volume of each gas sub-cavity into the node corresponding to the same gas sub-cavity, so that the node corresponding to the same gas sub-cavity forms node compensation volume data containing node temperature changes, node pressure changes, and sub-cavity volumes.

10. The deep learning-based method for determining the airbag compensation volume of a marine power battery pack according to claim 6, characterized in that, The formation of the capacity constraint convergence relationship includes: when the airbag compensation volume output value is lower than the cold shrink negative pressure compensation volume reference value, updating the convergence relationship of the local compensation volume contribution value based on the capacity shortage to the airbag compensation volume output value; when the airbag compensation volume output value is higher than the cold shrink negative pressure compensation volume reference value, updating the convergence relationship based on the capacity surplus; setting the update effect of the loss corresponding to the capacity shortage on the convergence relationship to be greater than the update effect of the loss corresponding to the capacity surplus on the convergence relationship; determining the updated convergence relationship as the capacity constraint convergence relationship; and converging the local compensation volume contribution value based on the capacity constraint convergence relationship to obtain the airbag compensation volume discrimination value.

Images