Battery pack life prediction method based on multi-sensing parameter fusion

Abstract

The invention discloses a battery pack life prediction method based on multi-sensing parameter fusion, and relates to the technical field of battery management. According to the method, a start-stop frequency density field fatigue pre-evaluation index Fpre serves as fatigue feature input and also serves as a core quantity of trigger control and life recalculation; a start-stop frequency density field fatigue pre-evaluation index Fpre, a reference fatigue level Fref, a historical accumulated fatigue damage degree Dold, a residual life prediction result RULnew and a life threshold RULth are connected in series to form a complete process of multi-sensor acquisition, fatigue evaluation, control triggering, damage accumulation and life early warning, so that high-fatigue start-stop working conditions can be identified in real time, and meanwhile, immediate intervention can be realized; and the control effect can be reflected to a residual life result, so that a traceable and adjustable full-life-cycle life management mechanism is formed, and the operation safety and reliability of the port automatic container tractor battery pack are improved.

Description

Technical Field

[0001] This invention relates to the field of battery management technology, specifically to a method for predicting battery pack lifespan by fusing multiple sensor parameters. Background Technology

[0002] With the rapid development of port automation operations, new energy transportation equipment, and smart grid technology, the importance of battery management and health status prediction technology in large-scale electrified port vehicle clusters is becoming increasingly prominent. In particular, in the assessment of the lifespan of power battery packs and the analysis of cycle fatigue under high-frequency start-stop conditions, more refined and scenario-based technical means are needed. This invention relates to the field of battery management and health status prediction technology, specifically to the problem of battery pack lifespan prediction for automated container tractors and other vehicles operating under conditions of frequent start-stop and short-distance heavy-load operation in closed port areas.

[0003] Currently, in the battery packs used in automated container tractors at ports, existing life prediction and health assessment methods generally rely on steady-state or statistical parameters such as the number of charge-discharge cycles, depth of discharge (DOD), state of charge (SOC), and average operating temperature as the main criteria. Some solutions only infer life degradation trends through periodic capacity and internal resistance tests under offline experimental conditions. These methods have significant shortcomings in application scenarios such as port operations with high-frequency start-stop and drastic load fluctuations: On the one hand, traditional models mainly focus on energy flow and SOC changes over long timescales, lacking the ability to capture short-term transient impact characteristics such as the steepness of current rise and temperature rise rate during vehicle startup, making it difficult to reflect the high-frequency cyclic fatigue caused by frequent start-stop operations on battery electrodes, electrolyte interfaces, and current collectors; on the other hand, existing technologies typically classify operating conditions simply into coarse-grained categories such as "heavy load / light load" and "high temperature / normal temperature," failing to provide a detailed characterization of the density, sequence, and transient coupling effects of start-stop events within a time window, resulting in insufficient sensitivity of life prediction results to actual port operating conditions.

[0004] Under the aforementioned technological conditions, when automated container tractors operate under conditions of high start-stop frequency, short intervals, heavy-load acceleration, and frequent braking, the metal negative electrode, electrode-current collector interface, and solder joints inside the battery pack will experience accelerated cyclic fatigue under the influence of high current rise steepness and rapid temperature fluctuations. This manifests as abnormal dendrite growth in the negative electrode, shedding of active materials, and microcracks and contact degradation between the current collector and the electrode. However, existing life assessment methods do not construct quantifiable fatigue characteristics from transient information such as start-stop event density, current rise steepness, and temperature rise rate. They rely solely on DOD-SOC operating condition classification and average temperature evaluation. Often, life degradation issues are only passively discovered after abnormal phenomena such as sudden increases in battery capacity decay, abnormal increases in internal resistance, aggravated fluctuations in single-cell voltage, or even power drops, forced speed reductions, or shutdowns of tractors during peak operating periods occur. This can easily lead to disruptions in port operation plans, reduced equipment utilization, and potential safety risks. There is an urgent need for a life prediction method that can use the start-stop event density field to map multi-sensor transient parameters such as start-up instantaneous messages, current rise steepness, and temperature rise rate to accurately deduce the cyclic fatigue accumulation caused by short-term repetitive loads, thereby more realistically reflecting the actual loss sources of battery packs under port operations. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a battery pack life prediction method based on the fusion of multiple sensing parameters, thus solving the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a battery pack life prediction method based on multi-sensor parameter fusion, comprising the following steps: S1. Several collection points are set up in the battery pack of the automated container tractor in the port to collect transient sequence sets in real time and transmit the transient sequence sets to the cloud prediction server. The cloud prediction server performs start-stop window division and window data extraction on the transient sequence sets to obtain the start-stop frequency transient dataset. S2. Based on the start-stop frequency transient dataset, calculate the start-stop frequency density field fatigue pre-evaluation index Fpre, and then set the benchmark fatigue level Fref and the start-stop frequency density field fatigue pre-evaluation index Fpre for preliminary comparison and evaluation. S3. Based on the preliminary comparison and evaluation results, generate a trigger flag TF, and then execute the start-stop control strategy and thermal management control strategy according to the trigger flag TF generation results; S4. After executing the start-stop control strategy and thermal management control strategy, update the historical cumulative fatigue damage degree Dold based on the fatigue pre-evaluation index Fpre of the start-stop frequency density field, calculate the battery pack remaining life prediction result RULnew, then compare the preset life threshold RULth with the battery pack remaining life prediction result RULnew, and generate a response strategy.

[0007] Preferably, S1 includes S11; S11. By setting up several collection points in the battery pack of the automated container tractor in the port, and collecting the transient sequence set of the battery pack in real time through the vehicle battery management system and vehicle controller during the operation of the battery pack. The transient sequence set is uploaded to the cloud prediction server via the vehicle communication gateway in the form of Hypertext Transfer Protocol messages. The data acquisition points include current acquisition points, temperature acquisition points, and start / stop event marker acquisition points; The transient sequence set includes current transient sequences, temperature transient sequences, and start / stop event marker sequences.

[0008] Preferably, S1 further includes S12; S12. In the cloud prediction server, the transient sequence set of each port automated container tractor battery pack is parsed and cached. The current transient sequence, temperature transient sequence and start-stop event marker sequence in the transient sequence set are reordered in chronological order according to a unified time axis. The sorted transient sequence set is then stored in the order of the port automated container tractor battery pack number as an index. Then, based on the vehicle start event and braking end event marked by the start-stop event marker sequence, the unified time axis is divided into several start-stop windows, each start-stop window corresponding to a time interval [t0,t1] between a set of vehicle start events and braking end events.

[0009] Preferably, S1 further includes S13; S13. For each start-stop window, extract the current transient sequence, temperature transient sequence, and start-stop event marker sequence that fall within the current start-stop window time interval from the transient sequence set of each port automated container tractor battery pack stored in the index, and form the start-stop window current subsequence, start-stop window temperature subsequence, and start-stop window voltage subsequence. The start-stop frequency is statistically analyzed based on the start-stop event marker sequence. The start-stop frequency transient dataset is composed of the start-stop window current subsequence, start-stop window temperature subsequence, start-stop window voltage subsequence and the corresponding start-stop frequency. The start-stop frequency transient dataset includes the current value I and the temperature value T within the start-stop window time interval [t0, t1]. Then, the physical quantities in the start-stop frequency transient dataset are numerically transformed according to the preset maximum value normalization rule to eliminate the differences in units and dimensions of all parameters in the start-stop frequency transient dataset.

[0010] Preferably, S2 includes S21; S21. The cloud prediction server uses numerical differentiation to obtain the current change rate dI(t) / dt and temperature change rate dT(t) / dt at each sampling time based on the start-stop frequency transient dataset. Then, the cloud prediction server performs integral calculations based on the current change rate reference value RI and temperature change rate reference value RT, which are calibrated in advance under standard smooth acceleration and normal heat dissipation conditions, to obtain the fatigue pre-evaluation index Fpre of the start-stop frequency density field, and quantitatively analyzes the fatigue strength caused by the combined current transient impact and temperature transient impact within the stop window. The fatigue pre-evaluation index Fpre based on the start-stop frequency density field is calculated and output using the following algorithm formula: ; In the formula, d represents the integral function, dt represents the time integral function, I(t) represents the current value at time t within the start-stop window time interval [t0,t1], and T(t) represents the temperature value at time t within the start-stop window time interval [t0,t1].

[0011] Preferably, S2 further includes S22; S22. During the initial safe operation of the battery pack, the fatigue pre-evaluation index Fpre of the start-stop frequency density field of multiple start-stop windows is statistically analyzed. The average fatigue pre-evaluation index Fpre of the start-stop frequency density field of the start-stop windows that are in normal working conditions and have not experienced abnormal decay is calculated to obtain the benchmark fatigue level Fref. During the online operation phase, the cloud-based prediction server calculates the ratio of the current start-stop frequency density field fatigue pre-assessment index Fpre to the baseline fatigue level Fref for each start-stop window, obtaining the fatigue relative coefficient RF. Based on the output of the fatigue relative coefficient RF, a preliminary comparative assessment is performed to determine the fatigue state of the current start-stop window. The specific assessment content is as follows: When the fatigue relative coefficient RF > 1, it indicates that the fatigue intensity of the current start-stop window exceeds the benchmark, and the current start-stop window is marked as a fatigue state. When the relative fatigue coefficient RF≤1, it means that the fatigue intensity of the current start-stop window has not exceeded the benchmark, and the current start-stop window is not marked.

[0012] Preferably, S3 includes S31; S31. If the preliminary comparative evaluation results indicate a fatigue state and there are more than two consecutive start-stop cycles, then the current trigger flag TF=1, and the start-stop control strategy and thermal management control strategy will be triggered. Conversely, if the current trigger flag TF is set to 0, the control strategy cannot be triggered, and only data is recorded; The start-stop control strategy transmits the fatigue relative coefficient RF to the traction motor controller and the vehicle dispatching system. The traction motor controller uses the current change rate reference value RI as the adjustment target for the current change rate during the start-up phase, as detailed below: When the relative fatigue coefficient RF is between 1.20 and 1.50, the current ramp-up time during the motor start-up phase is increased by 10%-30% from the original setting, and the actual current change rate dI(t) / dt is limited to the range of 0.80×RI to 1.00×RI by adjusting the ramp-up curve of the PWM duty cycle. When the relative fatigue coefficient RF≥1.50, the current ramp-up time during the motor start-up phase is increased by 30%-60% from the original setting, and the actual current change rate dI(t) / dt is limited to the range of 0.50×RI to 0.80×RI by limiting the current rise slope. The vehicle dispatching system adjusts the lower limit of the continuous start-stop event interval based on the fatigue relative coefficient RF: when the fatigue relative coefficient RF is between 1.20 and 1.50, the minimum start-stop time interval Δtmin is increased by 20%-50% from the original setting; When the relative fatigue coefficient RF ≥ 1.50, the minimum start-stop time interval Δtmin is increased by 50%-100% based on the original setting.

[0013] Preferably, S3 further includes S32; S32. The thermal management control strategy sends the fatigue relative coefficient RF to the thermal management controller via a cloud prediction server. The thermal management controller adjusts the operating parameters of the cooling device in stages based on the transient dataset of start-stop frequencies. The specific adjustments are as follows: When the start-stop frequency in the start-stop window is ≥25 times and the current average temperature of the battery pack is higher than 30℃, the thermal management controller executes pre-cooling control, increasing the cooling fan and cooling plate from the basic operating condition to 120%-150% of the basic operating power; Within the start / stop window, when the temperature change rate dT(t) / dt of any battery pack is detected to be greater than the temperature rise rate reference value RT, the thermal management controller performs cooling optimization control on the local cooling unit corresponding to the current battery pack, increasing the fan speed and cooling power of the current local cooling unit by 30%-60% based on the basic operating conditions, until the difference between the battery pack temperature and the average battery pack temperature drops to within 3℃, and the temperature change rate dT(t) / dt falls back to the range of 0.60×RT to 0.90×RT.

[0014] Preferably, S4 includes S41; S41. After the start-stop control strategy and thermal management control strategy are executed, the fatigue pre-evaluation index Fpre of the start-stop frequency density field is recalculated and generated in the next start-stop window. The cloud prediction server accumulates the fatigue damage of the historical start-stop evaluation window to obtain the historical cumulative fatigue damage degree Dold, which is a dimensionless quantity between 0 and 1. The cloud-based prediction server uses the design life Lf as the benchmark for calculating the remaining life, and calculates a new remaining life prediction value RULnew by combining it with the historical cumulative fatigue damage degree Dold and the fatigue pre-evaluation index Fpre of the start-stop frequency density field. The new remaining lifetime prediction value RULnew is calculated and output using the following algorithm formula: .

[0015] Preferably, S4 further includes S42; S42. Set a preset lifespan threshold RULth based on 20% of the design lifespan Lf. Then compare the new remaining lifespan prediction value RULnew with the lifespan threshold RULth to determine the current battery pack lifespan safety status. The specific comparison is as follows: When the new remaining life prediction value RULnew ≤ life threshold RULth, it indicates that the current battery pack's remaining life is abnormal. At this time, a life warning is issued to the operation and maintenance system, and the current battery pack is subjected to in-depth inspection and replacement. When the new remaining life prediction value RULnew > life threshold RULth, it indicates that the current start-stop control strategy and thermal management control strategy are experiencing accelerated fatigue, but the battery pack is still in an acceptable lifespan state, and monitoring should continue at this time.

[0016] This invention provides a method for predicting battery pack lifespan by fusing multiple sensor parameters. It has the following beneficial effects: (1) This method constructs a start-stop frequency transient dataset and calculates the start-stop frequency density field fatigue pre-evaluation index Fpre. It transforms the current transient sequence, current change rate and temperature change rate within the start-stop window into a single fatigue strength index. Compared with the traditional method based only on DOD, SOC and average temperature, it can directly reflect the high-frequency fatigue effect of the current rise steepness and temperature rise rate at the moment of start-up on the battery metal anode and electrode-current collector interface. It can more accurately depict the real loss source of the port automated container tractor under frequent start-stop conditions and improve the discrimination and physical reliability of the life prediction input features.

[0017] (2) This method sets a baseline fatigue level Fref and calculates the fatigue relative coefficient RF. Combined with the trigger flag TF of continuous fatigue window triggering, it automatically triggers start-stop control strategy and thermal management control strategy. When the start-stop frequency and fatigue intensity are too high, it dynamically adjusts the start-up current ramp-up time, limits the current change rate dI(t) / dt, increases the minimum start-stop time interval, and upgrades the power of cooling fan and cooling plate in stages. This significantly reduces the current impact and temperature impact in the subsequent start-stop window, thereby slowing down the growth of negative electrode dendrites, thickening of interface film and accumulation of current collector fatigue from the source, and achieving active fatigue suppression without changing the hardware.

[0018] (3) This method uses the fatigue pre-evaluation index Fpre of the start-stop frequency density field as both a fatigue characteristic input and a core quantity for triggering control and life recalculation. It connects the fatigue pre-evaluation index Fpre of the start-stop frequency density field, the baseline fatigue level Fref, the historical cumulative fatigue damage degree Dold, the remaining life prediction result RULnew, and the life threshold RULth into a complete process of "multi-sensor acquisition, fatigue assessment, control triggering, damage accumulation and life warning". It can not only identify high fatigue start-stop conditions in real time and intervene immediately, but also reflect the control effect in the remaining life result, forming a traceable and adjustable full life cycle life management mechanism, and improving the safety and reliability of the battery pack operation of the automated container tractor in the port. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the steps in the battery pack lifetime prediction method based on multi-sensor parameter fusion of the present invention. Figure 2 This is a schematic diagram showing the location of the battery pack data collection points for the automated container tractor in the port. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example 1

[0021] This invention provides a battery pack lifetime prediction method based on multi-sensor parameter fusion. Please refer to [link / reference]. Figure 1 This includes the following steps: S1. Several collection points are set up in the battery pack of the automated container tractor in the port to collect transient sequence sets in real time and transmit the transient sequence sets to the cloud prediction server. The cloud prediction server performs start-stop window division and window data extraction on the transient sequence sets to obtain the start-stop frequency transient dataset. S2. Based on the start-stop frequency transient dataset, calculate the start-stop frequency density field fatigue pre-evaluation index Fpre, and then set the benchmark fatigue level Fref and the start-stop frequency density field fatigue pre-evaluation index Fpre for preliminary comparison and evaluation. S3. Based on the preliminary comparison and evaluation results, generate a trigger flag TF, and then execute the start-stop control strategy and thermal management control strategy according to the trigger flag TF generation results; S4. After executing the start-stop control strategy and thermal management control strategy, update the historical cumulative fatigue damage degree Dold based on the fatigue pre-evaluation index Fpre of the start-stop frequency density field, calculate the battery pack remaining life prediction result RULnew, then compare the preset life threshold RULth with the battery pack remaining life prediction result RULnew, and generate a response strategy.

[0022] In this embodiment, the method involves deploying current acquisition points, temperature acquisition points, and start-stop event marker points within the battery pack of the automated container tractor in port S1. Real-time acquisition of current transient sequences, temperature transient sequences, and start-stop event marker sequences is achieved on the vehicle side and uploaded to a cloud prediction server. In the cloud, a unified timeline is divided into start-stop windows based on start-stop events. Corresponding current and temperature subsequences are extracted from each window to construct a start-stop frequency transient dataset that corresponds one-to-one with each start-stop action. This avoids the problem of traditional methods that only consider average daily battery capacity and total cycle count without identifying "which start-stop events are abnormally damaging the battery." In S2, the cloud prediction server uses the start-stop frequency transient dataset... The fatigue pre-assessment index Fpre for the start-stop frequency density field is calculated. The current rise steepness and temperature rise rate within each start-stop window are compressed into a single fatigue strength index through an integral formula, and compared with the benchmark fatigue level Fref to obtain the fatigue relative coefficient. This allows for a clear numerical determination of whether a start-stop window belongs to an "exceeding fatigue condition," solving the problem that traditional DOD-SOC grading cannot reflect short-term impact fatigue. In S3, when the fatigue relative coefficient of multiple consecutive start-stop windows exceeds a set threshold, a trigger flag TF=1 is generated, automatically triggering the start-stop control strategy and thermal management control strategy. On the one hand, this extends the motor starting current ramp-up time... Increasing the minimum start-stop interval reduces the current change rate. On the other hand, pre-cooling and local enhanced cooling suppress the temperature change rate, thereby actively "reducing the load and cooling the temperature" of the electrode, electrolyte interface, and current collector under high start-stop frequency and high impact conditions, suppressing the trend of further increases in Fpre. In S4, the fatigue pre-evaluation index Fpre of the start-stop frequency density field is recalculated in the new start-stop window after the control strategy is executed, and the Fpre / Fref corresponding to each start-stop window is accumulated to form the historical cumulative fatigue damage degree Dold. Then, using Dold and the Fpre / Fref of the current window as input, combined with the design life Lf, a new remaining life prediction result is calculated. RULnew is compared with the preset lifespan threshold RULth. When RULnew is lower than RULth, strategies such as reduced load operation, limited start-stop frequency, and early maintenance or replacement are triggered. When RULnew is higher than RULth, the current control conditions are maintained or moderately relaxed. This achieves closed-loop management from "multi-sensor data acquisition, quantitative identification of start-stop fatigue, active flow and temperature control, quantitative assessment of remaining lifespan, and maintenance decisions based on lifespan thresholds." This not only suppresses the hidden cyclic fatigue caused by frequent start-stops at the physical level, but also transforms abstract lifespan risks into executable control and maintenance actions at the decision-making level, improving battery pack lifespan utilization and the safety and reliability of port operations. Example 2

[0023] Please see Figure 1 and Figure 2 Specifically: S1 includes S11; S11. By setting up several collection points in the battery pack of the automated container tractor in the port, and collecting the transient sequence set of the battery pack in real time through the vehicle battery management system and vehicle controller during the operation of the battery pack. The transient sequence set is uploaded to the cloud prediction server via the vehicle communication gateway in the form of Hypertext Transfer Protocol messages. The data acquisition points include current acquisition points, temperature acquisition points, and start / stop event marker acquisition points; The transient sequence set includes current transient sequences, temperature transient sequences, and start / stop event marker sequences; The current acquisition point is set at the main circuit bus in the battery pack, and the current transient sequence is obtained through the current sensor of the vehicle battery management system. Temperature acquisition points are set in high heat load areas of each battery pack, such as near the tabs or in cooling blind areas, and the transient temperature sequence is acquired by the temperature acquisition sensor of the vehicle battery management system. The start-stop event marker collection point is set on the on-board controller of the automated container tractor in the port. When the motor is started and braking ends, the start-stop flag message is written to the BMS to obtain the start-stop event marker sequence.

[0024] S1 also includes S12; S12. In the cloud prediction server, the transient sequence set of each port automated container tractor battery pack is parsed and cached. The current transient sequence, temperature transient sequence and start-stop event marker sequence in the transient sequence set are reordered in chronological order according to a unified time axis. The sorted transient sequence set is then stored in the order of the port automated container tractor battery pack number as an index. Then, based on the vehicle start event and braking end event marked by the start-stop event marker sequence, the unified time axis is divided into several start-stop windows, each start-stop window corresponding to a time interval [t0,t1] between a set of vehicle start events and braking end events.

[0025] S1 also includes S13; S13. For each start-stop window, extract the current transient sequence, temperature transient sequence, and start-stop event marker sequence that fall within the current start-stop window time interval from the transient sequence set of each port automated container tractor battery pack stored in the index, and form the start-stop window current subsequence, start-stop window temperature subsequence, and start-stop window voltage subsequence. The start-stop frequency is statistically analyzed based on the start-stop event marker sequence. The start-stop frequency transient dataset is composed of the start-stop window current subsequence, start-stop window temperature subsequence, start-stop window voltage subsequence and the corresponding start-stop frequency. The start-stop frequency transient dataset includes the current value I and the temperature value T within the start-stop window time interval [t0, t1]. Then, the physical quantities in the start-stop frequency transient dataset are numerically transformed according to the preset maximum value normalization rule to eliminate the differences in units and dimensions of all parameters in the start-stop frequency transient dataset.

[0026] In this embodiment, method S11 sets up current acquisition points, temperature acquisition points, and start-stop event marker acquisition points within the battery pack of the automated container tractor in the port. It then relies on the on-board battery management system and on-board controller to acquire the current transient sequence, temperature transient sequence, and start-stop event marker sequence in real time and uploads them to the cloud prediction server. This transforms the originally scattered monitoring data, used only for protection alarms, into raw transient data sources that can be directly utilized for subsequent fatigue modeling. S12, on the cloud prediction server side, rearranges the aforementioned transient sequence set according to a unified time axis and precisely divides the time axis into start-stop windows [t0, t1] based on the start-stop event markers. This ensures that each "start and stop" operation is performed correctly. The process involves clear and continuous data slices to avoid cross-interference between data from different vehicles and at different times, accurately anchoring the start-stop behavior from a time dimension. S13 then extracts current, temperature, and start-stop event subsequences within each start-stop window, counts the start-stop frequency, and constructs a transient dataset of start-stop frequency. At the same time, physical quantities such as current value I and temperature value T are normalized according to a preset maximum value rule. This is to unify physical quantities such as "current magnitude, temperature change, and number of start-stops" into a comparable dimensionless space within the same window, avoiding bias caused by different units and magnitudes, and providing clean and structured data input for the subsequent construction of the start-stop frequency density field fatigue pre-evaluation index Fpre. This data collection, alignment, windowing, and normalization process, from onboard to cloud, essentially refines coarse-grained information like "how many trips a vehicle made today" into a quantitative description of "how much current and temperature surged and accelerated during each start-stop process." This significantly improves the observability and modelability of real loss sources under frequent start-stop conditions and provides a more reliable data foundation than the traditional DOD-SOC method for subsequent fatigue assessment, control strategy triggering, and life prediction. Example 3

[0027] Please see Figure 1 Specifically: S2 includes S21; S21. The cloud prediction server uses numerical differentiation to obtain the current change rate dI(t) / dt and temperature change rate dT(t) / dt at each sampling time based on the start-stop frequency transient dataset. Then, the cloud prediction server performs integral calculations based on the current change rate reference value RI and temperature change rate reference value RT, which are calibrated in advance under standard smooth acceleration and normal heat dissipation conditions, to obtain the fatigue pre-evaluation index Fpre of the start-stop frequency density field, and quantitatively analyzes the fatigue strength caused by the combined current transient impact and temperature transient impact within the stop window. The fatigue pre-evaluation index Fpre for start-stop frequency density field is calculated and output using the following algorithm formula: ; In the formula, d represents the integral function, dt represents the time integral function, I(t) represents the current value at time t within the start-stop window time interval [t0,t1], and T(t) represents the temperature value at time t within the start-stop window time interval [t0,t1]. This formula provides a unified quantitative characterization of the electrical and thermal shocks experienced by a battery during start-up and shutdown. It selects the current change rate dI(t) / dt and the temperature change rate dT(t) / dt as fundamental quantities because the current change rate dI / dt reflects the magnitude of changes in electromagnetic stress and electrochemical reaction rates experienced by the battery during start-up and shutdown, and is directly related to the current density impact on the electrodes. The temperature change rate dT / dt reflects the rate of change in thermal stress caused by uneven heating and cooling within the battery's internal structure, and is related to the expansion and contraction of electrode materials and changes in interfacial stress. By using only the absolute values ​​of current I(t) and temperature T(t), the rate of change is more sensitive to capturing the characteristics of "impact damage." Mathematically, the normalized rates of change of current and temperature are squared, integrated, and then square-rooted, respectively, which is a typical L2 norm structure. The squaring operation is used to unify positive and negative changes into an "intensity" quantity, avoiding cancellation due to different signs, and corresponding to the energy-type measurement form. The integration operation is used to accumulate this intensity over the entire start-stop evaluation window, reflecting "the cumulative impact amount over this period" in the time dimension. The square root operation restores the integral result to the "strength order of magnitude," making F_pre a unified fatigue strength index that facilitates comparison across different time windows, vehicles, or operating conditions. Regarding the coupling of multiple physical quantities, the normalized current rate of change component and the normalized temperature rate of change component are added in square root form as a sum of squares. This can be viewed as treating electrical shock fatigue and thermal shock fatigue as two orthogonal components in a two-dimensional space, respectively. The magnitude of the composite vector is then calculated using a method similar to the Pythagorean theorem, thus achieving [the desired result] without relying on subjective weight settings. The symmetry and interpretability of electrical and thermal stresses are integrated. Based on the above design, the larger the value of the fatigue pre-evaluation index Fpre of the start-stop frequency density field, the faster the current rises, the higher the peak value, the more rapid the temperature rise, and the steeper the thermal gradient within the corresponding start-stop evaluation window. The start-stop process is more "intense". In long-term operation, this type of high start-stop frequency density field fatigue pre-evaluation index Fpre window will significantly accelerate the initiation and growth of negative electrode dendrites, the thickening of electrolyte interface film, and the cyclic fatigue of current collectors and solder joints. Macroscopically, this manifests as accelerated battery capacity decay and intensified internal resistance increase.

[0028] S2 also includes S22; S22. In the early stage of safe operation of the battery pack, the fatigue pre-evaluation index Fpre of the start-stop frequency density field of multiple start-stop windows is statistically analyzed. The fatigue pre-evaluation index Fpre of the start-stop frequency density field of the start-stop windows under normal operating conditions and without abnormal decay is averaged to obtain the benchmark fatigue level Fref, so that the benchmark fatigue level Fref reflects the standard start-stop fatigue level of the battery pack within the design allowable range. During the online operation phase, the cloud-based prediction server calculates the ratio of the current start-stop frequency density field fatigue pre-assessment index Fpre to the baseline fatigue level Fref for each start-stop window, obtaining the fatigue relative coefficient RF. Based on the output of the fatigue relative coefficient RF, a preliminary comparative assessment is performed to determine the fatigue state of the current start-stop window. The specific assessment content is as follows: When the fatigue relative coefficient RF > 1, it indicates that the fatigue intensity of the current start-stop window exceeds the benchmark, and the current start-stop window is marked as a fatigue state. When the relative fatigue coefficient RF≤1, it means that the fatigue intensity of the current start-stop window has not exceeded the benchmark, and the current start-stop window is not marked.

[0029] In this embodiment, step S21 of the method performs numerical differentiation on the transient dataset of start-stop frequency, first converting the current I(t) and temperature T(t) within the start-stop window into the current change rate dI(t) / dt and the temperature change rate dT(t) / dt, and then combining the reference values ​​RI and RT calibrated under standard smooth acceleration conditions and normal heat dissipation conditions for normal integration, to obtain the fatigue pre-evaluation index Fpre of the start-stop frequency density field. Essentially, it compresses the two types of transient impacts, "how fast the current surges and how rapidly the temperature rises," into a measurable fatigue strength quantity, rather than just looking at the peak current or average temperature. Subsequently, in step S22, in the early stage of safe operation of the battery pack, a batch of normal start-stop windows without abnormal decay are selected, and the average of their Fpre is taken to obtain the benchmark fatigue level Fref. Then, in the online stage, Fpre / Fref is calculated for each start-stop window to obtain the fatigue relative coefficient RF. When RF>1, the window is marked as fatigued, and RF≤1 is considered normal. This is equivalent to setting a "normal upper limit for start-stop fatigue" for the battery. The reason for this design is that in port scenarios, drivers often experience prolonged periods of heavy-load, short-distance reversing and frequent intermittent braking. On the surface, the total DOD and SOC changes are not significant, but the current slope and temperature rise rate during start-stop are much higher than usual. If only traditional cycle counting and average temperature are used, it is easy to "miss" these high-impact conditions, resulting in serious dendrite growth, interface film thickening, and current collector fatigue, while the model still shows "healthy lifespan". By using Fpre to quantify transient fatigue and using Fref and RF to provide a clear criterion for "whether it exceeds the standard", these hidden high fatigue windows can be directly identified numerically. This provides an accurate and reliable threshold for subsequent triggering control strategies and lifespan recalculation, thus effectively avoiding the distortion of "seemingly low cycle count but sudden rapid decay". In fact, it improves the sensitivity and reliability of lifespan assessment to the real start-stop loss sources. Example 4

[0030] Please see Figure 1 Specifically: S3 includes S31; S31. If the preliminary comparative evaluation results indicate a fatigue state and there are more than two consecutive start-stop cycles, then the current trigger flag TF=1, and the start-stop control strategy and thermal management control strategy will be triggered. Conversely, if the current trigger flag TF is set to 0, the control strategy cannot be triggered, and only data is recorded; The start-stop control strategy sends the fatigue relative coefficient RF to the traction motor controller and the vehicle dispatching system. The traction motor controller uses the current change rate reference value RI as the adjustment target for the current change rate during the startup phase, as detailed below: When the relative fatigue coefficient RF is between 1.20 and 1.50, the current ramp-up time during the motor start-up phase is increased by 10%-30% from the original setting, and the actual current change rate dI(t) / dt is limited to the range of 0.80×RI to 1.00×RI by adjusting the ramp-up curve of the PWM duty cycle. When the relative fatigue coefficient RF≥1.50, the current ramp-up time during the motor start-up phase is increased by 30%-60% from the original setting, and the actual current change rate dI(t) / dt is limited to the range of 0.50×RI to 0.80×RI by limiting the current rise slope, thereby reducing the intensity of transient current impact during start-up and shutdown. The vehicle dispatching system adjusts the lower limit of the interval between consecutive start-stop events based on the relative fatigue coefficient RF: when the relative fatigue coefficient RF is between 1.20 and 1.50, the minimum start-stop interval Δtmin is increased by 20%-50% from the original setting; When the fatigue relative coefficient RF≥1.50, the minimum start-stop time interval Δtmin is increased by 50%-100% on the basis of the original setting, so as to reduce the average start-stop frequency in the transient sequence of adjacent start-stop event intervals in the start-stop frequency transient dataset, so that the two dimensions of current change rate and start-stop frequency jointly suppress the increase of the fatigue pre-evaluation index Fpre of the start-stop frequency density field in the subsequent start-stop evaluation window.

[0031] S3 also includes S32; S32. The thermal management control strategy sends the fatigue relative coefficient RF to the thermal management controller via a cloud-based prediction server. The thermal management controller then adjusts the operating parameters of the cooling device in stages based on the transient dataset of start-stop frequencies. The specific adjustments are as follows: When the start-stop frequency in the start-stop window is ≥25 times and the current average temperature of the battery pack is higher than 30℃, the thermal management controller executes pre-cooling control, increasing the cooling fan and cooling plate from the basic operating condition to 120%-150% of the basic operating power; Within the start / stop window, when the temperature change rate dT(t) / dt of any battery pack is detected to be greater than the temperature rise rate reference value RT, the thermal management controller performs cooling optimization control on the local cooling unit corresponding to the current battery pack, increasing the fan speed and cooling power of the current local cooling unit by 30%-60% based on the basic operating conditions, until the difference between the battery pack temperature and the average battery pack temperature drops to within 3℃, and the temperature change rate dT(t) / dt falls back to the range of 0.60×RT to 0.90×RT; By combining the above-mentioned pre-cooling control with local enhanced cooling control, under the premise of ensuring that the peak value of the temperature change rate dT(t) / dt is controlled and the duration of high temperature is shortened, the contribution of temperature transient shock to the fatigue pre-evaluation index Fpre of the start-stop frequency density field is weakened. Thus, in conjunction with the start-stop control strategy, electrical stress and thermal stress are jointly suppressed.

[0032] In this embodiment, step S31 of the method first stipulates that the trigger flag TF is only set to 1 when the fatigue relative coefficient RF exceeds two consecutive start-stop cycles and is judged to be in a fatigue state. This is to filter out occasional single heavy loads or abnormal operations, avoid frequent system interventions due to a single "accidental high current", and ensure that control actions only intervene when a "continuous high fatigue trend" occurs, which is truly meaningful for long-term lifespan. Under the condition of TF=1, on the one hand, the start-stop control strategy adjusts the motor starting current ramp-up time and minimum start-stop time interval according to RF levels: when RF is between 1.20 and 1.50, the ramp-up time and Δtmin are only moderately extended; when RF ≥ 1.50, the ramp-up time is significantly extended and d is reduced. I(t) / dt and multiplying Δtmin essentially "apply the brakes" to the two variables that directly determine the current density surge of the electrode plates: "how quickly the current climbs up" and "how frequently it starts and stops." This prevents the tractor from being in a state of high slope and high frequency of start-stop for a long time during peak operations. On the other hand, the thermal management control strategy in step S32 does the same thing from the temperature dimension. When the start-stop frequency is ≥25 times and the average temperature is >30℃ within the start-stop window, pre-cooling is performed in advance to lower the baseline temperature before subsequent start-stop. When it is detected that dT(t) / dt of a certain module is >RT or the local temperature difference is too large, local enhanced cooling is performed to "pull down" the highest temperature and maximum temperature rise, preventing a certain module from becoming a "high temperature island" for a long time. This combination of hot and cold lines controls both the rate of change of current and the rate of change of temperature, directly corresponding to the two physical components in Fpre. This effectively weakens the "impact energy" of each start-stop cycle. Without this approach, in real-world port scenarios, drivers or dispatchers could easily make dozens of short-distance starts and stops in quick succession to meet deadlines, or vehicles could repeatedly start and stop abruptly in high-temperature environments. While the vehicle might still be able to run in the short term, in the long term, this would significantly accelerate the growth of dendrites in the negative electrode, interface fatigue, and solder joint cracking. However, through this graded start-stop + thermal management coordinated control system triggered by RF and TF, these high-risk conditions can be automatically "softened and thinned" without changing the hardware. This significantly reduces the level of Fpre in subsequent start-stop windows, thereby substantially slowing down the accumulation of battery pack cycle fatigue and improving the overall vehicle lifespan and operational reliability. Example 5

[0033] Please see Figure 1 Specifically: S4 includes S41; S41. After the start-stop control strategy and thermal management control strategy are executed, the fatigue pre-evaluation index Fpre of the start-stop frequency density field is recalculated and generated in the next start-stop window. The cloud-based prediction server accumulates the fatigue damage of historical start-stop assessment windows to obtain the historical cumulative fatigue damage degree Dold. The historical cumulative fatigue damage degree Dold is a dimensionless quantity between 0 and 1, used to characterize the proportion of the battery pack's lifespan that has been consumed before the current moment. The total equivalent damage degree of the battery pack in the current start-stop assessment window is Dold plus the damage increment Fpre / Fref of the current window. The cloud-based prediction server uses the design life Lf as the benchmark for calculating the remaining life. The design life Lf is in units of time or number of cycles. For example, it is converted into a unified life scale based on 4,000 cycles or 8 years of operation. It is then combined with the historical cumulative fatigue damage degree Dold and the fatigue pre-assessment index Fpre of the start-stop frequency density field to calculate a new remaining life prediction value RULnew. The new remaining lifetime prediction value RULnew is calculated and output using the following algorithm formula: ; In the formula, Fpre and Fref are both dimensionless values, and the ratio of the two, Fpre / Fref, is a dimensionless quantity. The historical cumulative fatigue damage degree, Dold, is also a dimensionless quantity. (Dold+Fpre / Fref) in parentheses represents the proportion of the battery pack's lifespan that has been consumed at the current moment, and (1-(Dold+Fpre / Fref)) represents the proportion of the battery pack's remaining lifespan. The new remaining lifespan prediction value, RULnew, obtained by multiplying this remaining lifespan proportion by the design lifespan Lf, is consistent with the design lifespan Lf in terms of dimensions. Both can be characterized by operating hours, operating years, or equivalent cycle count.

[0034] S4 also includes S42; S42. Set a preset lifespan threshold RULth based on 20% of the design lifespan Lf. Then compare the new remaining lifespan prediction value RULnew with the lifespan threshold RULth to determine the current battery pack lifespan safety status. The specific comparison is as follows: When the new remaining life prediction value RULnew ≤ life threshold RULth, it indicates that the current battery pack's remaining life is abnormal. At this time, a life warning is issued to the operation and maintenance system, and the current battery pack is subjected to in-depth inspection and replacement. When the new remaining life prediction value RULnew > life threshold RULth, it indicates that the current start-stop control strategy and thermal management control strategy are experiencing accelerated fatigue, but the battery pack is still in an acceptable lifespan state, and monitoring should continue at this time.

[0035] In this embodiment, S41 of the method is to "calculate the overall effect" after the start-stop control strategy and thermal management control strategy are executed: the cloud prediction server recalculates the fatigue pre-evaluation index Fpre of the start-stop frequency density field in the next start-stop window, and superimposes the damage increment Fpre / Fref of the current window onto the historical cumulative fatigue damage degree Dold, compressing all the start-stop conditions that have occurred into a dimensionless life consumption ratio between 0 and 1, and then calculates a new remaining life prediction result RULnew in combination with the design life Lf (e.g., 4000 cycles or 8 years). Essentially, it uses "how much has been used plus how much has been used up in the current cycle" to directly convert "how much is left to be used safely"; then in S42, RULnew is compared with the life threshold RULth set at 20% of Lf. When RULnew≤RULth, the system actively issues a warning and requires in-depth inspection or replacement of the battery pack. When RULnew>RULth, it is considered that the life is still safe under the current control strategy, and only continued monitoring is needed. The reason for introducing the entire mechanism of Dold, Fpre / Fref, and RULth is that in high-intensity scenarios like ports, if we only look at "how many years it has been used" and "how many trips it has made," we often find that some battery packs are actually close to their limits due to frequent high-impact start-stop cycles, but their apparent lifespan and cycle count are not high. As a result, they suddenly lose power or even fail to shut down during peak operating periods. By implementing the above method, the fatigue of each start-stop cycle can be truly recorded in the lifespan ledger, and the decision to reduce load or replace the battery pack can be triggered before its lifespan drops to the 20% safety threshold. This transforms the approach of "using it until something goes wrong and then dealing with it" into "proactive intervention before it reaches its limit," thereby significantly reducing the risk of sudden failure and improving the utilization rate of battery pack lifespan and the continuity and safety of port operations.

[0036] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended technical solutions and their equivalents.

Claims

1. A battery pack lifetime prediction method based on multi-sensor parameter fusion, characterized in that: Includes the following steps: S1. Several collection points are set up in the battery pack of the automated container tractor in the port to collect transient sequence sets in real time and transmit the transient sequence sets to the cloud prediction server. The cloud prediction server performs start-stop window division and window data extraction on the transient sequence sets to obtain the start-stop frequency transient dataset. S2. Based on the start-stop frequency transient dataset, calculate the start-stop frequency density field fatigue pre-evaluation index Fpre, and then set the benchmark fatigue level Fref and the start-stop frequency density field fatigue pre-evaluation index Fpre for preliminary comparison and evaluation. S3. Based on the preliminary comparison and evaluation results, generate a trigger flag TF, and then execute the start-stop control strategy and thermal management control strategy according to the trigger flag TF generation results; S4. After executing the start-stop control strategy and thermal management control strategy, update the historical cumulative fatigue damage degree Dold based on the fatigue pre-evaluation index Fpre of the start-stop frequency density field, calculate the battery pack remaining life prediction result RULnew, then compare the preset life threshold RULth with the battery pack remaining life prediction result RULnew, and generate a response strategy.

2. The battery pack lifetime prediction method based on multi-sensor parameter fusion according to claim 1, characterized in that: S1 includes S11; S11. By setting up several collection points in the battery pack of the automated container tractor in the port, and collecting the transient sequence set of the battery pack in real time through the vehicle battery management system and vehicle controller during the operation of the battery pack. The transient sequence set is uploaded to the cloud prediction server via the vehicle communication gateway in the form of Hypertext Transfer Protocol messages. The data acquisition points include current acquisition points, temperature acquisition points, and start / stop event marker acquisition points; The transient sequence set includes current transient sequences, temperature transient sequences, and start / stop event marker sequences.

3. The battery pack lifetime prediction method based on multi-sensor parameter fusion according to claim 1, characterized in that: S1 further includes S12; S12. In the cloud prediction server, the transient sequence set of each port automated container tractor battery pack is parsed and cached. The current transient sequence, temperature transient sequence and start-stop event marker sequence in the transient sequence set are reordered in chronological order according to a unified time axis. The sorted transient sequence set is then stored in the order of the port automated container tractor battery pack number as an index. Then, based on the vehicle start event and braking end event marked by the start-stop event marker sequence, the unified time axis is divided into several start-stop windows, each start-stop window corresponding to a time interval [t0,t1] between a set of vehicle start events and braking end events.

4. The battery pack lifetime prediction method based on multi-sensor parameter fusion according to claim 3, characterized in that: S1 also includes S13; S13. For each start-stop window, extract the current transient sequence, temperature transient sequence, and start-stop event marker sequence that fall within the current start-stop window time interval from the transient sequence set of each port automated container tractor battery pack stored in the index, and form the start-stop window current subsequence, start-stop window temperature subsequence, and start-stop window voltage subsequence. The start-stop frequency is statistically analyzed based on the start-stop event marker sequence. The start-stop frequency transient dataset is composed of the start-stop window current subsequence, start-stop window temperature subsequence, start-stop window voltage subsequence and the corresponding start-stop frequency. The start-stop frequency transient dataset includes the current value I and the temperature value T within the start-stop window time interval [t0, t1]. Then, the physical quantities in the start-stop frequency transient dataset are numerically transformed according to the preset maximum value normalization rule to eliminate the differences in units and dimensions of all parameters in the start-stop frequency transient dataset.

5. The battery pack lifetime prediction method based on multi-sensor parameter fusion according to claim 4, characterized in that: S2 includes S21; S21. The cloud prediction server uses numerical differentiation to obtain the current change rate dI(t) / dt and temperature change rate dT(t) / dt at each sampling time based on the start-stop frequency transient dataset. Then, the cloud prediction server performs integral calculations based on the current change rate reference value RI and temperature change rate reference value RT, which are calibrated in advance under standard smooth acceleration and normal heat dissipation conditions, to obtain the fatigue pre-evaluation index Fpre of the start-stop frequency density field, and quantitatively analyzes the fatigue strength caused by the combined current transient impact and temperature transient impact within the stop window. The fatigue pre-evaluation index Fpre, which is the start-stop frequency density field, is calculated and output using the following algorithm formula: ； In the formula, d represents the integral function, dt represents the time integral function, I(t) represents the current value at time t within the start-stop window time interval [t0,t1], and T(t) represents the temperature value at time t within the start-stop window time interval [t0,t1].

6. The battery pack lifetime prediction method based on multi-sensor parameter fusion according to claim 5, characterized in that: S2 further includes S22; S22. During the initial safe operation of the battery pack, the fatigue pre-evaluation index Fpre of the start-stop frequency density field of multiple start-stop windows is statistically analyzed. The average fatigue pre-evaluation index Fpre of the start-stop frequency density field of the start-stop windows that are in normal working conditions and have not experienced abnormal decay is calculated to obtain the benchmark fatigue level Fref. During the online operation phase, the cloud-based prediction server calculates the ratio of the current start-stop frequency density field fatigue pre-assessment index Fpre to the baseline fatigue level Fref for each start-stop window, obtaining the fatigue relative coefficient RF. Based on the output of the fatigue relative coefficient RF, a preliminary comparative assessment is performed to determine the fatigue state of the current start-stop window. The specific assessment content is as follows: When the fatigue relative coefficient RF > 1, it indicates that the fatigue intensity of the current start-stop window exceeds the benchmark, and the current start-stop window is marked as a fatigue state. When the relative fatigue coefficient RF≤1, it means that the fatigue intensity of the current start-stop window has not exceeded the benchmark, and the current start-stop window is not marked.

7. The battery pack lifetime prediction method based on multi-sensor parameter fusion according to claim 6, characterized in that: S3 includes S31; S31. If the preliminary comparative evaluation results indicate a fatigue state and there are more than two consecutive start-stop cycles, then the current trigger flag TF=1, and the start-stop control strategy and thermal management control strategy will be triggered. Conversely, if the current trigger flag TF is set to 0, the control strategy cannot be triggered, and only data is recorded; The start-stop control strategy transmits the fatigue relative coefficient RF to the traction motor controller and the vehicle dispatching system. The traction motor controller uses the current change rate reference value RI as the adjustment target for the current change rate during the startup phase, as detailed below: When the relative fatigue coefficient RF is between 1.20 and 1.50, the current ramp-up time during the motor start-up phase is increased by 10%-30% from the original setting, and the actual current change rate dI(t) / dt is limited to the range of 0.80×RI to 1.00×RI by adjusting the ramp-up curve of the PWM duty cycle. When the relative fatigue coefficient RF≥1.50, the current ramp-up time during the motor start-up phase is increased by 30%-60% from the original setting, and the actual current change rate dI(t) / dt is limited to the range of 0.50×RI to 0.80×RI by limiting the current rise slope. The vehicle dispatching system adjusts the lower limit of the continuous start-stop event interval based on the fatigue relative coefficient RF: when the fatigue relative coefficient RF is between 1.20 and 1.50, the minimum start-stop time interval Δtmin is increased by 20%-50% from the original setting; When the relative fatigue coefficient RF ≥ 1.50, the minimum start-stop time interval Δtmin is increased by 50%-100% based on the original setting.

8. The battery pack lifetime prediction method based on multi-sensor parameter fusion according to claim 7, characterized in that: S3 further includes S32; S32. The thermal management control strategy sends the fatigue relative coefficient RF to the thermal management controller via a cloud prediction server. The thermal management controller adjusts the operating parameters of the cooling device in stages based on the transient dataset of start-stop frequencies. The specific adjustments are as follows: When the start-stop frequency in the start-stop window is ≥25 times and the current average temperature of the battery pack is higher than 30℃, the thermal management controller executes pre-cooling control, increasing the cooling fan and cooling plate from the basic operating condition to 120%-150% of the basic operating power; Within the start / stop window, when the temperature change rate dT(t) / dt of any battery pack is detected to be greater than the temperature rise rate reference value RT, the thermal management controller performs cooling optimization control on the local cooling unit corresponding to the current battery pack, increasing the fan speed and cooling power of the current local cooling unit by 30%-60% based on the basic operating conditions, until the difference between the battery pack temperature and the average battery pack temperature drops to within 3℃, and the temperature change rate dT(t) / dt falls back to the range of 0.60×RT to 0.90×RT.

9. The battery pack lifetime prediction method based on multi-sensor parameter fusion according to claim 8, characterized in that: S4 includes S41; S41. After the start-stop control strategy and thermal management control strategy are executed, the fatigue pre-evaluation index Fpre of the start-stop frequency density field is recalculated and generated in the next start-stop window. The cloud prediction server accumulates the fatigue damage of the historical start-stop evaluation window to obtain the historical cumulative fatigue damage degree Dold, which is a dimensionless quantity between 0 and 1. The cloud-based prediction server uses the design life Lf as the benchmark for calculating the remaining life, and calculates a new remaining life prediction value RULnew by combining it with the historical cumulative fatigue damage degree Dold and the fatigue pre-evaluation index Fpre of the start-stop frequency density field. The new remaining lifetime prediction value RULnew is calculated and output using the following algorithm formula: 。 10. The battery pack lifetime prediction method based on multi-sensor parameter fusion according to claim 9, characterized in that: S4 also includes S42; S42. Set a preset lifespan threshold RULth based on 20% of the design lifespan Lf. Then compare the new remaining lifespan prediction value RULnew with the lifespan threshold RULth to determine the current battery pack lifespan safety status. The specific comparison is as follows: When the new remaining life prediction value RULnew ≤ life threshold RULth, it indicates that the current battery pack's remaining life is abnormal. At this time, a life warning is issued to the operation and maintenance system, and the current battery pack is subjected to in-depth inspection and replacement. When the new remaining life prediction value RULnew > life threshold RULth, it indicates that the current start-stop control strategy and thermal management control strategy are experiencing accelerated fatigue, but the battery pack is still in an acceptable lifespan state, and monitoring should continue at this time.

Images