A thermoelectric coupling energy storage device predictive maintenance method, apparatus and program product

Abstract

The present disclosure belongs to the technical field of energy storage, and particularly relates to a thermoelectric coupling energy storage device predictive maintenance method, device and program product, comprising: calculating a thermoelectric coupling coefficient and a thermoelectric coupling entropy of the thermoelectric coupling energy storage device; taking an electric core in the thermoelectric coupling energy storage device as a node, a thermal path and an electric path of the thermoelectric coupling energy storage device as edges, constructing a thermoelectric weighted directed graph of the thermoelectric coupling energy storage device, and calculating an inconsistency influence factor of each node in the thermoelectric weighted directed graph; performing multi-level sorting on the inconsistency influence factors of all nodes in the thermoelectric weighted directed graph according to a level of the electric core in the thermoelectric coupling energy storage device; inputting the thermoelectric coupling coefficient, the thermoelectric coupling entropy and the multi-level sorting result of the inconsistency influence factors into a preset classifier, and performing predictive maintenance on the thermoelectric coupling energy storage device according to a fault determination result and a predictive maintenance instruction output by the classifier.

Description

Technical Field

[0001] This disclosure pertains to the field of energy storage technology, and specifically relates to a predictive maintenance method, device, and program product for thermoelectric coupled energy storage equipment. Background Technology

[0002] 100MW-level lithium iron phosphate containerized energy storage power stations need to maintain cell consistency over the long term to fulfill the guarantee of ≥90% available capacity in frequency regulation and spot arbitrage contracts. Temperature and voltage inconsistencies are the main causes of capacity decay and thermal runaway. Therefore, the industry commonly configures BMS equalization and air conditioning systems, relying on maintenance personnel to conduct regular random inspections within the containers. In existing technologies, data sensing involves the BMS polling voltage at 100ms–1s cycles, temperature probes reading single-point surface temperatures every 30s, and the cooling system independently reporting inlet and outlet temperatures. Consistency calculations involve calculating the maximum–minimum voltage difference ΔV and the maximum–minimum temperature difference ΔT in the EMS or local controller, using fixed thresholds: ΔV > 10 mV or ΔT > 3℃ triggers an alarm. Sorting is done simply by descending the ΔV value to form a voltage sequence list; temperature is only used for over-limit alarms and does not participate in the sorting. Maintenance decisions rely on maintenance personnel manually recharging or replacing fan filters on the cells ranked higher according to the sequence list, and strategy updates depend on semi-annual experience calibration.

[0003] Existing methods employ a voltage-temperature separation and a one-size-fits-all threshold judgment framework, resulting in several drawbacks: First, they ignore the coupling relationship between ΔV and ΔT. High temperatures are often accompanied by increased polarization voltage, easily misjudging thermal-electrical symbiotic anomalies as purely voltage problems, leading to low anomaly location accuracy. Second, they only perform numerical sorting at the cell level, lacking a unified evaluation scale across multiple levels from cell to module to cluster to container, making it impossible to compare the health status of different locations horizontally. Third, the judgment thresholds are fixed, failing to consider the drift of operating conditions such as SOC, SOH, and charge / discharge rate. Using the same temperature threshold in summer and winter results in a false alarm rate >15%. Fourth, the sorting results are disconnected from maintenance actions, requiring manual explanation of the sorting reasons on-site, leading to delayed maintenance response and significant losses due to passive shutdowns. These problems make it difficult to quantify thermal-electrical inconsistencies within energy storage containers, delay early degradation location, and prevent the automatic generation of maintenance strategies, seriously affecting the safe and stable operation and capacity guarantee capabilities of power plants. Summary of the Invention

[0004] To address the aforementioned problems, this disclosure provides a predictive maintenance method for thermoelectric coupled energy storage devices, which accurately predicts faults in these devices. The method includes: Calculate the thermoelectric coupling coefficient and thermoelectric coupling entropy of the thermoelectric coupled energy storage device; Using the battery cells in the thermoelectric coupled energy storage device as nodes and the thermal and electrical paths of the thermoelectric coupled energy storage device as edges, a thermoelectric weighted phase graph of the thermoelectric coupled energy storage device is constructed, and the inconsistency influence factor of each node in the thermoelectric weighted directed graph is calculated. Based on the cell hierarchy in the thermoelectric coupled energy storage device, the inconsistency influence factors of all nodes in the thermoelectric weighted directed graph are sorted in multiple levels. The thermoelectric coupling coefficient, the thermoelectric coupling entropy, and the multi-level sorting results of the inconsistency influence factors are input into a preset classifier. Based on the fault judgment results and predictive maintenance instructions output by the classifier, predictive maintenance is performed on the thermoelectric coupling energy storage device.

[0005] Further, the thermoelectric coupling coefficient and thermoelectric coupling entropy of the thermoelectric coupled energy storage device are calculated, including: The standardized temperature difference and standardized voltage difference are calculated based on the operating parameters of the thermoelectric coupled energy storage device; wherein, the operating parameters include: cell voltage, cell temperature, cell current, and cell cooling flow rate; the standardized voltage difference is expressed as follows: ΔVnorm = ΔV_raw (k1 ΔT_env + k2 ΔSOC + k3 ΔSOH + k4 C_rate) Where ΔVnorm represents the standardized voltage difference; ΔV_raw represents the collected voltage difference; ΔT_env represents the ambient temperature deviation; ΔSOC represents the SOC deviation of the battery cell; ΔSOH represents the battery cell health; C_rate represents the charge / discharge rate of the battery cell; k1, k2, k3, and k4 are all fitting coefficients; The standardized temperature difference, and the corresponding expression, includes: ΔT = T_raw (m1 ΔT_env + m2 ΔSOC + m3 ΔSOH + m4 C_rate) Where ΔT represents the standardized temperature difference; T_raw represents the collected temperature difference; and m1, m2, m3, and m4 are all fitting coefficients.

[0006] Furthermore, the thermoelectric coupling coefficient is the ratio of the standardized voltage difference to the standardized temperature difference, expressed as: Cvt = ΔT / ΔVnorm Wherein, Cvt represents the thermoelectric coupling coefficient; The expression corresponding to the thermoelectric coupling entropy is: Hvt = Σ(p i ln p i ) Where Hvt represents the thermoelectric coupling entropy; p i This represents the percentage of battery cells in the i-th interval of the statistical histogram of the thermoelectric coupling coefficient.

[0007] Further, the inconsistency impact factor of each node in the thermoelectric weighted directed graph is calculated, including: iteratively calculating the inconsistency impact factor based on the PageRank-Heat algorithm, and the corresponding expression includes: IF i (new) = (1 d) IF i (init) + d Σ(IF j (old) W ji ) Among them, IF i (new) represents the inconsistency impact factor of node i after iteration; IF j (old) represents the inconsistency impact factor of node i before iteration; IF i (init) represents the inconsistency impact factor of the initial node i; W ji d represents the edge weight from node i to node j; d is the damping coefficient.

[0008] Furthermore, the battery cell hierarchy in the thermoelectric coupled energy storage device includes a progressively higher level of battery cell, module, cluster, and energy storage device. Based on the battery cell hierarchy in the thermoelectric coupled energy storage device, the inconsistency influence factors of all nodes in the thermoelectric weighted directed graph are sorted in multiple levels, including: Based on weighted average and K-value penalty, the inconsistency influence factors of all nodes in each level are aggregated to the next level to achieve the multi-level sorting.

[0009] Furthermore, the multi-level ranking results also include: the comprehensive score and confidence interval of each health level, which are calculated based on the TOPSIS-Entropy method.

[0010] Furthermore, the classifier is constructed based on the TabNet neural network and the Isolation Forest algorithm.

[0011] This disclosure also proposes a predictive maintenance system for thermoelectric coupled energy storage devices for implementing the above-mentioned predictive maintenance method, comprising: The data layer, feature layer, network and ranking layer, and decision layer; among them, The data layer includes a synchronous sampling unit deployed on the thermoelectric coupled energy storage device, which is configured to collect the operating parameters of the thermoelectric coupled energy storage device. The feature layer is configured to calculate the thermoelectric coupling coefficient and thermoelectric coupling entropy based on the operating parameters; The network and sorting layer are used to construct a thermoelectric weighted phase graph of the thermoelectric coupled energy storage device, with the battery cells in the thermoelectric coupled energy storage device as nodes and the thermal and electrical paths of the thermoelectric coupled energy storage device as edges, and to calculate the inconsistency influence factor of each node in the thermoelectric weighted directed graph; and to perform multi-level sorting of the inconsistency influence factors of all nodes in the thermoelectric weighted directed graph according to the hierarchy of the battery cells in the thermoelectric coupled energy storage device. The decision layer, including a classifier, is configured to generate fault determination results and predictive maintenance instructions based on the multi-level ranking results of the thermoelectric coupling coefficient, the thermoelectric coupling entropy, and the inconsistency influence factors, and to perform predictive maintenance on the thermoelectric coupling energy storage device.

[0012] This disclosure also proposes a computer-readable storage medium storing a computer program or instructions, which, when executed by a processor, are at least used to implement the above-described predictive maintenance method for thermoelectric coupled energy storage devices.

[0013] This disclosure also proposes a computer program product stored in a computer-readable storage medium, which, when executed by a processor, is used to at least implement the above-described predictive maintenance method for thermoelectric coupled energy storage devices.

[0014] Compared with the prior art, this disclosure has the following advantages: This disclosure proposes a thermoelectric coupling coefficient and a "coupling entropy" index, which realizes the synchronous quantification of temperature and voltage deviations, solves the misjudgment and omission caused by the traditional method of treating the two independently, improves the accuracy of anomaly location, and achieves multi-level scale unification of thermoelectric coupled energy storage devices by sorting the inconsistency influence factors of thermoelectric weighted phase diagram nodes. Furthermore, it realizes automated predictive maintenance of thermoelectric coupled energy storage devices through a pre-trained classifier.

[0015] Other features and advantages of this disclosure will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the disclosure. The objects and other advantages of this disclosure may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A predictive maintenance method for a thermoelectric coupled energy storage device according to an embodiment is shown; Figure 2 A predictive maintenance system for a thermoelectric coupled energy storage device according to an embodiment is shown; Figure 3 A computer-readable storage medium according to an embodiment is shown. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0019] Figure 1 The predictive maintenance method for thermoelectric coupled energy storage devices proposed in this disclosure includes: Calculate the thermoelectric coupling coefficient and thermoelectric coupling entropy of the thermoelectric coupled energy storage device; Using the battery cells in the thermoelectric coupled energy storage device as nodes and the thermal and electrical paths of the thermoelectric coupled energy storage device as edges, a thermoelectric weighted phase graph of the thermoelectric coupled energy storage device is constructed, and the inconsistency influence factor of each node in the thermoelectric weighted directed graph is calculated. Based on the cell hierarchy in the thermoelectric coupled energy storage device, the inconsistency influence factors of all nodes in the thermoelectric weighted directed graph are sorted in multiple levels. The thermoelectric coupling coefficient, the thermoelectric coupling entropy, and the multi-level sorting results of the inconsistency influence factors are input into a preset classifier. Based on the fault judgment results and predictive maintenance instructions output by the classifier, predictive maintenance is performed on the thermoelectric coupling energy storage device.

[0020] According to some embodiments of this disclosure, the predictive maintenance method for thermoelectric coupled energy storage devices proposed in this disclosure synchronously collects multi-source data such as voltage, temperature, current, and cooling flow rate within the thermoelectric coupled energy storage device (e.g., an energy storage container). After four-factor correction based on ambient temperature, SOC, SOH, and rate of return, the thermoelectric coupling coefficient Cvt and coupling entropy Hvt are constructed and calculated as quantitative indicators. A thermoelectric weighted directed graph is established, with the battery cell as a node and the thermal / electric path as an edge. The inconsistency influence factor of each node is calculated and the consistency is sorted. The sorting result is input into a classifier, which outputs the fault judgment result and predictive maintenance instructions, thereby realizing predictive maintenance of the thermoelectric coupled energy storage device.

[0021] Further, the thermoelectric coupling coefficient and thermoelectric coupling entropy of the thermoelectric coupled energy storage device are calculated, including: The standardized temperature difference and standardized voltage difference are calculated based on the operating parameters of the thermoelectric coupled energy storage device; wherein, the operating parameters include: cell voltage, cell temperature, cell current, and cell cooling flow rate; the standardized voltage difference is expressed as follows: ΔVnorm = ΔV_raw (k1 ΔT_env + k2 ΔSOC + k3 ΔSOH + k4 C_rate) Where ΔVnorm represents the standardized voltage difference; ΔV_raw represents the collected voltage difference; ΔT_env represents the ambient temperature deviation; ΔSOC represents the SOC deviation of the battery cell; ΔSOH represents the battery cell health; C_rate represents the charge / discharge rate of the battery cell; k1, k2, k3, and k4 are all fitting coefficients; The standardized temperature difference, and the corresponding expression, includes: ΔT = T_raw (m1 ΔT_env + m2 ΔSOC + m3 ΔSOH + m4 C_rate) Where ΔT represents the standardized temperature difference; T_raw represents the collected temperature difference; and m1, m2, m3, and m4 are all fitting coefficients.

[0022] According to some embodiments of this disclosure, the standardized difference based on ambient temperature, SOC, SOH, and operating rate can eliminate the influence of ambient temperature, battery SOC (state of charge), SOH (health status), and operating rate on the original measured values, thus obtaining a standardized difference that better reflects the battery's own state: The formulas for calculating the standardized voltage difference and standardized temperature difference are as follows: ΔVnorm = ΔV_raw (k1 ΔT_env + k2 ΔSOC + k3 ΔSOH + k4 C_rate) ΔT = T_raw (m1 ΔT_env + m2 ΔSOC + m3 ΔSOH + m4 C_rate) Wherein, ΔVnorm represents the standardized voltage difference; ΔV_raw represents the collected voltage difference; ΔT_env represents the ambient temperature deviation; ΔSOC represents the cell's SOC deviation; ΔSOH represents the cell's health; C_rate represents the cell's charge / discharge rate; k1, k2, k3, k4, m1, m2, m3, and m4 are all fitting coefficients; ΔT represents the standardized temperature difference; T_raw represents the collected temperature difference, and the deviations of the above parameters are all the differences between the current value of the parameter and the preset standard reference value.

[0023] Furthermore, the thermoelectric coupling coefficient is the ratio of the standardized voltage difference to the standardized temperature difference, expressed as: Cvt = ΔT / ΔVnorm Wherein, Cvt represents the thermoelectric coupling coefficient; The expression corresponding to the thermoelectric coupling entropy is: Hvt = Σ(p i ln p i ) Where Hvt represents the thermoelectric coupling entropy; p i This represents the percentage of battery cells in the i-th interval of the statistical histogram of the thermoelectric coupling coefficient.

[0024] According to some embodiments of this disclosure, the thermoelectric coupling coefficient is used to quantify the correlation between temperature and voltage changes, and the formula for calculating the thermoelectric coupling coefficient is as follows: Cvt = ΔT / ΔVnorm Wherein, Cvt represents the thermoelectric coupling coefficient; the value of the thermoelectric coupling coefficient represents the temperature difference associated with a unit voltage difference, and the larger the value, the more abnormal the thermoelectric correlation.

[0025] According to some embodiments of this disclosure, thermoelectric coupling entropy is used to assess the uniformity and dispersion of the entire battery pack. The formula for calculating thermoelectric coupling entropy is as follows: Hvt = Σ(p i ln p i ) Where Hvt represents the thermoelectric coupling entropy; p i This represents the percentage of battery cells in the i-th interval of a statistical histogram representing the thermoelectric coupling coefficient. For example, it involves statistically analyzing the Cvt value distribution of all cells in the current battery pack, dividing it into 64 intervals as a histogram, and calculating the probability p for each interval. i (Number of cells in this interval / Total number of cells), with probability p for 64 intervals. i Substituting into the above equation, we can obtain the thermoelectric coupling entropy.

[0026] Further, the inconsistency impact factor of each node in the thermoelectric weighted directed graph is calculated, including: iteratively calculating the inconsistency impact factor based on the PageRank-Heat algorithm, and the corresponding expression includes: IF i (new) = (1 d) IF i (init) + d Σ(IF j (old) W ji ) Among them, IF i (new) represents the inconsistency impact factor of node i after iteration; IF j (old) represents the inconsistency impact factor of node i before iteration; IF i (init) represents the inconsistency impact factor of the initial node i; W ji d represents the edge weight from node i to node j; d is the damping coefficient.

[0027] According to some embodiments of this disclosure, graph sorting is based on the mutual influence of nodes in the battery network to identify the battery cell that poses the greatest threat to the overall consistency of the system. The process is as follows: 1. Construct a thermo-electric weighted directed graph: Treat each cell as a node. If two cells are physically adjacent (sharing a heat dissipation path or having a tight electrical connection), then establish an edge between them. Assign a weight to each edge, which can be designed as W = 1 / (thermal conductivity). Electrical conductivity (or conductivity) indicates the ease or difficulty of transmission.

[0028] 2. Iteratively calculate the inconsistency impact factor (IF), using the following formula: IF i (new) = (1 d) IF i (init) + d Σ(IF j (old) W ji ) Among them, IF i (new) represents the inconsistency impact factor of node i after iteration; IF j (old) represents the inconsistency impact factor of node i before iteration; IF i (init) represents the inconsistency impact factor of the initial node i; W ji d represents the edge weight from node i to node j; d is the damping coefficient.

[0029] 3. After performing the above iterations several times, the IF value of each node is finally converged. The higher the IF value, the greater the influence of the battery cell on the overall inconsistency of the thermoelectric coupled energy storage device, and the higher the ranking.

[0030] Furthermore, the battery cell hierarchy in the thermoelectric coupled energy storage device includes a progressively higher level of battery cell, module, cluster, and energy storage device. Based on the battery cell hierarchy in the thermoelectric coupled energy storage device, the inconsistency influence factors of all nodes in the thermoelectric weighted directed graph are sorted in multiple levels, including: Based on weighted average and K-value penalty, the inconsistency influence factors of all nodes in each level are aggregated to the next level to achieve the multi-level sorting.

[0031] According to some embodiments of this disclosure, after obtaining the inconsistency impact factor IF for each node, weighted aggregation is performed in the order of cell level to module level, module level to cluster level, and cluster level to energy storage device level (combined with K-value penalty to ensure that extreme individual cells are not submerged by the average value). The inconsistency impact factors of all nodes in each level are aggregated to the next level, that is, the inconsistency impact factor of the next level is obtained by weighted summation of all inconsistency impact factors of the previous level.

[0032] Furthermore, the multi-level ranking results also include: the comprehensive score and confidence interval of each health level, which are calculated based on the TOPSIS-Entropy method.

[0033] According to some embodiments of this disclosure, after aggregating the inconsistency influence factors of all nodes in each level to the next level to achieve multi-level sorting, the comprehensive health score and confidence interval of each level that can be compared horizontally are calculated based on the TOPSIS-Entropy method.

[0034] Furthermore, the classifier is constructed based on the TabNet neural network and the Isolation Forest algorithm.

[0035] According to some embodiments of this disclosure, after performing multi-level sorting of inconsistency impact factors, the sorting results are input into a TabNet+Isolation Fores classifier to generate predictive maintenance work orders for predictive maintenance of thermoelectric coupled energy storage equipment.

[0036] like Figure 2 As shown, this disclosure also proposes a predictive maintenance system for thermoelectric coupled energy storage devices to implement the above-described predictive maintenance method, comprising: The data layer, feature layer, network and ranking layer, and decision layer; among them, The data layer includes a synchronous sampling unit deployed on the thermoelectric coupled energy storage device, which is configured to collect the operating parameters of the thermoelectric coupled energy storage device. The feature layer is configured to calculate the thermoelectric coupling coefficient and thermoelectric coupling entropy based on the operating parameters; The network and sorting layer are used to construct a thermoelectric weighted phase graph of the thermoelectric coupled energy storage device, with the battery cells in the thermoelectric coupled energy storage device as nodes and the thermal and electrical paths of the thermoelectric coupled energy storage device as edges, and to calculate the inconsistency influence factor of each node in the thermoelectric weighted directed graph; and to perform multi-level sorting of the inconsistency influence factors of all nodes in the thermoelectric weighted directed graph according to the hierarchy of the battery cells in the thermoelectric coupled energy storage device. The decision layer, including a classifier, is configured to generate fault determination results and predictive maintenance instructions based on the multi-level ranking results of the thermoelectric coupling coefficient, the thermoelectric coupling entropy, and the inconsistency influence factors, and to perform predictive maintenance on the thermoelectric coupling energy storage device.

[0037] According to some embodiments of this disclosure, a predictive maintenance system for thermoelectric coupled energy storage devices includes: a data layer, a feature layer, a network and sorting layer, and a decision layer. 1. The data layer includes a synchronous sampling unit and a four-way correction module: (1) The synchronous sampling unit uses the LTC6810-1 chip to realize the acquisition of 192 series cells through daisy chain topology, with a polling period of 1ms and a resolution of 16-bit, to ensure high real-time and high accuracy of voltage data; Surface / terminal temperature: a 10kΩ±1%NTC thermistor is used, with a temperature range of -40℃~125℃, and synchronous acquisition in 1s to ensure the consistency of temperature data; Working current: a CH704 Hall sensor is used, with a range of 500A / 5V and a linearity of ±0.5%, to achieve distortion-free current acquisition; Cooling flow rate: a YF-S401 water flow sensor is used, with a range of 0.3-6L / min and a measurement error of ±3%, to complete the flow monitoring of the cooling system.

[0038] (2) The four-dimensional correction module is based on the built-in algorithm of the RK3568 processor. It takes ambient temperature, SOC, SOH and charge / discharge rate as input variables and corrects data drift through a four-dimensional linear regression model to eliminate data offset and error caused by complex working conditions, environment and aging.

[0039] 2. The feature layer includes a coupling coefficient calculation kernel, a coupling entropy calculator, and a high-frequency spike extractor: (1) The coupling coefficient calculation kernel is implemented based on FPGA logic, using a 30s sliding window, and the normalization benchmark is the laboratory calibration curve under the current SOC-SOH. (2) The coupling entropy calculator plots a 64-box histogram of the Cvt distribution and achieves discrete quantification consistency through the formula Hvt=-Σplnp. (3) The high-frequency spike extractor is based on wavelet IP and uses Daubechies-4 three-level decomposition to detect spikes with a temperature rise of >2σ, which is used for early warning of micro short circuits.

[0040] 3. The network and ranking layers include a graph builder, a PageRank-Heat kernel, a hierarchical aggregator, and a TOPSIS-Entropy sorter: (1) The graph builder is implemented based on FPGA. It uses battery cells as nodes and edge weights as the reciprocal of thermal conductivity × the reciprocal of resistance to generate a directed adjacency list. (2) The PageRank-Heat core is implemented on FPGA, iterates 20 times, multiplies and adds in parallel 256 channels, and outputs the node IF in 20µs; (3) The hierarchical aggregator is implemented based on ARM. The cell to module adopts the weighted average and K maximum penalty method, and the module to cluster and then to container adopt the same method; (4) The TOPSIS-Entropy sorter outputs healthrank and 95% confidence interval with IF, Hvt, number of thrusts, and cooldown ΔP as attributes.

[0041] 4. The decision-making layer includes the TabNet classifier, policy rule engine, and work order interface: (1) The TabNet classifier is implemented based on NPU. It takes an IF sequence as input and outputs four types of fault probabilities. (2) The policy rule engine executes the following rules: If IF > threshold 1 and cooling ΔP < threshold 2, it is determined that the cooling plate needs to be cleaned. If the IF value is greater than the threshold of 1 and the probability of a micro-short circuit is greater than 0.8, it is determined that power reduction is required and unpacking should be scheduled. The remaining decisions involve dynamically adjusting the BMS balancing and fan PID control. (3) The work order interface automatically generates Modbus-TCP commands and CSV work orders, and pushes them to the operation and maintenance SaaS.

[0042] 5. Acceleration and twin layers include FPGA accelerator cards and cloud-based digital twins: (1) The FPGA acceleration card uses Intel Cyclone VSoC, encapsulates the above 2a-3b logic, has a clock frequency of 64MHz, and the sorting of thousands of cells is ≤20ms; (2) The cloud-based digital twin receives the rank sequence through KubeEdge and uses FATE federated learning to update the RUL on a rolling basis, with a prediction error of <5%.

[0043] The above solution will be further explained below with reference to the embodiments: Example 1: A grid-side energy storage power station consists of 40 standard 40ft containers, each with a capacity of 2.5MW·h and 1536 280Ah lithium iron phosphate cells, using liquid cooling. The power station operates on a two-charge-two-discharge cycle per day, utilizing peak-valley arbitrage and frequency regulation, with an average load of 1C and a peak load of 2C. The highest temperature inside the containers reached 48℃ in summer. Traditional BMS only triggered alarms based on a 50mV voltage difference or a 5℃ temperature difference, leading to two thermal runaway shutdowns in 2022 caused by increased connection impedance, resulting in a direct loss of approximately 1.2 million yuan in revenue.

[0044] Data layer deployment: Each box is equipped with 64 AFE daughterboards, daisy chain 2MHz, FPGA generates 1PPS synchronization pulse, 1536 voltage channels, 3072 temperature channels (surface and pole), liquid cooling inlet and outlet temperatures, and flow rates are all refreshed in ≤1s; EMS forwards SOC, SOH, rate, and ambient temperature in real time, and ARM edge gateway completes four-way calibration to eliminate operating condition drift.

[0045] Feature layer calculation: ΔTnorm / ΔVnorm is calculated using a 300s sliding window to obtain Cvt; Hvt is generated statistically using a 64×64 two-dimensional grid; Espike is extracted from the 0.5-2Hz frequency band of wavelet packets. At 14:32:03 on July 13, 2023, the predictive maintenance system for thermoelectric coupled energy storage equipment detected Cvt > 1.2℃ / mV and Espike being higher than the background 3σ for 4 consecutive seconds, triggering a "micro-short circuit suspected" flag.

[0046] Multi-level sorting: Construct a 1536-node thermal-electric weighted directed graph. The FPGA acceleration card completes the PageRank-Heat iteration in 20ms and outputs IF_cell. After the weighted average of the modules and the K-value penalty, we get IF_module. Then we aggregate them into IF_rack and IF_container.

[0047] This box has an IF_container value of 78, ranking 37th out of 40 boxes on the entire site, and is in the red healthrank.

[0048] Predictive maintenance: TabNet input IF sequence and cooling pressure drop ΔP=8kPa (normal 12kPa), output P(C)=68%, mode "micro short circuit"; The rules engine automatically issued the following: reduce power to 0.3C and schedule unpacking during the nighttime low electricity price period three days later; The maintenance personnel disassembled the battery according to the work order and found that a 10μm lithium dendrite appeared on the negative terminal of the 17th cell in the 3rd module, which was consistent with the 11% increase in AC internal resistance. After replacing the module, the IF_container in the box rose back to 92, avoiding a potential thermal runaway.

[0049] Benefit assessment: Early detection of degradation 72 hours in advance reduces downtime by 16 hours, increases discharge capacity by 40 MW·h, and generates additional revenue of RMB 320,000; balanced power consumption is reduced by 18%, resulting in annual power savings of approximately 1.5 MW·h; the overall station lifespan is expected to be extended by 12%, resulting in an increase in net present value of approximately RMB 9 million.

[0050] Example 2: A 2MW / 4MW·h thermoelectric energy storage system in an industrial park uses eight 1MW·h liquid-cooled containers to provide transformer capacity expansion and replacement for 2C fast charging stations. During the midday peak in summer, the battery temperature often exceeds 45℃. Traditional BMS lacks cooling state quantification, and multiple instances of module temperature differences exceeding 8℃ due to cooling plate blockage have forced the system to operate at reduced power, resulting in a poor customer experience.

[0051] Hardware layer: 896 cells per box, AFE synchronous sampling period of 0.5s; added 0.1-level Hall current array and coolant ΔP sensor.

[0052] Operational event: 2023-08-06 11:15:00, the thermoelectric energy storage device IF_container dropped from 85 to 74, healthrank turned yellow, ΔP dropped from 11kPa to 7kPa, and Hvt increased by 22%.

[0053] Decision-making closed loop: The classifier outputs P(D)=72%, the pattern is "cooling plate blockage"; the rule engine automatically issues a "clean cooling plate" work order and simultaneously lowers the liquid cooling target temperature by 2℃; the maintenance personnel backwash on site 30 minutes later, ΔP recovers to 10kPa, IF_container rises back to 83, and the charging service is uninterrupted throughout the process.

[0054] Benefit assessment: Cooling fault repair time is reduced from 8 hours to 0.5 hours, resulting in an additional 1.6 MW·h of electricity sold and an additional revenue of RMB 12,000; customer demand electricity costs are reduced by approximately RMB 500,000 per year; the availability of thermoelectric energy storage equipment is increased from 96.2% to 99.1%, meeting the high reliability requirements of UPS.

[0055] In summary, the two embodiments verify that the predictive maintenance scheme for thermoelectric coupled energy storage devices proposed in this disclosure can complete thermo-electric consistency sorting in ≤1s under different operating conditions and automatically generate accurate maintenance work orders, significantly reducing downtime losses and extending battery life, and has broad application value.

[0056] like Figure 3 As shown, this disclosure also proposes a computer-readable storage medium storing a computer program or instructions, which, when executed by a processor, are at least used to implement the above-described predictive maintenance method for thermoelectric coupled energy storage devices.

[0057] This disclosure also proposes a computer program product stored in a computer-readable storage medium, which, when executed by a processor, is used to at least implement the above-described predictive maintenance method for thermoelectric coupled energy storage devices.

[0058] Although the present disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure.

Claims

1. A predictive maintenance method for thermoelectric coupled energy storage devices, characterized in that, include: Calculate the thermoelectric coupling coefficient and thermoelectric coupling entropy of the thermoelectric coupled energy storage device; Using the battery cells in the thermoelectric coupled energy storage device as nodes and the thermal and electrical paths of the thermoelectric coupled energy storage device as edges, a thermoelectric weighted phase graph of the thermoelectric coupled energy storage device is constructed, and the inconsistency influence factor of each node in the thermoelectric weighted directed graph is calculated. Based on the cell hierarchy in the thermoelectric coupled energy storage device, the inconsistency influence factors of all nodes in the thermoelectric weighted directed graph are sorted in multiple levels. The thermoelectric coupling coefficient, the thermoelectric coupling entropy, and the multi-level sorting results of the inconsistency influence factors are input into a preset classifier. Based on the fault judgment results and predictive maintenance instructions output by the classifier, predictive maintenance is performed on the thermoelectric coupling energy storage device.

2. The method as described in claim 1, characterized in that, The calculation of the thermoelectric coupling coefficient and thermoelectric coupling entropy of the thermoelectric coupled energy storage device includes: The standardized temperature difference and standardized voltage difference are calculated based on the operating parameters of the thermoelectric coupled energy storage device; wherein, the operating parameters include: cell voltage, cell temperature, cell current, and cell cooling flow rate; the standardized voltage difference is expressed as follows: ΔVnorm = ΔV_raw (k1 ΔT_env + k2 ΔSOC + k3 ΔSOH + k4 C_rate) Where ΔVnorm represents the standardized voltage difference; ΔV_raw represents the collected voltage difference; ΔT_env represents the ambient temperature deviation; ΔSOC represents the SOC deviation of the battery cell; ΔSOH represents the battery cell health; C_rate represents the charge / discharge rate of the battery cell; k1, k2, k3, and k4 are all fitting coefficients; The standardized temperature difference, and the corresponding expression, includes: ΔT = T_raw (m1 ΔT_env + m2 ΔSOC + m3 ΔSOH + m4 C_rate) Where ΔT represents the standardized temperature difference; T_raw represents the collected temperature difference; and m1, m2, m3, and m4 are all fitting coefficients.

3. The method as described in claim 2, characterized in that, The thermoelectric coupling coefficient is the ratio of the standardized voltage difference to the standardized temperature difference, expressed as: Cvt = ΔT / ΔVnorm Wherein, Cvt represents the thermoelectric coupling coefficient; The expression corresponding to the thermoelectric coupling entropy is: Hvt = Σ(p i in p i ) Where Hvt represents the thermoelectric coupling entropy; p i This represents the percentage of battery cells in the i-th interval of the statistical histogram of the thermoelectric coupling coefficient.

4. The method as described in claim 1, characterized in that, Calculating the inconsistency impact factor for each node in the thermoelectric weighted directed graph includes: iteratively calculating the inconsistency impact factor based on the PageRank-Heat algorithm, with corresponding expressions including: IF i (new) = (1 d) IF i (init) + d Σ(IF j (old) W ji ) Among them, IF i (new) represents the inconsistency impact factor of node i after iteration; IF j (old) represents the inconsistency impact factor of node i before iteration; IF i (init) represents the inconsistency impact factor of the initial node i; W ji d represents the edge weight from node i to node j; d is the damping coefficient.

5. The method as described in claim 1, characterized in that, The battery cell hierarchy in the thermoelectric coupled energy storage device includes a progressively higher level of battery cell, module, cluster, and energy storage device. Based on the battery cell hierarchy in the thermoelectric coupled energy storage device, the inconsistency influence factors of all nodes in the thermoelectric weighted directed graph are sorted in multiple levels, including: Based on weighted average and K-value penalty, the inconsistency influence factors of all nodes in each level are aggregated to the next level to achieve the multi-level sorting.

6. The method as described in claim 5, characterized in that, The multi-level ranking results also include: a comprehensive score and confidence interval for each level of health, which are calculated based on the TOPSIS-Entropy method.

7. The method according to any one of claims 1-6, characterized in that, The classifier is built based on the TabNet neural network and the Isolation Forest algorithm.

8. A predictive maintenance system for a thermoelectric coupled energy storage device for implementing the predictive maintenance method for any one of claims 1-7, characterized in that, include: The data layer, feature layer, network and ranking layer, and decision layer; among them, The data layer includes a synchronous sampling unit deployed on the thermoelectric coupled energy storage device, which is configured to collect the operating parameters of the thermoelectric coupled energy storage device. The feature layer is configured to calculate the thermoelectric coupling coefficient and thermoelectric coupling entropy based on the operating parameters; The network and sorting layer are used to construct a thermoelectric weighted phase graph of the thermoelectric coupled energy storage device, with the battery cells in the thermoelectric coupled energy storage device as nodes and the thermal and electrical paths of the thermoelectric coupled energy storage device as edges, and to calculate the inconsistency influence factor of each node in the thermoelectric weighted directed graph; and to perform multi-level sorting of the inconsistency influence factors of all nodes in the thermoelectric weighted directed graph according to the hierarchy of the battery cells in the thermoelectric coupled energy storage device. The decision layer, including a classifier, is configured to generate fault determination results and predictive maintenance instructions based on the multi-level ranking results of the thermoelectric coupling coefficient, the thermoelectric coupling entropy, and the inconsistency influence factors, and to perform predictive maintenance on the thermoelectric coupling energy storage device.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program or instructions, which, when executed by a processor, are used to implement at least the predictive maintenance method for thermoelectric coupled energy storage devices according to any one of claims 1-7.

10. A computer program product, said computer program product being stored in a computer-readable storage medium, characterized in that, When the computer program product is executed by a processor, it is used to implement at least the predictive maintenance method for thermoelectric coupled energy storage devices according to any one of claims 1-7.

Images