A communication base station machine room remote intelligent detection system based on digital twinning

Abstract

This application provides a remote intelligent detection system for communication base station equipment rooms based on digital twins, relating to the field of communication detection technology. It includes a digital twin thermodynamic simulation model running on a cloud server. Taking real-time acquired equipment room air conditioning operating status parameters as input, it outputs the expected surface temperature value of each battery cell under the assumption of no internal faults. An edge computing gateway acquires the equipment room air conditioning operating status parameters and collects the real-time surface temperature values ​​of each battery cell. For each battery cell, a residual temperature time series is generated and a trend test is performed to determine whether it exhibits a statistically significant monotonically increasing trend. The correlation between the residual temperature time series and the start / stop status signal of the air conditioning compressor is calculated to determine whether the two exhibit statistical decoupling. When both the monotonically increasing trend and statistical decoupling conditions are met simultaneously, an early warning signal for the corresponding battery cell's internal abnormal heat source is generated, achieving software-based removal of environmental interference.

Description

Technical Field

[0001] This invention belongs to the field of communication detection technology, specifically relating to a remote intelligent detection system for communication base station equipment rooms based on digital twins. Background Technology

[0002] Valve-regulated lead-acid battery packs in communication base station equipment rooms are in a floating charge standby state for a long time. The abnormal heat generation caused by early faults such as micro-short circuits, plate sulfation, and water loss is extremely low. It is usually manifested as the surface temperature of the battery cells slowly rising at a rate of 0.5°C to 1.0°C per hour.

[0003] Meanwhile, to reduce operational energy consumption, communication base stations generally adopt an intermittent start-stop and fixed-point air supply operation strategy for their equipment room air conditioners. The periodic start-stop of the air conditioner compressor and the directional blowing of cold air cause drastic fluctuations in the surface temperature of individual battery cells, ranging from 2°C to 5°C, with a cycle of several minutes to tens of minutes.

[0004] Under the aforementioned operating conditions, the weak temperature rise signal indicating early thermal runaway of the battery is drowned out in both time and amplitude by the ambient temperature fluctuations caused by the intermittent operation of the air conditioner. Existing technologies, such as the absolute temperature threshold alarm method and the fixed time window temperature rise rate alarm method, completely fail in the early stages of thermal runaway—when the absolute temperature or temperature rise rate triggers the alarm threshold, a large amount of irreversible heat has already accumulated inside the battery, and thermal runaway has entered an uncontrollable phase.

[0005] Therefore, there is an urgent need for a remote intelligent detection system that can accurately identify early abnormal internal heating of battery cells under the interference of intermittent air conditioning operation. Summary of the Invention

[0006] In view of the above-mentioned defects or deficiencies in the existing technology, a remote intelligent detection system for communication base station equipment rooms based on digital twins is provided, including a cloud server and an edge computing gateway deployed in the communication base station equipment room; wherein, The cloud server runs a digital twin thermodynamic simulation model, which is a simulation model that couples the operating status parameters of the computer room air conditioner with the thermodynamic parameters of the battery cells. It is used to calculate and output the expected surface temperature value of each battery cell under the preset assumption of no internal faults, taking the real-time acquired operating status parameters of the computer room air conditioner as input. The assumption of no internal faults is that the battery cells only have Joule heating of the float charge current and no abnormal internal heating. The edge computing gateway is used for: Real-time acquisition of the operating status parameters of the computer room air conditioner, including the start / stop status signal of the air conditioner compressor; Real-time surface temperature values ​​are collected from temperature sensors deployed on the surface of each battery cell. The operating status parameters of the computer room air conditioner and the real-time surface temperature value are uploaded to the cloud server after being stamped with a unified timestamp. The cloud server also includes a residual calculation unit and an anomaly detection unit; The residual calculation unit is used to calculate the difference between the real-time surface temperature value and the expected surface temperature value at the same timestamp for each battery cell, and generate a residual temperature time series. The anomaly detection unit is used to perform the following detection logic on the residual temperature time series: A trend test is performed on the residual temperature time series to determine whether it exhibits a statistically significant monotonically increasing trend. Calculate the correlation between the residual temperature time series and the start / stop status signal of the air conditioner compressor, and determine whether the two exhibit statistical decoupling; When both the monotonically increasing trend and the statistical decoupling conditions are met simultaneously, an early warning signal for the internal abnormal heat source of the corresponding battery cell is generated.

[0007] According to the technical solution provided in this application, the digital twin thermodynamic simulation model is embedded with a spatial distribution function of the heat transfer coefficient. The spatial distribution function is used to characterize the spatial difference in the local heat transfer capacity of each battery cell caused by the attenuation of the airflow from the computer room air conditioner along the battery pack arrangement direction. The cloud server also includes a spatial distribution function calibration module. The spatial distribution function calibration module is used to extract the temperature response rate difference of each battery cell to the same air conditioner status change event by using the dynamic response data of the surface temperature of each battery cell collected by the edge computing gateway before and after the air conditioner operation status change event, and calibrate the parameters of the spatial distribution function based on the temperature response rate difference. When calculating the expected surface temperature of each battery cell, the digital twin thermodynamic simulation model determines the corresponding heat transfer coefficient based on the spatial location of the battery cell in the battery pack using the calibrated spatial distribution function, and then uses the heat transfer coefficient to calculate the expected surface temperature.

[0008] According to the technical solution provided in this application, the spatial distribution function calibration module is specifically configured for: After the air conditioner compressor is in a continuous shutdown state and the surface temperature of each battery cell reaches thermal equilibrium, the edge computing gateway is controlled to trigger the air conditioner compressor to start and collect data on the process of the decrease in surface temperature of each battery cell after the compressor starts. The temperature drop data of each battery cell are fitted to extract the characteristic temperature drop rate of each battery cell. The characteristic temperature drop rate of the battery cell at the starting point of the air supply airflow is selected as the reference rate. The ratio of the characteristic temperature drop rate of the remaining battery cells to the reference rate is calculated, and this ratio is used as the heat transfer coefficient attenuation factor for the corresponding spatial location. Using the row and column position coordinates of the battery cells on the battery rack as independent variables and the heat transfer coefficient attenuation factor as the dependent variable, the functional expression and parameters of the spatial distribution function are obtained by fitting.

[0009] According to the technical solution provided in this application, the cloud server further includes a spatial distribution function failure self-checking module, specifically configured for: During normal system operation, the residual temperature time series of each battery cell within a preset self-test cycle is obtained from the residual calculation unit, and the average residual temperature of each battery cell within that cycle is calculated. All battery cells in the battery pack located in the first half of the air supply stroke are divided into the front row of cells, and all battery cells located in the second half of the air supply stroke are divided into the back row of cells. Calculate the arithmetic mean of the residual temperature values ​​of all battery cells in the front row set as the representative value of the front row residual; calculate the arithmetic mean of the residual temperature values ​​of all battery cells in the back row set as the representative value of the back row residual. When the representative value of the front row residual is less than the first preset threshold and the representative value of the back row residual is greater than the second preset threshold, the spatial distribution function is determined to be invalid, and the spatial distribution function calibration module is triggered to re-execute the calibration process; wherein, the first preset threshold is a negative value and the second preset threshold is a positive value.

[0010] According to the technical solution provided in this application, the cloud server further includes an abnormal heat source inversion unit and a backup capability assessment unit; The abnormal heat source inversion unit is configured to, after the abnormality determination unit generates an internal abnormal heat source warning signal for the target battery cell, invert the equivalent abnormal heat generation power of the target battery cell at the current moment based on the residual temperature time series of the target battery cell. The backup capability assessment unit is configured to obtain the current state of charge of the battery pack to which the target battery cell belongs and the typical load power of the communication base station equipment room. It takes the equivalent abnormal heat generation power, the current state of charge, and the typical load power as inputs to drive the digital twin thermodynamic simulation model to perform virtual discharge simulation under a preset mains power outage scenario, obtain the estimated backup time of the target battery cell under the condition of the presence of the internal abnormal heat source, and output the backup capability assessment result based on the estimated backup time.

[0011] According to the technical solution provided in this application, the abnormal heat source inversion unit is specifically configured for: Obtain the residual temperature time series of the target battery cell within a first preset time period before the warning time, and the expected surface temperature time series output by the digital twin thermodynamic simulation model within the first preset time period; Using the lumped heat capacity parameter and surface heat transfer coefficient of the target battery cell as known quantities, and the residual temperature time series as observed values, the equivalent abnormal heating power time series inside the target battery cell within the first preset time period is solved by using the heat balance equation in reverse. The equivalent abnormal heating power time series is subjected to trend fitting, and the equivalent abnormal heating power value at the current moment is extracted as the equivalent abnormal heating power obtained by inversion.

[0012] According to the technical solution provided in this application, the backup capability assessment unit is specifically configured for: Obtain the preset mains power interruption backup time requirement of the battery pack to which the target battery cell belongs; The equivalent abnormal heat generation power value at the current moment, the average growth rate of the equivalent abnormal heat generation power within the second preset time period, the current state of charge, and the typical load power are used as boundary conditions to drive the digital twin thermodynamic simulation model to perform a virtual discharge simulation. The virtual discharge simulation simulates the time it takes for the terminal voltage of the target battery cell to drop to the cutoff voltage due to the continuous evolution of internal abnormal heat generation under a preset mains power interruption scenario. This time period is used as the estimated backup time. The estimated backup time is compared with the preset backup time requirement for mains power outage, and the backup capability assessment result of the target battery cell is output. The backup capability assessment result includes at least one of the following levels: the backup time can be met, there is a risk of insufficient backup time, and it is recommended to replace it immediately.

[0013] According to the technical solution provided in this application, the abnormal heat source inversion unit is further configured for; After obtaining the residual temperature time series within a first preset time period before the warning time and before performing the reverse solution of the heat balance equation, the residual temperature time series is purified. The purification process includes: Obtain the time series of battery pack float charging current collected by the edge computing gateway within the first preset time period; The floating charge current time series is input into the digital twin thermodynamic simulation model, which drives the digital twin thermodynamic simulation model to calculate the time series of the floating charge current temperature change component of the target battery cell under the floating charge current fluctuation condition and the assumption of no internal fault. Subtracting the float charge current-induced temperature variation component time series from the residual temperature time series yields the purified residual temperature time series. The abnormal heat source inversion unit uses the purified residual temperature time series as the observed value to perform the reverse solution of the heat balance equation.

[0014] According to the technical solution provided in this application, the backup capability assessment unit is further configured to perform the following operations before executing the virtual discharge simulation: Determine whether the equivalent abnormal heating power at the current moment is lower than a preset power threshold; When it is determined that the equivalent abnormal heating power is lower than the power threshold, a preset conservative growth rate is used instead of the average growth rate of the equivalent abnormal heating power within a second preset time period as the boundary condition for the virtual discharge simulation. The preset conservative growth rate is a growth rate value with a preset upper limit of confidence level, extracted by statistical analysis of the historical evolution data of abnormal heat generation power of the same type of battery under micro short-circuit fault mode.

[0015] According to the technical solution provided in this application, the backup capability assessment unit is further configured to perform the following operations before executing the virtual discharge simulation: Determine whether the value of the equivalent abnormal heating power at the current moment reaches or exceeds a preset power threshold; When it is determined that the equivalent abnormal heating power reaches or exceeds the power threshold, the average growth rate of the equivalent abnormal heating power within a second preset time period is nonlinearly compensated and corrected. The nonlinear compensation correction includes: Obtain the second-order difference statistics of the time series of the equivalent abnormal heating power within the second preset time period, and determine whether the evolution of the equivalent abnormal heating power shows an accelerated characteristic based on the second-order difference statistics. When acceleration characteristics are determined, the acceleration factor is determined based on the value of the second-order difference statistic, and the average growth rate is multiplied by the acceleration factor as the boundary condition for the virtual discharge simulation. When it is determined that no acceleration characteristics are present, the average growth rate is directly used as the boundary condition for the virtual discharge simulation.

[0016] Compared with the prior art, the beneficial effects of this application are as follows: I. A digital twin thermodynamic simulation model is used to generate a target surface temperature benchmark under the assumption of no internal faults, achieving software-based removal of environmental interference. This system does not directly rely on absolute temperature values ​​or temperature rise rates for fault determination. Instead, it utilizes a simulation model that couples the operating status parameters of the computer room air conditioning with the thermodynamic parameters of the battery to dynamically calculate the target surface temperature value that each battery cell should have under the assumption of only float charging Joule heat and no internal abnormal heating. This target temperature value naturally includes the influence of air conditioning start-stop on battery temperature, providing a dynamic and adaptive comparison benchmark for subsequent residual calculations. In principle, this solves the problem that the fixed threshold alarm method cannot adapt to varying operating conditions.

[0017] II. Through decoupling the residual temperature from environmental driving signals, the system achieves ultra-early identification of weak internal abnormal heat sources. This system defines the difference between the measured and expected temperatures as the residual temperature. Using a dual-judgment logic of trend testing and correlation analysis, it identifies the component of the residual temperature that is statistically decoupled from the air conditioning compressor start / stop signal and exhibits a monotonically increasing trend. This component, after removing environmental interference from the air conditioning system, is a pure temperature rise signal contributed solely by abnormal internal battery heating. It can detect precursors to thermal runaway weeks to months in advance, even when the battery's absolute temperature is still within the normal range and traditional alarm methods are completely ineffective, providing maintenance personnel with ample response time.

[0018] Third, no new hardware sensors are required; the system is based on existing monitoring data and computing resources. The required operating status parameters of the computer room air conditioning can be directly read from the air conditioning controller via the Modbus / SNMP protocol, or indirectly obtained through non-intrusive smart meters. The surface temperature of individual battery cells is collected using temperature sensors from the existing battery monitoring system. All computing tasks are handled by the cloud server; the edge computing gateway is only responsible for data acquisition and timestamp alignment. This does not increase on-site hardware deployment costs or construction complexity, making it suitable for the intelligent transformation of existing communication base stations. Attached Figure Description

[0019] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 A schematic diagram of the structure of the remote intelligent detection system for communication base station equipment rooms based on digital twins provided in this application.

[0020] The text labels in the image represent: 1. Cloud server; 2. Edge computing gateway. Detailed Implementation

[0021] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0022] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] As mentioned in the background section, this application proposes a remote intelligent detection system for communication base station equipment rooms based on digital twins, such as... Figure 1 As shown, it includes a cloud server 1 and an edge computing gateway 2 deployed in the communication base station equipment room; wherein, The cloud server 1 runs a digital twin thermodynamic simulation model, which is a simulation model that couples the operating status parameters of the computer room air conditioner with the thermodynamic parameters of the battery cell. It is used to calculate and output the expected surface temperature value of each battery cell under the preset assumption of no internal faults, taking the real-time acquired operating status parameters of the computer room air conditioner as input. The assumption of no internal faults is that the battery cell only has Joule heating of the float charge current and no abnormal internal heating. The edge computing gateway 2 is used for: Real-time acquisition of the operating status parameters of the computer room air conditioner, including the start / stop status signal of the air conditioner compressor; Real-time surface temperature values ​​are collected from temperature sensors deployed on the surface of each battery cell. The operating status parameters of the computer room air conditioner and the real-time surface temperature value are uploaded to the cloud server 1 after being stamped with a unified timestamp. The cloud server 1 also includes a residual calculation unit and an anomaly detection unit; The residual calculation unit is used to calculate the difference between the real-time surface temperature value and the expected surface temperature value at the same timestamp for each battery cell, and generate a residual temperature time series. The anomaly detection unit is used to perform the following detection logic on the residual temperature time series: A trend test is performed on the residual temperature time series to determine whether it exhibits a statistically significant monotonically increasing trend. Calculate the correlation between the residual temperature time series and the start / stop status signal of the air conditioner compressor, and determine whether the two exhibit statistical decoupling; When both the monotonically increasing trend and the statistical decoupling conditions are met simultaneously, an early warning signal for the internal abnormal heat source of the corresponding battery cell is generated.

[0024] Specifically, the system is physically deployed in a two-tier architecture: an edge computing gateway 2 deployed in the communication base station equipment room, and a remote cloud server 1. The edge computing gateway 2 is responsible for data aggregation and preprocessing, and its hardware can be an industrial-grade embedded computer, a programmable logic controller, or a smart IoT gateway device with multi-channel data access capabilities. This gateway establishes data links with the air conditioning controller and battery monitoring unit in the equipment room via protocols such as RS485 bus, Modbus TCP, SNMP, or wireless LoRa.

[0025] The core component running on cloud server 1 is a digital twin thermodynamic simulation model, specifically a mathematical simulation model based on physical mechanisms, which models the coupling relationship between the thermodynamic behavior of individual battery cells and the thermal environment of the data center. The first part of the model is a lumped-parameter thermal model of the individual battery cells, which treats each cell as an object with a uniform internal temperature distribution, using an equivalent heat capacity parameter to describe its heat storage capacity and a surface heat transfer coefficient to describe the intensity of convective heat transfer between it and the surrounding air. The second part of the model is a simplified model of the data center thermal environment, which maps the air conditioning operating status parameters to the effective heat transfer coefficient correction applied to the surface of each individual battery cell. The aforementioned air conditioning operating status parameters include at least the start / stop status signal of the air conditioning compressor, which can be obtained by reading the operating status register of the air conditioning controller or indirectly determined by installing a non-invasive current sensor on the air conditioning power supply circuit and analyzing its current harmonic characteristics. When the compressor is running, the model assumes that forced convection exists in the computer room, and the heat transfer coefficient of the battery surface is taken as a higher value; when the compressor is stopped, the model assumes that only natural convection exists in the computer room, and the heat transfer coefficient is taken as a lower value.

[0026] The core function of the digital twin thermodynamic simulation model is to dynamically calculate and output the expected surface temperature value of each battery cell under a preset assumption of no internal faults, using real-time acquired air conditioning operating status parameters as input excitation. This assumption of no internal faults is specifically defined as follows: it is assumed that only Joule heat generated by the float charge current flowing through the internal resistance of the battery cell exists, and there is no additional abnormal heating caused by internal faults such as micro-short circuits, plate sulfation, or water loss. Under this assumption, the model's output temperature trajectory is entirely determined by environmental heat exchange, reflecting the normal thermal response of the battery to changes in the air conditioning environment under healthy conditions. Edge computing gateway 2 is also responsible for collecting real-time surface temperature values ​​output by temperature sensors deployed on the surface of each battery cell. The sensor types can be PT100 platinum resistance thermometers, NTC thermistors, or digital temperature sensors. The gateway adds a unified timestamp to the collected air conditioning operating status parameters and real-time surface temperature values. This operation aims to ensure that the two data streams are precisely aligned on the time axis during subsequent residual calculations. The timestamp accuracy should reach the second or millisecond level, and then the data is uploaded to the cloud server 1 via a 4G or 5G network.

[0027] The cloud server 1 contains a residual calculation unit and an anomaly detection unit. The residual calculation unit uses timestamps as indexes to calculate the arithmetic difference between the real-time surface temperature value and the expected surface temperature value at the same moment, forming a residual temperature time series that changes over time. The physical meaning of this residual temperature series is that it characterizes the degree of deviation between the actual thermal state of the battery and its ideal healthy thermal state. The anomaly detection unit performs a dual-judgment logic on this residual temperature time series. The first judgment is a trend test, specifically implemented using the Mann-Kendall nonparametric trend test algorithm to analyze the residual series within the sliding time window. When the standardized value of the test statistic exceeds a preset significance level, such as the threshold corresponding to a 95% confidence interval, the residual series is judged to show a statistically significant monotonically increasing trend. The second judgment is a correlation test, specifically implemented by calculating the Pearson correlation coefficient between the residual temperature series and the air conditioner compressor start-stop state series within the same time window. When the absolute value of the correlation coefficient is less than a preset threshold, such as 0.2, the two are judged to show statistical decoupling, meaning that the change pattern of the residual is no longer synchronized with the air conditioner's operation. The anomaly detection unit generates an internal abnormal heat source warning signal for the battery cell only when both of the above conditions are met.

[0028] The technical advantage of this solution lies in generating a predictable temperature baseline that dynamically adapts to the air conditioning status by constructing a digital twin model, thus separating the interference from ambient temperature fluctuations from fault characteristics. The underlying principle is that temperature changes caused by the air conditioning's start-up and shutdown simultaneously affect both the actual battery and the virtual model. Subtracting the two results cancels out the common interference component, leaving only the temperature rise signal contributed by abnormal internal battery heating in the residual. This allows the system to detect milliwatt-level precursors of abnormal internal heating even when the battery's surface absolute temperature is still within the normal range, achieving ultra-early warning.

[0029] In a preferred embodiment, the digital twin thermodynamic simulation model is embedded with a spatial distribution function of the heat transfer coefficient. The spatial distribution function is used to characterize the spatial differences in the local heat transfer capacity of each battery cell caused by the attenuation of the airflow from the computer room air conditioner along the battery pack arrangement direction. The cloud server 1 also includes a spatial distribution function calibration module. The spatial distribution function calibration module is used to extract the temperature response rate difference of each battery cell to the same air conditioner status change event by using the dynamic response data of the surface temperature of each battery cell collected by the edge computing gateway 2 before and after the air conditioner operation status change event, and calibrate the parameters of the spatial distribution function based on the temperature response rate difference. When calculating the expected surface temperature of each battery cell, the digital twin thermodynamic simulation model determines the corresponding heat transfer coefficient based on the spatial location of the battery cell in the battery pack using the calibrated spatial distribution function, and then uses the heat transfer coefficient to calculate the expected surface temperature.

[0030] Specifically, the core improvement of this embodiment lies in addressing the issue of varying heat transfer intensities faced by battery cells in different spatial locations within the same equipment room due to airflow attenuation. Battery packs in communication base station equipment rooms are typically installed in multiple layers and columns on standard battery racks. After the air conditioning airflow passes through the front row of batteries, its velocity decreases and temperature increases due to airflow resistance and heat exchange with the battery casings, resulting in a significantly worse local heat transfer environment for the rear row of batteries compared to the front row. If the digital twin thermodynamic simulation model uses the same heat transfer coefficient for all cells, it will lead to a systematic deviation in the prediction of the expected temperature of the rear row of batteries. To solve this problem, this embodiment defines a spatial distribution function of the heat transfer coefficient embedded in the digital twin thermodynamic simulation model. This spatial distribution function is a mathematical mapping relationship, with the independent variable being the spatial position of the battery cell on the battery rack, which can be represented by row and column coordinates or by the Euclidean distance from the air conditioning outlet; and the dependent variable being the attenuation factor of the local heat transfer coefficient corresponding to that cell relative to the reference heat transfer coefficient, which is a value less than or equal to 1. The physical essence of this spatial distribution function is to characterize the spatial differences in the local heat exchange capacity of each battery cell caused by the attenuation of the airflow from the computer room air conditioner along the battery pack arrangement direction.

[0031] The cloud server 1 also includes a spatial distribution function calibration module. This module automatically determines the spatial distribution function parameters during system initialization or upon receiving a recalibration command. Its specific workflow is as follows: The calibration module continuously monitors the air conditioning operating status parameters uploaded by the edge computing gateway 2. When an air conditioning operating status change event is detected, such as the air conditioning compressor switching from a stopped state to a started state, the calibration module begins capturing the dynamic response data of the surface temperature of each battery cell before and after the event. For a compressor start-up event, the surface temperature of each cell will undergo a decrease process from an initial steady state to a new low-temperature steady state. The calibration module extracts the characteristic temperature decrease rate of each cell during this transition process. This extraction method can be to fit an exponential function to the temperature decrease curve and take the time constant of the fitted function as the response rate index, or to directly calculate the slope of the initial linear segment of the temperature decrease curve as the response rate. Because each cell has a different spatial position on the battery rack, the perceived airflow intensity is different, thus the temperature decrease rate also exhibits spatial distribution differences. The calibration module extracts the temperature response rate differences of each cell to the same air conditioning status change event and calibrates the parameters of the spatial distribution function based on these differences. The specific calibration logic will be further elaborated later.

[0032] After the system is put into normal operation, the digital twin thermodynamic simulation model calculates the expected surface temperature of each battery cell by first using the cell's row and column coordinates within the battery pack, then calling the calibrated spatial distribution function, and either looking up a table or substituting it into a formula to calculate the heat transfer coefficient attenuation factor corresponding to that location. Next, the baseline heat transfer coefficient is multiplied by this attenuation factor to obtain the actual heat transfer coefficient applicable to that cell. Finally, this actual heat transfer coefficient is substituted into the heat balance equation to calculate the expected surface temperature of that cell.

[0033] The technical solution achieves its effect by correcting, purely through software, the inconsistency in assessing the heat exchange capacity of batteries at different locations caused by uneven airflow organization within the computer room. The underlying principle is to utilize the inherent thermal excitation signal of changes in air conditioning status. By observing the response speed of each individual cell to this excitation, the relative heat exchange intensity distribution at various points in the space is derived, and this distribution is then embedded as a spatial distribution function into a digital twin model. This ensures that the model's calculation of the expected temperature for each battery takes into account its unique local thermal environment, significantly improving the baseline accuracy of residual calculations and avoiding false alarms caused by insufficient systemic heat exchange in the rear battery system.

[0034] In a preferred embodiment, the spatial distribution function calibration module is specifically configured to: After the air conditioner compressor is in a continuous shutdown state and the surface temperature of each battery cell reaches thermal equilibrium, the edge computing gateway 2 is controlled to trigger the air conditioner compressor to start and collect data on the process of the decrease in surface temperature of each battery cell after the compressor starts. The temperature drop data of each battery cell are fitted to extract the characteristic temperature drop rate of each battery cell. The characteristic temperature drop rate of the battery cell at the starting point of the air supply airflow is selected as the reference rate. The ratio of the characteristic temperature drop rate of the remaining battery cells to the reference rate is calculated, and this ratio is used as the heat transfer coefficient attenuation factor for the corresponding spatial location. Using the row and column position coordinates of the battery cells on the battery rack as independent variables and the heat transfer coefficient attenuation factor as the dependent variable, the functional expression and parameters of the spatial distribution function are obtained by fitting.

[0035] Specifically, the core of this implementation method is to use the specific event of the air conditioning compressor switching from a continuously stopped state to a started state as a standard excitation source to stimulate the dynamic temperature response of each unit.

[0036] The first step in the calibration process is condition judgment and preparation. The spatial distribution function calibration module analyzes the start / stop status signal of the air conditioner compressor uploaded by the edge computing gateway 2 to confirm that the air conditioner compressor is in a continuous shutdown state and the shutdown time has exceeded a preset threshold, such as two hours. During this period, the calibration module continuously monitors the surface temperature of each battery cell. When it is determined that the maximum fluctuation of the surface temperature of all cells in the past fifteen minutes is less than 0.1 degrees Celsius, it is considered that each cell has reached thermal equilibrium, that is, there is no significant temperature difference between cells, and the entire battery pack is in a uniform ambient temperature field.

[0037] The second step involves applying the stimulus and acquiring data. After confirming that the thermal equilibrium state has been achieved, the calibration module sends a compressor start command to the air conditioning controller through the edge computing gateway 2, or waits for the air conditioning control system to start the compressor according to its own logic. Taking the compressor start time as the zero point, the edge computing gateway 2 synchronously acquires the data on the temperature drop process of each battery cell surface within a preset time period, such as ten minutes, at a sampling frequency of no less than once per second, forming a temperature drop curve for each cell.

[0038] The third step is feature extraction. The calibration module performs mathematical fitting on the temperature drop data of each battery cell. A preferred fitting method is to use a first-order exponential decay model, whose expression is T(t) = T_target + (T_start - T_target) × e (-t / τ) Where T_start is the initial temperature at startup, T_target is the final approached steady-state temperature, and τ is a time constant characterizing the rate of temperature decrease. The measured data are fitted using the least squares method to extract the time constant τ for each individual cell. A smaller τ value indicates a faster temperature decrease, meaning a stronger heat transfer capacity at that location. This τ value, or its reciprocal, can be used as the characteristic temperature decrease rate for that individual cell.

[0039] The fourth step is the relative ratio calculation. The calibration module automatically identifies, or manually presets, the battery cell at the starting point of the airflow path of the air conditioner, based on the actual physical layout of the battery pack. This starting cell is typically the cell in the first row and first column closest to the air conditioner vent. The characteristic temperature drop rate corresponding to this starting cell is used as the reference rate. For each of the remaining battery cells, its characteristic temperature drop rate is divided by the reference rate; the resulting ratio is the heat transfer coefficient attenuation factor for that cell's spatial location. Since the temperature drop rate of cells in the rear row is necessarily less than the reference rate of the front row, this ratio is always less than or equal to 1.

[0040] The fifth step is function fitting. The calibration module uses the row and column position coordinates of the battery cells on the battery rack as independent variables, for example, the first column of the first row as the origin, the row direction as the X-axis, and the column direction as the Y-axis. The heat transfer coefficient attenuation factor calculated above is used as the dependent variable. Polynomial surface fitting or exponential decay function fitting methods are used to obtain the function expression and parameters of the spatial distribution function. For example, the heat transfer coefficient attenuation factor is denoted as α, and its fitting function relationship with the row and column position coordinates X and Y of the battery cell is: α(X, Y) = exp(-a·X - b·Y); where: α is the heat transfer coefficient attenuation factor at the corresponding spatial position (X, Y), dimensionless, and its value range is (0, 1]; X is the row direction coordinate of the battery cell on the battery rack, starting from the row coordinate of the cell at the starting point of the airflow path as 0, and increasing along the airflow direction; Y is the column direction coordinate of the battery cell on the battery rack, starting from the reference column as 0, and increasing to both sides; a is the row direction attenuation coefficient, the parameter to be fitted, a>0; b is the column direction attenuation coefficient, the parameter to be fitted, b≥0; exp represents an exponential function with the natural constant e as the base. After fitting, the function expression and parameters are stored in the cloud server 1 for use by the digital twin thermodynamic simulation model during the running phase.

[0041] The technical advantage of this solution lies in providing a fully automated, non-manually-intervened method for calibrating spatial distribution functions, without relying on any additional specialized measuring instruments. The underlying principle is that, under uniform initial conditions of thermal equilibrium, the cooling excitation applied by the air conditioner upon startup begins synchronously for each individual unit. The difference in the rate of temperature decrease among each unit uniquely reflects the difference in its local heat transfer intensity. By capturing the relative differences in this dynamic process, the spatial heat transfer distribution can be non-destructively derived. This method is particularly suitable for existing communication base stations already in operation, eliminating the need for additional wind speed or airflow sensors on the battery racks, resulting in zero construction costs.

[0042] In a preferred embodiment, the cloud server 1 further includes a spatial distribution function failure self-checking module, specifically configured for: During normal system operation, the residual temperature time series of each battery cell within a preset self-test cycle is obtained from the residual calculation unit, and the average residual temperature of each battery cell within that cycle is calculated. All battery cells in the battery pack located in the first half of the air supply stroke are divided into the front row of cells, and all battery cells located in the second half of the air supply stroke are divided into the back row of cells. Calculate the arithmetic mean of the residual temperature values ​​of all battery cells in the front row set as the representative value of the front row residual; calculate the arithmetic mean of the residual temperature values ​​of all battery cells in the back row set as the representative value of the back row residual. When the representative value of the front row residual is less than the first preset threshold and the representative value of the back row residual is greater than the second preset threshold, the spatial distribution function is determined to be invalid, and the spatial distribution function calibration module is triggered to re-execute the calibration process; wherein, the first preset threshold is a negative value and the second preset threshold is a positive value.

[0043] Specifically, the spatial distribution function failure self-check module addresses the issue of calibrated spatial distribution functions failing during long-term system operation due to changes in air conditioning airflow direction or organization caused by human or natural factors. Typical scenarios leading to failure include: maintenance personnel moving battery racks for equipment repair, changes in the louver angle of air outlets after cleaning or repairing air conditioners, and the addition or removal of other heat-generating equipment in the server room altering the airflow path. Once the airflow organization changes, the previously calibrated spatial distribution of the heat transfer coefficient will no longer conform to physical reality, and continued use will lead to systematic deviations in the expected temperature calculation. To solve this problem, the cloud server 1 includes a spatial distribution function failure self-check module. This module runs continuously in the background during normal system operation, obtaining the residual temperature time series of each battery cell within a preset self-check cycle, such as every 30 minutes or hour, from the residual calculation unit. For each cell, the self-check module calculates the arithmetic mean of the residual temperature values ​​of all sampling points within that cycle, using this as the average residual temperature of that cell within that cycle.

[0044] Subsequently, the self-test module divides all individual cells in the battery pack into two sets based on their spatial location. The division is based on the relative position of the cells within the airflow path: all battery cells in the first half of the airflow path are grouped into the front row set, and all battery cells in the second half are grouped into the rear row set. Specific division rules can include sorting by the Euclidean distance of the cells from the air conditioning vent, with the first 50% of cells having the smallest distances forming the front row set and the last 50% forming the rear row set; alternatively, it can be based on the number of rows and columns of the battery rack. For example, if there are six rows of batteries, the first to third rows form the front row set, and the fourth to sixth rows form the rear row set. After completing the set division, the self-test module calculates the arithmetic mean of the residual temperature values ​​of all cells in the front row set, using this value as the representative residual value for the front row. Similarly, it calculates the arithmetic mean of the residual temperature values ​​of the rear row set, using this value as the representative residual value for the rear row. These two representative values ​​reflect the overall level and polarity of the residuals in the front and rear row regions, respectively.

[0045] Finally, the failure determination logic is entered. The self-test module compares the representative value of the front row residual with a preset first threshold, and the representative value of the rear row residual with a preset second threshold. The first preset threshold is set to a negative value, for example, -0.3 degrees Celsius; the second preset threshold is set to a positive value, for example, 0.3 degrees Celsius. When both conditions are met—the representative value of the front row residual being less than the first preset threshold and the representative value of the rear row residual being greater than the second preset threshold—the spatial distribution function is determined to have failed. The physical basis for the above determination condition is that the previously calibrated spatial distribution function is based on a basic assumption: the airflow decreases unidirectionally from the front row to the rear row along the battery pack arrangement direction, and the heat transfer coefficient of the front row is higher than that of the rear row. When this assumption matches physical reality, the residual should exhibit a random distribution without spatial aggregation. However, when the airflow direction or airflow organization changes significantly, causing the originally set front row area to no longer be on the windward side of the airflow, while the original rear row area receives stronger forced convection, the spatial distribution function becomes mismatched with the current situation. In this mismatch state, the digital twin thermodynamic simulation model still assigns a high heat transfer coefficient to the front row of units according to the original function, underestimating their true temperature. This results in the expected surface temperature being lower than the measured surface temperature, with the residual being positive. Simultaneously, the model assigns a low heat transfer coefficient to the rear row of units, overestimating their true temperature. This results in the expected surface temperature being higher than the measured surface temperature, with the residual being negative. However, in this embodiment, the sign of the threshold condition is set so that the residual value for the front row is negative and the residual value for the rear row is positive. This corresponds to another typical case of spatial distribution function failure: the airflow direction is not completely reversed, but rather deflected or dispersed. This causes the heat transfer capacity of the original front row area to be significantly enhanced compared to the model's expectation, while the heat transfer capacity of the original rear row area is significantly weakened compared to the model's expectation. For example, when the louvers of an air conditioning vent are adjusted to deflect upwards or downwards, causing the cooling energy to concentrate on one side of the battery rack instead of being delivered from the front, some of the front-row cells on that side may experience abnormally enhanced localized heat transfer, resulting in an actual temperature lower than the model's prediction and a negative residual. Conversely, the rear-row cells in areas where the airflow is blocked or bypassed experience weaker heat transfer than the model's prediction, resulting in an actual temperature higher than the model's prediction and a positive residual. This spatial polarity reversal residual distribution pattern is a unique characteristic of spatial distribution function failure and is fundamentally different from the individual cell residuals caused by internal faults in the battery cells.

[0046] Once a failure is detected, the self-test module immediately triggers the spatial distribution function calibration module to re-execute the entire calibration process described above, obtaining a new spatial distribution function that matches the current airflow organization. This recalibration process is fully automated and requires no manual intervention.

[0047] The technical advantage of this solution lies in endowing the digital twin system with the ability to autonomously perceive and self-correct environmental changes. The underlying principle is that the failure of the spatial distribution function leads to systematic residuals with opposite polarities and spatial clustering in different regions, a spatial distribution pattern that is fundamentally different from the individual residuals caused by internal battery faults. By monitoring the macroscopic statistical characteristics of the residuals in the spatial dimension, the system can accurately distinguish between model mismatch and equipment failure, and automatically initiate a recalibration process upon confirming model mismatch, restoring model accuracy without manual intervention. This ensures the long-term availability and robustness of the system throughout the entire lifecycle of the communication base station.

[0048] In a preferred embodiment, the cloud server 1 further includes an abnormal heat source inversion unit and a backup capability assessment unit; The abnormal heat source inversion unit is configured to, after the abnormality determination unit generates an internal abnormal heat source warning signal for the target battery cell, invert the equivalent abnormal heat generation power of the target battery cell at the current moment based on the residual temperature time series of the target battery cell. The backup capability assessment unit is configured to obtain the current state of charge of the battery pack to which the target battery cell belongs and the typical load power of the communication base station equipment room. It takes the equivalent abnormal heat generation power, the current state of charge, and the typical load power as inputs to drive the digital twin thermodynamic simulation model to perform virtual discharge simulation under a preset mains power outage scenario, obtain the estimated backup time of the target battery cell under the condition of the presence of the internal abnormal heat source, and output the backup capability assessment result based on the estimated backup time.

[0049] Specifically, the system can identify an internal abnormal heat source in a single battery cell, but only outputs a binary warning signal. Maintenance personnel, upon receiving this signal, cannot determine the severity of the anomaly or whether immediate replacement is necessary. For communication base stations located in remote areas, the overall cost of a single on-site maintenance is high. Immediate replacement for all warnings would waste maintenance resources; conversely, delaying all cases could lead to base station failures due to rapid degradation of individual batteries during power outages. To address these issues, this implementation specifies that the cloud server 1 also includes an abnormal heat source inversion unit and a backup capability assessment unit. The abnormal heat source inversion unit's function is to restore the residual temperature signal to its physical root cause, namely, the equivalent abnormal heat generation power generated inside the battery cell. Its trigger condition is that the anomaly detection unit has generated an internal abnormal heat source warning signal for a specific battery cell, which is subsequently referred to as the target battery cell. The basic principle of the inversion is based on the battery's thermal balance equation, which describes the quantitative relationship between the battery surface temperature change rate, internal heat generation power, and environmental heat transfer power. Given the known thermodynamic parameters of the battery and real-time heat exchange conditions, by analyzing the dynamic characteristics of the residual temperature time series, the intensity of the internal heat source required to generate the temperature deviation can be calculated in reverse. This intensity, measured in watts, is called the equivalent abnormal heating power. The physical meaning of this equivalent abnormal heating power is that it is a normalized fault intensity index, decoupled from specific ambient temperature and air conditioning operating conditions, reflecting only the severity of the internal battery fault itself.

[0050] The backup capability assessment unit is designed to map the fault intensity indicators obtained from the inversion into decision-making information that directly guides the operation and maintenance of communication base stations. The core function of a communication base station battery pack is to provide uninterrupted backup power during mains power outages. Therefore, the primary concern for maintenance personnel is whether a faulty battery can support the specified backup duration during the next mains power outage. The backup capability assessment unit first obtains the current state of charge (SOC) of the battery pack to which the target battery cell belongs via the edge computing gateway 2 or the data interface with the base station's environmental monitoring system. This SOC is expressed as a percentage and can be obtained through the battery management system's ampere-hour integration or open-circuit voltage lookup table. Simultaneously, the unit obtains the typical load power of the communication base station's equipment room. This parameter can be the average DC load power calculated based on the base station's main equipment's rated power and traffic volume model, or it can be the statistical value of discharge current recorded from historical mains power outage events. The backup capability assessment unit uses the aforementioned equivalent abnormal heat generation power, current SOC, and typical load power obtained from the inversion as input parameters to drive the constructed digital twin thermodynamic simulation model to perform a virtual discharge simulation under a preset mains power outage scenario. The boundary conditions for this simulation are set as follows: a mains power outage occurs, the switching power supply stops outputting, and the battery pack begins supplying power to the load. During the simulation, the digital twin model not only simulates the normal discharge voltage drop process of a battery, but also injects the effect of the target battery cell continuously consuming its own energy due to abnormal internal heating. The simulation terminates when the terminal voltage of the target battery cell drops to a preset cutoff voltage, and the virtual time length experienced during the simulation is the estimated backup duration. The backup capability assessment unit compares this estimated backup duration with the backup duration requirements specified in the operation and maintenance specifications, and outputs a clearly tiered backup capability assessment result. For example, if the estimated duration is longer than the required duration, the output indicates that the backup duration is sufficient; if the estimated duration is close to the required duration, the output indicates a risk of insufficient backup duration; and if the estimated duration is significantly shorter than the required duration, the output recommends immediate replacement.

[0051] The technical effectiveness of this solution lies in establishing a complete decision-making loop from fault detection to risk assessment, transforming abnormal temperature signals at the technical level into differentiated operation and maintenance instructions that management can directly execute. Its technical principle is based on leveraging the bidirectional capabilities of a digital twin model: forward simulation for calculating expected temperatures, reverse identification for fault intensity inversion, and further forward deduction for performance prediction under faulty conditions. This purely software-based approach completes a comprehensive rehearsal of the faulty battery's performance in real emergency scenarios, avoiding any safety risks to the online battery pack caused by actual discharge testing.

[0052] In a preferred embodiment, the abnormal heat source inversion unit is specifically configured to: Obtain the residual temperature time series of the target battery cell within a first preset time period before the warning time, and the expected surface temperature time series output by the digital twin thermodynamic simulation model within the first preset time period; Using the lumped heat capacity parameter and surface heat transfer coefficient of the target battery cell as known quantities, and the residual temperature time series as observed values, the equivalent abnormal heating power time series inside the target battery cell within the first preset time period is solved by using the heat balance equation in reverse. The equivalent abnormal heating power time series is subjected to trend fitting, and the equivalent abnormal heating power value at the current moment is extracted as the equivalent abnormal heating power obtained by inversion.

[0053] Specifically, the core of the workflow for the abnormal heat source inversion unit is to use the heat balance equation to solve the residual temperature time series piece by piece in the time domain, thereby reconstructing the complete trajectory of the equivalent abnormal heating power inside the target battery cell as a function of time.

[0054] The first step in the inversion process is data preparation. The abnormal heat source inversion unit extracts two time-series datasets from the system database. Both datasets span a preset time period before the warning time, which can be set to 30 or 60 minutes depending on the system configuration. The first dataset is the residual temperature time series of the target battery cell within this time period, generated and continuously stored by the residual calculation unit. The second dataset is the expected surface temperature time series output by the digital twin thermodynamic simulation model within the same time period, synchronously stored during normal model operation. Both datasets have a unified timestamp to ensure point-to-point alignment.

[0055] The second step in the inversion process is parameter acquisition. The abnormal heat source inversion unit reads two thermodynamic characteristic parameters of the target battery cell from the system parameter library. The first parameter is the lumped heat capacity parameter, which physically means the amount of heat absorbed by the battery cell to increase its overall temperature by one degree Celsius, measured in joules per degree Celsius. This parameter can be theoretically estimated by multiplying the mass of the battery cell by the specific heat capacity of the casing material, or it can be obtained through laboratory calibration by performing step thermal response tests on batteries of the same model. The second parameter is the surface heat transfer coefficient, which physically means the heat transfer power per unit area between the surface of the battery cell and the surrounding air per unit temperature difference, measured in watts per square meter per degree Celsius. This parameter has been dynamically determined by a digital twin thermodynamic simulation model combined with a spatial distribution function. During the inversion process, both parameters are considered known quantities.

[0056] The third step in the inversion process is the core solution. The abnormal heat source inversion unit uses the residual temperature time series as the observed value and substitutes it into the thermal balance equation of the battery cell for inverse solution. The physical form of this thermal balance equation is: the net internal heat generation power of the battery equals the battery heat capacity multiplied by the rate of temperature change, plus the surface heat transfer coefficient multiplied by the surface area, and then multiplied by the temperature difference between the battery and the air. In the application scenario of this application, the internal net heat generation power consists of two parts: one is the reference Joule heat power generated by the float charge current, which is already included in the expected surface temperature output by the digital twin model; the other is the internal equivalent abnormal heat generation power to be solved. Therefore, by substituting the residual temperature series point by point into the discretized thermal balance difference equation, the equivalent abnormal heat generation power value inside the target battery cell at each sampling time within the first preset time period can be solved, thus forming an equivalent abnormal heat generation power time series of the same length as the residual series.

[0057] The fourth step in the inversion process is trend extraction. Since the original inversion results may contain short-term fluctuations due to measurement noise or minor model errors, the abnormal heat source inversion unit performs trend fitting on the aforementioned equivalent abnormal heat generation power time series. The fitting method can be a first-order linear regression after moving average filtering, assuming that the abnormal heat generation power changes approximately linearly within the first preset time period, and determining the slope and intercept of the best-fitting line using the least squares method. After fitting, the value of the fitted line at the time corresponding to the warning moment is taken as the equivalent abnormal heat generation power value at the current moment, which is the final output of the inversion process.

[0058] The technical advantage of this solution lies in providing a fault intensity quantification method entirely based on the first principles of thermodynamics, without relying on empirical formulas or statistical regression models. The principle is that abnormal heating inside the battery macroscopically manifests as a contribution to the deviation of the surface temperature from the expected value. By solving the inverse thermal equilibrium problem describing this deviation process, the temperature-domain characteristic can be reduced to a power-domain fault intensity quantity. The advantage of this reduction process is that the equivalent abnormal heating power is a physical quantity decoupled from the thermal environment. While the temperature residual caused by the same internal fault may differ under different air conditioning operating conditions and different computer room ambient temperatures, the corresponding equivalent abnormal heating power remains stable. Therefore, this index can be used for horizontal comparisons and trend tracking between different sites and seasons.

[0059] In a preferred embodiment, the backup capability assessment unit is specifically configured to: Obtain the preset mains power interruption backup time requirement of the battery pack to which the target battery cell belongs; The equivalent abnormal heat generation power value at the current moment, the average growth rate of the equivalent abnormal heat generation power within the second preset time period, the current state of charge, and the typical load power are used as boundary conditions to drive the digital twin thermodynamic simulation model to perform a virtual discharge simulation. The virtual discharge simulation simulates the time it takes for the terminal voltage of the target battery cell to drop to the cutoff voltage due to the continuous evolution of internal abnormal heat generation under a preset mains power interruption scenario. This time period is used as the estimated backup time. The estimated backup time is compared with the preset backup time requirement for mains power outage, and the backup capability assessment result of the target battery cell is output. The backup capability assessment result includes at least one of the following levels: the backup time can be met, there is a risk of insufficient backup time, and it is recommended to replace it immediately.

[0060] Specifically, the first step in the evaluation process is parameter acquisition. The backup capability evaluation unit reads the preset mains power outage backup duration requirements of the battery pack to which the target battery cell belongs from the communication base station's operation and maintenance configuration database. These requirements are usually determined by the communication operator based on the base station level and grid reliability. For example, core node base stations require a backup duration of no less than three hours, ordinary macro base stations require no less than one hour, and micro base stations require no less than thirty minutes. This parameter serves as the benchmark for the evaluation.

[0061] The second step in the evaluation process is boundary condition construction. The backup capability assessment unit collects and organizes the following four boundary condition parameters. The first parameter is the equivalent abnormal heat generation power value at the current moment, which is provided by the abnormal heat source inversion unit. The second parameter is the average growth rate of the equivalent abnormal heat generation power over a second preset time period. This second preset time period is typically set to a longer time span than the first preset time period, such as six or twelve hours, to capture the mid-term trend of fault evolution. The average growth rate is calculated by obtaining the equivalent abnormal heat generation power time series obtained from the inversion over the second preset time period, extracting its slope through linear fitting, and expressing it in watts per hour. The third parameter is the current state of charge, the method of which has been described in claim 5. The fourth parameter is the typical load power, the method of which has been described.

[0062] The third step in the evaluation process is the execution of a virtual discharge simulation. The backup capability evaluation unit calls the digital twin thermodynamic simulation model and switches its operating mode from real-time simulation to simulation mode. In simulation mode, the model's input is no longer the real-time collected air conditioner operating status parameters, but a set of preset virtual scenario boundary conditions. Specifically, the model is set to simulate a preset mains power outage scenario: at the initial moment after the mains power outage, the battery pack begins constant current discharge at a typical load power, and the terminal voltage of the target battery cell is calculated in real time based on its state of charge and discharge current. At the same time, the model injects an internal abnormal heating power term that increases over time into the thermodynamic equation of the target battery cell. The initial value of this term is the equivalent abnormal heating power value at the current moment, and its growth slope is set to the aforementioned average growth rate. As virtual time progresses, this internal abnormal heating continuously consumes the battery's active material and charge, manifested as an accelerated drop in terminal voltage. When the calculated terminal voltage value drops to the preset discharge cutoff voltage, such as 1.8 volts for a lead-acid battery cell, the virtual discharge simulation terminates. The virtual time elapsed from the start to the end of the simulation is the estimated backup time of the target battery cell under the current condition of an abnormal internal heat source.

[0063] The fourth step in the assessment process is result comparison and tiered output. The backup capability assessment unit compares the estimated backup duration with the preset mains power interruption backup duration requirement obtained in the first step. The comparison result is mapped to an assessment conclusion with at least three levels. The first level indicates that the backup duration is met. When the estimated backup duration is greater than or equal to the backup duration requirement multiplied by a preset safety factor, such as 1.2, this conclusion is output, indicating that the current fault does not pose a substantial threat to backup support capability, and the unit can be included in the next routine inspection plan. The second level indicates a risk of insufficient backup duration. When the estimated backup duration is between the backup duration requirement and the backup duration requirement multiplied by the safety factor, this conclusion is output, indicating that the fault has begun to affect backup capability, and it is recommended to pay close attention and prepare spare parts in advance. The third level recommends immediate replacement. When the estimated backup duration is less than the backup duration requirement, this conclusion is output, indicating that the faulty state of the unit can no longer meet basic backup support needs, and a work order should be dispatched immediately for replacement.

[0064] The technical advantage of this solution lies in transforming abstract fault power values ​​into concrete backup time losses, enabling a quantifiable assessment of the fault's impact. The underlying principle is that the actual performance of a battery during a mains power outage is the result of both its initial energy storage state and internal losses during discharge. Abnormal internal heating, acting as a continuous parasitic load, not only consumes effective energy storage during discharge but also accelerates side reactions due to increased temperature, creating a vicious cycle. By faithfully reproducing this physical process in a digital twin model, the system can predict the actual performance of a faulty battery under extreme scenarios without conducting a real discharge, providing a highly reliable quantitative basis for operational and maintenance decisions.

[0065] In a preferred embodiment, the abnormal heat source inversion unit is further configured for; After obtaining the residual temperature time series within a first preset time period before the warning time and before performing the reverse solution of the heat balance equation, the residual temperature time series is purified. The purification process includes: Obtain the time series of battery pack float charging current collected by the edge computing gateway 2 within the first preset time period; The floating charge current time series is input into the digital twin thermodynamic simulation model, which drives the digital twin thermodynamic simulation model to calculate the time series of the floating charge current temperature change component of the target battery cell under the floating charge current fluctuation condition and the assumption of no internal fault. Subtracting the float charge current-induced temperature variation component time series from the residual temperature time series yields the purified residual temperature time series. The abnormal heat source inversion unit uses the purified residual temperature time series as the observed value to perform the reverse solution of the heat balance equation.

[0066] Specifically, the core improvement lies in addressing the issue of potential interference from floating charge current fluctuations in the residual temperature time series before the abnormal heat source inversion unit performs the inverse solution of the thermal balance equation. In the inversion process, the abnormal heat source inversion unit directly substitutes the residual temperature time series as observed values ​​into the thermal balance equation for inverse solution, implicitly assuming that the residual temperature is entirely contributed by abnormal heating within the battery. However, in the actual operating environment of a communication base station, the switching power supply continuously and slightly adjusts the floating charge current based on load changes, grid fluctuations, and the temperature compensation characteristics of the battery pack's floating charge voltage to maintain bus voltage stability. Although these floating charge current fluctuations are small, typically only at the milliampere level, the resulting Joule heat power changes directly affect individual battery cells, generating temperature disturbance components that may overlap with internal abnormal heating in the time domain. If this disturbance component is not removed, the equivalent abnormal heating power time series obtained through inversion will be superimposed with operating noise unrelated to the fault, thus affecting the accuracy of the virtual discharge simulation. To address this issue, the abnormal heat source inversion unit is configured to perform purification processing on the residual temperature time series. This purification processing occurs after the residual temperature time series within a first preset duration prior to the acquisition of the warning time and before the inverse solution of the heat balance equation, serving as an independent signal preprocessing step.

[0067] The first step in the purification process is to acquire float charge current data. The abnormal heat source inversion unit extracts the battery pack float charge current time series that corresponds exactly to the first preset duration from the historical data cache uploaded by the edge computing gateway 2. This float charge current time series is acquired by a Hall current sensor or shunt deployed on the main circuit of the battery pack in conjunction with a high-precision analog-to-digital converter. The sampling frequency is consistent with the temperature sensor sampling frequency, for example, once per second, and each data record has a unified timestamp synchronized with the temperature data.

[0068] The second step in the purification process is to simulate and calculate the temperature variation component caused by the float charge current. The abnormal heat source inversion unit takes the aforementioned float charge current time series as input and drives the digital twin thermodynamic simulation model point by point. In this simulation, the model adopts the same assumption of no internal faults as before, that is, it assumes that the target battery cell only has reference Joule heat generated by the float charge current flowing through its internal resistance, and there is no abnormal internal heating. Based on the instantaneous value of the float charge current at each sampling time, the model calculates the corresponding Joule heat power using Joule's law, and then simulates the surface temperature fluctuation component of the target battery cell due to the change of float charge current under the conditions of float charge current fluctuation and the assumption of no internal faults, using the lumped heat capacity parameter and surface heat transfer coefficient of the battery cell. This component is called the float charge current-induced temperature variation component time series, and its physical meaning is: even if the battery is completely healthy, its surface temperature will fluctuate slightly due to the fluctuation of the float charge current itself.

[0069] The third step in the purification process is residual purification. The abnormal heat source inversion unit subtracts the float charge current-induced temperature variation component time series calculated in the second step from the original acquired residual temperature time series, point by point. Since the original residual temperature time series contains contributions from both internal abnormal heating and float charge current fluctuations, subtracting the latter leaves a pure temperature deviation signal contributed only by internal abnormal heating. This signal is called the purified residual temperature time series. After purification, the abnormal heat source inversion unit uses this purified residual temperature time series as the new observation value, replacing the original residual temperature time series, and continues to execute the inverse solution process of the heat balance equation.

[0070] The technical effect of this solution is that it eliminates the interference of operating condition disturbances on fault feature extraction at the signal input end, and improves the signal-to-noise ratio and accuracy of equivalent abnormal heating power inversion.

[0071] In a preferred embodiment, the backup capability assessment unit is further configured to perform the following operations before executing the virtual discharge simulation: Determine whether the equivalent abnormal heating power at the current moment is lower than a preset power threshold; When it is determined that the equivalent abnormal heating power is lower than the power threshold, a preset conservative growth rate is used instead of the average growth rate of the equivalent abnormal heating power within a second preset time period as the boundary condition for the virtual discharge simulation. The preset conservative growth rate is a growth rate value with a preset upper limit of confidence level, extracted by statistical analysis of the historical evolution data of abnormal heat generation power of the same type of battery under micro short-circuit fault mode.

[0072] Specifically, the core improvement lies in addressing the insufficient reliability of the average growth rate parameter in the early stages of a fault during virtual discharge simulation. In the evaluation process, the backup capability assessment unit uses the average growth rate of the equivalent abnormal heating power over a second preset time period as a boundary condition to drive the digital twin model to simulate the continuous evolution of the fault during future discharge. However, in the early stages of internal abnormal heating, the absolute value of the equivalent abnormal heating power is small, typically at the milliwatt level. At this point, the fluctuations in the power estimate caused by sensor background noise, data acquisition quantization errors, and minor environmental disturbances in the derived equivalent abnormal heating power time series may be comparable to or even larger than the actual growth rate in relative magnitude. If a linear fit is directly applied to this low signal-to-noise ratio data to extract the average growth rate, the result is easily contaminated by noise, leading to an overestimation or underestimation of the growth rate. This results in a significant deviation between the estimated backup time given by the virtual discharge simulation and the actual backup time, affecting the reliability of the evaluation conclusion.

[0073] To address this issue, the backup capability assessment unit is required to perform a branching decision on the processing strategy for the growth rate parameter before executing the virtual discharge simulation. The core idea of ​​this strategy is: when the fault intensity is insufficient to overcome the noise floor, the pursuit of the growth rate in the measured data is abandoned, and instead a conservative preset value based on statistical safety principles is adopted.

[0074] The backup capability assessment unit first performs a power threshold judgment. It obtains the equivalent abnormal heat generation power value at the current moment from the abnormal heat source inversion unit and compares this value with a preset power threshold. This power threshold is determined based on the upper limit of the fluctuation range of the estimated equivalent abnormal heat generation power caused by various non-fault factors in a healthy state for the same type of battery, and multiplied by a safety factor, for example, setting the power threshold to 50 milliwatts. When the current equivalent abnormal heat generation power value is determined to be lower than this power threshold, it indicates that the fault is still in its early stages and the signal-to-noise ratio is insufficient.

[0075] When the power level is determined to be below the threshold, the backup capability assessment unit does not use the average growth rate fitted from the measured data within the second preset time period. Instead, it retrieves a preset conservative growth rate from the system parameter library and uses this conservative growth rate instead of the measured average growth rate as the boundary condition for virtual discharge simulation. The method for determining this conservative growth rate is as follows: During the system development phase, a large number of battery samples of the same model undergo accelerated aging tests under micro-short-circuit fault mode in the laboratory. By artificially introducing a controllable micro-short-circuit fault, the evolution data of abnormal heat generation power is continuously monitored and recorded throughout the entire process from fault initiation to complete failure. Statistical analysis is performed on the growth rate data of all collected samples at each evolution stage, and a preset upper quantile value of its probability distribution is extracted, for example, the upper limit value at a 95% confidence level. The economic meaning of this upper limit value is that, in historical test samples, only 5% of the faulty batteries have an abnormal heat generation power growth rate exceeding this value. Using this value as the conservative growth rate can cover the worst evolution scenario with a high probability. This conservative growth rate is pre-stored in the system parameter library in watts per hour.

[0076] The technical advantage of this solution is that, in the early stage of a fault when the signal-to-noise ratio is insufficient, a conservative safety assessment is used instead of an unreliable precise estimate. This ensures that the backup capability assessment conclusion given by the system at this stage has sufficient safety margin and avoids overly optimistic backup duration estimates due to underestimating the speed of fault evolution.

[0077] In a preferred embodiment, the backup capability assessment unit is further configured to perform the following operations before executing the virtual discharge simulation: Determine whether the value of the equivalent abnormal heating power at the current moment reaches or exceeds a preset power threshold; When it is determined that the equivalent abnormal heating power reaches or exceeds the power threshold, the average growth rate of the equivalent abnormal heating power within a second preset time period is nonlinearly compensated and corrected. The nonlinear compensation correction includes: Obtain the second-order difference statistics of the time series of the equivalent abnormal heating power within the second preset time period, and determine whether the evolution of the equivalent abnormal heating power shows an accelerated characteristic based on the second-order difference statistics. When acceleration characteristics are determined, the acceleration factor is determined based on the value of the second-order difference statistic, and the average growth rate is multiplied by the acceleration factor as the boundary condition for the virtual discharge simulation. When it is determined that no acceleration characteristics are present, the average growth rate is directly used as the boundary condition for the virtual discharge simulation.

[0078] Specifically, when the equivalent abnormal heat generation power reaches or exceeds a preset power threshold, the fault has developed to a certain scale, and the time series of the equivalent abnormal heat generation power obtained by inversion has a high signal-to-noise ratio. At this point, the average growth rate extracted from the measured data is reliable. However, as the fault further intensifies, the physicochemical processes such as dendrite growth at the micro-short circuit point and partition melting may enter a positive feedback acceleration stage. The evolution of the equivalent abnormal heat generation power no longer follows the early approximate linear law, but exhibits nonlinear acceleration characteristics. If the linear average growth rate is still directly substituted into the virtual discharge simulation and assumed to remain constant throughout the simulation, the rate of fault deterioration in the later stages of the simulation will be systematically underestimated, leading to an overly optimistic estimate of the backup duration. To solve this problem, the backup capability assessment unit is limited to performing nonlinear compensation correction on the average growth rate after determining that the equivalent abnormal heat generation power has reached or exceeded the power threshold.

[0079] The backup capability assessment unit first determines whether the equivalent abnormal heat generation power at the current moment reaches or exceeds a preset power threshold. This power threshold is the same parameter as previously mentioned. When it is determined that the threshold has been reached or exceeded, a nonlinear compensation correction process is triggered.

[0080] The first step in nonlinear compensation correction is to obtain the second-order difference statistics. The backup capability assessment unit retrieves the equivalent abnormal heating power time series obtained from the inversion within the second preset time period. For this time series, its first-order difference sequence is first calculated, which is the change in power value between each adjacent sampling point. The first-order difference sequence reflects the change in the growth rate of abnormal heating power over time. Subsequently, the difference is calculated again for the first-order difference sequence to obtain the second-order difference sequence. Each value in the second-order difference sequence represents the change in the growth rate itself per unit time, and its unit is watts per square hour. The second-order difference statistics can be the arithmetic mean of the sequence within the second preset time period, or its weighted average or median.

[0081] The second step in nonlinear compensation correction is acceleration characteristic determination. The backup capability assessment unit determines whether the evolution of the equivalent abnormal heat generation power exhibits acceleration characteristics based on the aforementioned second-order difference statistics. The determination rule is as follows: when the value of the second-order difference statistics is significantly greater than zero, for example, greater than 0.01 watts per square hour, and this positive value is statistically significantly different from zero, the evolution is determined to exhibit acceleration characteristics; when the value of the second-order difference statistics is close to zero or is negative, it is determined that no acceleration characteristics are exhibited.

[0082] When acceleration characteristics are identified, the backup capability assessment unit determines the acceleration factor based on the value of the second-order difference statistic. A specific method for determining this factor is as follows: The average value of the first-order difference sequence within a second preset time period is calculated as the baseline growth rate. The second-order difference statistic is divided by this baseline growth rate, and the resulting ratio is the acceleration per unit growth rate. This ratio is then multiplied by the estimated duration of the virtual discharge simulation to obtain the acceleration factor. For example, if the estimated duration is approximately two hours and the acceleration per unit growth rate is 0.2, the acceleration factor is 1.4. The average growth rate obtained from the original linear fitting is then multiplied by this acceleration factor, and the product is used as the corrected growth rate, which is then substituted into the boundary conditions of the virtual discharge simulation.

[0083] When no acceleration characteristics are observed, the backup capability assessment unit does not correct the average growth rate and directly uses the average growth rate obtained from linear fitting as the boundary condition for virtual discharge simulation.

[0084] The technical advantage of this solution is that when a fault enters the accelerated evolution stage, it can automatically identify the acceleration trend and dynamically compensate the inference parameters, making the estimated backup time closer to the real physical process.

[0085] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A digital-twin-based remote intelligent detection system for a communication base station machine room, characterized in that, This includes a cloud server (1) and an edge computing gateway (2) deployed in the communication base station equipment room; among which, The cloud server (1) runs a digital twin thermodynamic simulation model. The digital twin thermodynamic simulation model is a simulation model that couples the operating status parameters of the computer room air conditioner with the thermodynamic parameters of the battery cell. It is used to calculate and output the expected surface temperature value of each battery cell under the preset assumption of no internal faults, with the real-time obtained operating status parameters of the computer room air conditioner as input. The assumption of no internal faults is that the battery cell only has Joule heating of the float charge current and no internal abnormal heating. The edge computing gateway (2) is used for: Real-time acquisition of the operating status parameters of the computer room air conditioner, including the start / stop status signal of the air conditioner compressor; Real-time surface temperature values ​​are collected from temperature sensors deployed on the surface of each battery cell. The operating status parameters of the computer room air conditioner and the real-time surface temperature value are uploaded to the cloud server (1) after being stamped with a unified timestamp. The cloud server (1) also includes a residual calculation unit and an anomaly detection unit; The residual calculation unit is used to calculate the difference between the real-time surface temperature value and the expected surface temperature value at the same timestamp for each battery cell, and generate a residual temperature time series. The anomaly detection unit is used to perform the following detection logic on the residual temperature time series: A trend test is performed on the residual temperature time series to determine whether it exhibits a statistically significant monotonically increasing trend. Calculate the correlation between the residual temperature time series and the start / stop status signal of the air conditioner compressor, and determine whether the two exhibit statistical decoupling; When both the monotonically increasing trend and the statistical decoupling conditions are met simultaneously, an early warning signal for the internal abnormal heat source of the corresponding battery cell is generated. The digital twin thermodynamic simulation model is embedded with a spatial distribution function of the heat transfer coefficient. The spatial distribution function is used to characterize the spatial difference in the local heat transfer capacity of each battery cell caused by the attenuation of the airflow from the computer room air conditioner along the battery pack arrangement direction. The cloud server (1) also includes a spatial distribution function calibration module. The spatial distribution function calibration module is used to extract the temperature response rate difference of each battery cell to the same air conditioner status change event by using the dynamic response data of the surface temperature of each battery cell collected by the edge computing gateway (2) before and after the air conditioner operation status change event, and to calibrate the parameters of the spatial distribution function based on the temperature response rate difference. When calculating the expected surface temperature of each battery cell, the digital twin thermodynamic simulation model determines the corresponding heat transfer coefficient based on the spatial location of the battery cell in the battery pack using the calibrated spatial distribution function, and then uses the heat transfer coefficient to calculate the expected surface temperature.

2. The digital-twin-based remote intelligent detection system for a communication base station room, according to claim 1, characterized in that, The spatial distribution function calibration module is specifically configured to: After the air conditioner compressor is in a continuous shutdown state and the surface temperature of each battery cell reaches thermal equilibrium, the edge computing gateway (2) is controlled to trigger the air conditioner compressor to start and collect data on the process of the decrease in surface temperature of each battery cell after the compressor starts. The temperature drop data of each battery cell are fitted to extract the characteristic temperature drop rate of each battery cell. The characteristic temperature drop rate of the battery cell at the starting point of the air supply airflow is selected as the reference rate. The ratio of the characteristic temperature drop rate of the remaining battery cells to the reference rate is calculated, and this ratio is used as the heat transfer coefficient attenuation factor for the corresponding spatial location. Using the row and column position coordinates of the battery cells on the battery rack as independent variables and the heat transfer coefficient attenuation factor as the dependent variable, the functional expression and parameters of the spatial distribution function are obtained by fitting.

3. The digital-twin-based remote intelligent detection system for a communication base station room, according to claim 2, wherein, The cloud server (1) also includes a spatial distribution function failure self-check module, specifically configured for: During normal system operation, the residual temperature time series of each battery cell within a preset self-test cycle is obtained from the residual calculation unit, and the average residual temperature of each battery cell within that cycle is calculated. All battery cells in the battery pack located in the first half of the air supply stroke are divided into the front row of cells, and all battery cells located in the second half of the air supply stroke are divided into the back row of cells. Calculate the arithmetic mean of the residual temperature values ​​of all battery cells in the front row set, and use it as the representative value of the front row residual. Calculate the arithmetic mean of the residual temperature values ​​of all battery cells in the rear row set, and use it as the representative value of the rear row residual. When the representative value of the front row residual is less than the first preset threshold and the representative value of the back row residual is greater than the second preset threshold, the spatial distribution function is determined to be invalid, and the spatial distribution function calibration module is triggered to re-execute the calibration process; wherein, the first preset threshold is a negative value and the second preset threshold is a positive value.

4. The remote intelligent detection system for communication base station equipment rooms based on digital twins according to claim 1, characterized in that, The cloud server (1) also includes an abnormal heat source inversion unit and a backup capability assessment unit; The abnormal heat source inversion unit is configured to, after the abnormality determination unit generates an internal abnormal heat source warning signal for the target battery cell, invert the equivalent abnormal heat generation power of the target battery cell at the current moment based on the residual temperature time series of the target battery cell. The backup capability assessment unit is configured to obtain the current state of charge of the battery pack to which the target battery cell belongs and the typical load power of the communication base station equipment room. It takes the equivalent abnormal heat generation power, the current state of charge, and the typical load power as inputs to drive the digital twin thermodynamic simulation model to perform virtual discharge simulation under a preset mains power outage scenario, obtain the estimated backup time of the target battery cell under the condition of the presence of the internal abnormal heat source, and output the backup capability assessment result based on the estimated backup time.

5. The remote intelligent detection system for communication base station equipment rooms based on digital twins according to claim 4, characterized in that, The abnormal heat source inversion unit is specifically configured for: Obtain the residual temperature time series of the target battery cell within a first preset time period before the warning time, and the expected surface temperature time series output by the digital twin thermodynamic simulation model within the first preset time period; Using the lumped heat capacity parameter and surface heat transfer coefficient of the target battery cell as known quantities, and the residual temperature time series as observed values, the equivalent abnormal heating power time series inside the target battery cell within the first preset time period is solved by using the heat balance equation in reverse. The equivalent abnormal heating power time series is subjected to trend fitting, and the equivalent abnormal heating power value at the current moment is extracted as the equivalent abnormal heating power obtained by inversion.

6. The remote intelligent detection system for communication base station equipment rooms based on digital twins according to claim 4, characterized in that, The backup capability assessment unit is specifically configured for: Obtain the preset mains power interruption backup time requirement of the battery pack to which the target battery cell belongs; The equivalent abnormal heat generation power value at the current moment, the average growth rate of the equivalent abnormal heat generation power within the second preset time period, the current state of charge, and the typical load power are used as boundary conditions to drive the digital twin thermodynamic simulation model to perform a virtual discharge simulation. The virtual discharge simulation simulates the time it takes for the terminal voltage of the target battery cell to drop to the cutoff voltage due to the continuous evolution of internal abnormal heat generation under a preset mains power interruption scenario. This time period is used as the estimated backup time. The estimated backup time is compared with the preset backup time requirement for mains power outage, and the backup capability assessment result of the target battery cell is output. The backup capability assessment result includes at least one of the following levels: the backup time can be met, there is a risk of insufficient backup time, and it is recommended to replace it immediately.

7. The remote intelligent detection system for communication base station equipment rooms based on digital twins according to claim 5, characterized in that, The abnormal heat source inversion unit is also configured to: After obtaining the residual temperature time series within a first preset time period before the warning time and before performing the reverse solution of the heat balance equation, the residual temperature time series is purified. The purification process includes: Obtain the time series of battery pack float charging current collected by the edge computing gateway (2) within the first preset time period; The floating charge current time series is input into the digital twin thermodynamic simulation model, which drives the digital twin thermodynamic simulation model to calculate the time series of the floating charge current temperature change component of the target battery cell under the floating charge current fluctuation condition and the assumption of no internal fault. Subtracting the float charge current-induced temperature variation component time series from the residual temperature time series yields the purified residual temperature time series. The abnormal heat source inversion unit uses the purified residual temperature time series as the observed value to perform the reverse solution of the heat balance equation.

8. The remote intelligent detection system for communication base station equipment rooms based on digital twins according to claim 6, characterized in that, The backup capability assessment unit is also configured to perform the following operations before executing the virtual discharge simulation: Determine whether the equivalent abnormal heating power at the current moment is lower than a preset power threshold; When it is determined that the equivalent abnormal heating power is lower than the power threshold, a preset conservative growth rate is used instead of the average growth rate of the equivalent abnormal heating power within a second preset time period as the boundary condition for the virtual discharge simulation. The preset conservative growth rate is a growth rate value with a preset upper limit of confidence level, extracted by statistical analysis of the historical evolution data of abnormal heat generation power of the same type of battery under micro short-circuit fault mode.

9. The remote intelligent detection system for communication base station equipment rooms based on digital twins according to claim 6, characterized in that, The backup capability assessment unit is also configured to perform the following operations before executing the virtual discharge simulation: Determine whether the value of the equivalent abnormal heating power at the current moment reaches or exceeds a preset power threshold; When it is determined that the equivalent abnormal heating power reaches or exceeds the power threshold, the average growth rate of the equivalent abnormal heating power within a second preset time period is nonlinearly compensated and corrected. The nonlinear compensation correction includes: Obtain the second-order difference statistics of the time series of the equivalent abnormal heating power within the second preset time period, and determine whether the evolution of the equivalent abnormal heating power shows an accelerated characteristic based on the second-order difference statistics. When acceleration characteristics are determined, the acceleration factor is determined based on the value of the second-order difference statistic, and the average growth rate is multiplied by the acceleration factor as the boundary condition for the virtual discharge simulation. When it is determined that no acceleration characteristics are present, the average growth rate is directly used as the boundary condition for the virtual discharge simulation.

Images