Data center battery health prediction method based on digital twinning

Abstract

The present application relates to the technical field of intelligent operation and maintenance, in particular to a data center battery health prediction method based on digital twinning. In the present application, a three-dimensional space attribute matrix is constructed based on multi-dimensional operating parameters and time sequence labels are continuously replaced, a dynamic association structure of high mapping between physical entities and virtual space is established, the evolution track of internal physical parameters of UPS battery is accurately copied, multi-feature normalization processing and time sequence neural network are combined to deeply mine deep space-time features of capacity degradation, the prediction sequence is bottomed out and constrained by extracting the theoretical decay lower limit, the life evaluation deviation caused by environmental and sudden load interference is effectively overcome, the collaborative mapping relationship between internal resistance and temperature fluctuation is deeply analyzed to divide the risk level, the simulation load flow is actively brought into the virtual space to pre-play the heat accumulation limit condition, the hysteresis disadvantage of static threshold judgment is effectively eliminated, the fundamental change from passive alarm to active predictive prevention is realized, and the long-term stability of the data center power supply base is effectively ensured.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology, and in particular to a data center battery health prediction method based on digital twins. Background Technology

[0002] The field of intelligent operation and maintenance technology involves the use of information processing technology to monitor, analyze and manage the operating status of various equipment and systems. It mainly includes data acquisition, status assessment, fault diagnosis and operation and maintenance decision support. Its purpose is to achieve full life cycle management of equipment through the processing of operating data. It is widely used in scenarios such as data centers, power systems, industrial equipment and communication infrastructure.

[0003] Among them, the data center battery health prediction method based on digital twins refers to a method for analyzing and evaluating the operating status of data center UPS battery systems. Traditional data center battery health prediction methods refer to evaluating the UPS battery status through historical operating data, threshold judgment, or statistical analysis. This method usually collects operating parameters such as UPS battery voltage, current, and temperature, and combines model analysis or rule judgment to predict and manage the health status of UPS batteries.

[0004] Traditional data center battery health prediction technologies rely excessively on static threshold judgments and statistical induction of discrete data. They lack the ability to dynamically correlate multi-dimensional physical states in the face of complex and ever-changing power loads in computer rooms. Simply extracting environmental and operating parameters often ignores the spatiotemporal coupling reactions caused by plate aging and uneven internal heat distribution. Under long-term operating conditions, they are easily affected by sudden load shocks and temperature drift, resulting in assessment biases. The one-way monitoring logic makes it difficult for the system to capture signs of irreversible physical capacity degradation inside the UPS battery. It can only make lagging warnings based on existing phenomena. It cannot predict the trend of heat generation and performance degradation in advance for future extreme operating conditions, which can easily lead to the failure to report potential thermal runaway hazards and serious distortion of lifespan estimates, thereby triggering the risk of backup power outage. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a data center battery health prediction method based on digital twins.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a data center battery health prediction method based on digital twins, comprising the following steps: S1: During the operation of the data center physical server room, acquire the voltage, temperature and charging / discharging current of individual UPS batteries, compare the corresponding thresholds, and extract the set of operating parameters. S2: Analyze the numerical change process of the UPS battery voltage, UPS battery temperature, and UPS battery charging and discharging current in the set of operating parameters within a preset three-dimensional geometry, and construct a battery digital twin; S3: Based on the UPS battery voltage, UPS battery temperature, and UPS battery charging and discharging current during the historical time period of the battery digital twin, construct a theoretical remaining capacity lower limit sequence and a time-series operation feature matrix. Input the time-series operation feature matrix into a long short-term memory neural network model to determine the predicted remaining capacity sequence. Based on the theoretical remaining capacity lower limit sequence and the predicted remaining capacity sequence, generate the quantitative result of the remaining service life of the UPS battery. S4: Obtain the relative voltage fluctuation of UPS battery voltage and UPS battery charging and discharging current, analyze the internal resistance of UPS battery, classify fault risk level labels based on the current UPS battery temperature, and determine the UPS battery health status early warning assessment level. S5: Based on the UPS battery health status early warning assessment level, set the simulated charging and discharging current parameters in the battery digital twin, determine the simulated operating temperature variable, and combine the quantification results of the remaining service life of the UPS battery to obtain the UPS battery health prediction result.

[0007] As a further aspect of the present invention, the set of operating parameters includes UPS battery voltage extraction, UPS battery temperature extraction, and UPS battery charge / discharge current retention. The battery digital twin includes a three-dimensional geometric entity architecture, internal attributes of UPS battery nodes, and time-series value change records. The quantification result of the remaining service life of the UPS battery is specifically calculated by the time difference between the predicted capacity termination time and the current time. The UPS battery health status early warning assessment level is specifically calculated by the internal resistance conversion score and fault risk classification label of the equipment. The UPS battery health prediction result includes the simulation operating temperature comparison conclusion, the digital twin simulation verification status, and the equipment pre-intervention maintenance instructions.

[0008] As a further aspect of the present invention, the step of obtaining the set of operating parameters specifically includes: S111: Obtain the voltage of a single UPS battery during the operation of the data center physical computer room, set a first voltage comparison threshold and a second voltage comparison threshold, and extract the UPS battery voltage between the first voltage comparison threshold and the second voltage comparison threshold; S112: Obtain the temperature of individual UPS batteries during the operation of the data center physical computer room, set the upper limit threshold for temperature safety, and extract the UPS battery temperatures that are lower than the upper limit threshold for temperature safety. S113: Synchronously acquire the charging and discharging current of a single UPS battery, integrate the UPS battery voltage, the UPS battery temperature, and the UPS battery charging and discharging current to generate a set of operating parameters.

[0009] As a further aspect of the present invention, the step of obtaining the battery digital twin specifically includes: S211: Extract the UPS battery voltage, UPS battery temperature, and UPS battery charging and discharging current from the set of operating parameters, construct a three-dimensional geometry, and input the UPS battery voltage, UPS battery temperature, and UPS battery charging and discharging current into the preset virtual UPS battery position of the three-dimensional geometry to obtain the initial attribute matrix of the three-dimensional geometry; S212: Obtain the time sequence labels of the UPS battery voltage, UPS battery temperature and UPS battery charging and discharging current parameters in the initial attribute matrix of the three-dimensional geometry, perform numerical replacement on each parameter according to the time sequence labels, and generate a node dynamic numerical replacement sequence. S213: Continuously input the dynamic numerical replacement sequence of the node, establish the data association mapping structure between the physical entity and the virtual model in the cloud service device, and construct a battery digital twin.

[0010] As a further aspect of the present invention, the steps for obtaining the quantitative result of the remaining lifespan of the UPS battery are specifically as follows: S311: Collect the UPS battery temperature and UPS battery charging and discharging current during the historical time period of the battery digital twin, and perform normalization processing. Calculate the product of the normalized UPS battery temperature and UPS battery charging and discharging current with the corresponding preset temperature attenuation coefficient and current attenuation coefficient to obtain the comprehensive physical attenuation rate. Calculate the product of the comprehensive physical attenuation rate with the preset future prediction time step sequence to obtain the cumulative capacity attenuation. Calculate the deviation between the preset rated initial capacity and the cumulative capacity attenuation to obtain the theoretical remaining capacity lower limit sequence. S312: Extract UPS battery voltages from the same historical time period and normalize them. Input the normalized UPS battery temperature, UPS battery charging and discharging current, and UPS battery voltage parameters into the long short-term memory neural network model to extract the predicted remaining capacity sequence. S313: Replace the parameters in the predicted remaining capacity sequence that are less than the theoretical remaining capacity lower limit sequence with the corresponding lower limit values ​​to generate a limited remaining capacity evolution sequence. Extract the time points in the limited remaining capacity evolution sequence that correspond to the parameters reaching the preset lifespan end capacity threshold, calculate the difference between the corresponding time point and the current time point, and obtain the quantitative result of the remaining lifespan of the UPS battery.

[0011] As a further aspect of the present invention, the steps for obtaining the UPS battery health status early warning assessment level are specifically as follows: S411: Obtain the numerical deviation of the UPS battery voltage at the current adjacent time point to obtain the relative voltage fluctuation, and the numerical deviation of the UPS battery charging and discharging current to obtain the relative current fluctuation. Calculate the ratio of the relative voltage fluctuation to the relative current fluctuation to obtain the internal resistance of the UPS battery. Input the internal resistance of the UPS battery and the UPS battery temperature into the gradient boosting tree model to classify the fault risk level labels. S412: Calculate the ratio of the internal resistance of the UPS battery to the preset standard internal resistance reference value to obtain the relative internal resistance ratio; calculate the ratio of the UPS battery temperature to the preset standard temperature reference value to obtain the relative temperature ratio; and combine the preset internal resistance weight component and temperature weight component to calculate the UPS battery health quantitative score. S413: Using the UPS battery health quantification score, the health value distribution limit of the UPS battery is calibrated, and the health value distribution limit is combined with the fault risk classification label to determine the UPS battery health status early warning assessment level.

[0012] As a further aspect of the present invention, the steps for obtaining the UPS battery health prediction result are specifically as follows: S511: Set matching simulated charge and discharge current parameters in the battery digital twin, control the battery digital twin to run simulated charge and discharge current parameters, extract the quadratic characteristic of the simulated charge and discharge current parameters and combine it with the internal resistance of the UPS battery to establish a heat generation power evaluation quantity, and perform ratio normalization processing on the heat generation power evaluation quantity based on the equivalent convection heat dissipation coefficient to generate a steady-state limit temperature rise amplitude reference value. At the same time, obtain the specific heat capacity coefficient of the UPS battery material and the physical mass of the UPS battery cell, construct the battery total heat capacity characteristic benchmark, establish the heat transfer dynamic response index evaluation item, use the heat transfer dynamic response index evaluation item to construct the corresponding exponential asymptotic decay factor, apply nonlinear weighted mapping correction to the steady-state limit temperature rise amplitude reference value, generate the actual heat generation temperature rise dynamic compensation quantity, and perform benchmark superposition and fusion calculation with the pre-acquired initial shell node temperature to obtain the simulated operating temperature variable. S512: Compare the simulated operating temperature variable with the preset safe operating temperature assessment threshold, and establish a temperature simulation safety status verification conclusion when the simulated operating temperature variable is less than the safe operating temperature assessment threshold. S513: Combine the quantification results of the remaining service life of the UPS battery, the early warning assessment level of the UPS battery health status, and the verification conclusion of the temperature simulation safety status to output the UPS battery health prediction result.

[0013] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, a three-dimensional spatial attribute matrix is ​​constructed based on multi-dimensional operating parameters and the time-series labels are continuously updated. A dynamic correlation structure that highly maps physical entities to virtual spaces is established, accurately replicating the evolution trajectory of internal physical parameters of UPS batteries. By combining multi-feature normalization processing and time-series neural networks, the deep spatiotemporal characteristics of capacity degradation are deeply mined. By extracting the theoretical attenuation lower limit, the predicted sequence is constrained, effectively overcoming the bias in life assessment caused by environmental and sudden load interference. The collaborative mapping relationship between internal resistance and temperature fluctuations is deeply analyzed to classify risk levels. Simulated load flow is actively introduced into the virtual space to pre-determine the extreme conditions of heat accumulation, effectively eliminating the lag of static threshold judgment, realizing a fundamental shift from passive alarm to proactive predictive prevention, and effectively ensuring the long-term stability of the power supply base of the data center. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the main steps of the present invention; Figure 2 This is a flowchart illustrating how the present invention obtains a set of operating parameters; Figure 3 This is a flowchart illustrating the process of obtaining a battery digital twin according to the present invention; Figure 4 This is a flowchart illustrating the quantitative results of obtaining the remaining lifespan of a UPS battery according to the present invention. Figure 5 This is a flowchart illustrating the process of obtaining the UPS battery health status early warning assessment hierarchy in this invention; Figure 6 This is a flowchart illustrating the process of obtaining UPS battery health prediction results according to the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0016] Please see Figure 1 This invention provides a technical solution: a data center battery health prediction method based on digital twins, comprising the following steps: S1: During the operation of the data center physical server room, acquire the voltage, temperature and charging / discharging current of individual UPS batteries, compare the corresponding thresholds, and extract the set of operating parameters. S2: Analyze the numerical change process of UPS battery voltage, UPS battery temperature and UPS battery charging and discharging current in the set of operating parameters within a preset three-dimensional geometry, and construct a digital twin of the battery. S3: Based on the UPS battery voltage, UPS battery temperature, and UPS battery charging and discharging current over a historical period of the battery digital twin, construct a theoretical remaining capacity lower limit sequence and a time-series operating feature matrix. Input the time-series operating feature matrix into a long short-term memory neural network model to determine the predicted remaining capacity sequence. Based on the theoretical remaining capacity lower limit sequence and the predicted remaining capacity sequence, generate the quantitative result of the remaining service life of the UPS battery. S4: Obtain the relative voltage fluctuation of UPS battery voltage and UPS battery charging and discharging current, analyze the internal resistance of UPS battery, classify fault risk level labels based on the current UPS battery temperature, and determine the UPS battery health status early warning assessment level. S5: Based on the UPS battery health status early warning assessment level, set the simulated charging and discharging current parameters in the battery digital twin, determine the simulated operating temperature variable, and combine the quantitative results of the remaining service life of the UPS battery to obtain the UPS battery health prediction result.

[0017] The set of operating parameters includes UPS battery voltage extraction, UPS battery temperature extraction, and UPS battery charge / discharge current retention. The battery digital twin includes a three-dimensional geometric entity architecture, internal attributes of UPS battery nodes, and time-series value change records. The quantification result of the remaining service life of the UPS battery is specifically calculated by the time difference between the predicted capacity termination time and the current time. The UPS battery health status early warning assessment level is specifically calculated by the internal resistance conversion score of the equipment and the fault risk classification label. The UPS battery health prediction result includes the simulation operating temperature comparison conclusion, the digital twin simulation verification status, and the equipment pre-intervention maintenance instructions.

[0018] Please see Figure 2 The specific steps for obtaining the set of runtime parameters are as follows: S111: Obtain the voltage of a single UPS battery during the operation of the data center physical computer room, set a first voltage comparison threshold and a second voltage comparison threshold, and extract the UPS battery voltage between the first voltage comparison threshold and the second voltage comparison threshold; The specific process of obtaining the voltage of a single UPS battery during the operation of a data center physical server room involves using voltage sensors deployed at the positive and negative terminals of the UPS battery to continuously collect raw voltage data at a fixed sampling frequency, and then transmitting the raw voltage data to the data processing terminal in the data center. Upon receiving the raw voltage data, the data processing terminal performs data cleaning to remove outliers with empty values ​​or values ​​exceeding the range. Subsequently, a median filtering algorithm is used to smooth the raw voltage data. The entire process of this algorithm is as follows: the data processing terminal collects continuous raw voltage data, sets a fixed number of sampling points in a sliding window, sorts the values ​​within the window by size, selects the value in the middle position as the filtered output, and as new data is continuously transmitted and arrives, the window slides forward along the time axis, continuously outputting the current single UPS battery voltage after eliminating high-frequency noise interference. The specific process of setting the first and second voltage comparison thresholds is as follows: first, the time series of UPS battery voltages operating normally in float charging mode within the historical period of the data center operation and maintenance system is retrieved, and the average and standard deviation of this time series are calculated. The average value minus a specific multiple of the standard deviation is set as the first voltage comparison threshold, and the average value plus a specific multiple of the standard deviation is set as the second voltage comparison threshold. The voltage of each individual UPS battery at the current moment is compared sequentially with the first and second voltage comparison thresholds. If the voltage of an individual UPS battery at the current moment is greater than the first voltage comparison threshold but less than the second voltage comparison threshold, it is extracted and stored in the effective voltage sequence.

[0019] S112: Obtain the temperature of individual UPS batteries during the operation of the data center physical computer room, set the upper limit threshold for temperature safety, and extract the UPS battery temperatures that are lower than the upper limit threshold for temperature safety. The temperature of individual UPS batteries during data center physical room operation is obtained by continuously collecting raw temperature data of the environment and UPS battery surface using thermocouple temperature sensors attached to the surface of the UPS battery casing, and sending this data to the data processing end. The data processing end performs sliding window averaging on the raw temperature data, calculating the average of multiple raw temperature data within a set window as the current individual UPS battery temperature. The rated maximum operating temperature provided by the equipment manufacturer is obtained, and combined with the critical temperature data of thermal runaway occurrences from the data center's historical fault record database. The average of the rated maximum operating temperature and the critical temperature data is calculated, and this average is multiplied by a temperature safety factor to obtain the upper temperature safety threshold. The current individual UPS battery temperature is compared with the upper temperature safety threshold. If the current individual UPS battery temperature is strictly lower than the upper temperature safety threshold, this temperature data is extracted into the safe temperature sequence.

[0020] S113: Synchronously acquire the charging and discharging current of individual UPS batteries, integrate the UPS battery voltage, UPS battery temperature and UPS battery charging and discharging current to generate a set of operating parameters; The set of operating parameters can be sent to data center cloud service equipment, data center energy management platform, and data center operation and maintenance management platform; The charging and discharging current of a single UPS battery is synchronously acquired through a current sensor connected in series in the main circuit of the data center UPS battery. The data processing unit performs zero-point calibration and linear compensation algorithms on the raw charging and discharging current data. The algorithm works as follows: when the UPS battery is in an open-circuit, static state, the system collects the sensor's noise floor current value as the zero-point offset. Then, under a known standard dummy load discharge test, the sensor's response slope is obtained to generate a linear compensation function. In actual operation, the real-time acquired raw data is subtracted from the zero-point offset and then divided by the response slope to obtain the accurate charging and discharging current of a single UPS battery. The UPS battery voltage, UPS battery temperature, and UPS battery charging and discharging current are integrated, using a unified timestamp as an alignment benchmark, to perform time-dimensional matching of the UPS battery voltage in the aforementioned effective voltage sequence, the UPS battery temperature in the safe temperature sequence, and the accurate charging and discharging current of a single UPS battery. For data with sampling frequency differences, a linear interpolation algorithm is used. The process is as follows: For target sampling points with missing values ​​due to inconsistent timestamps, the two closest valid sampling points before and after the target timestamp are found. A linear function equation is constructed using the known time difference and numerical difference between the two points. The target timestamp is substituted into this equation to calculate the corresponding interpolation result, filling in the missing values ​​and ensuring that the corresponding voltage, temperature, and current values ​​are available at the same timestamp. The UPS battery voltage, UPS battery temperature, and UPS battery charging and discharging current at the same timestamp are concatenated into a multi-dimensional data record according to a fixed field order. As time progresses, multiple multi-dimensional data records are continuously collected and concatenated to form a complete set of operating parameters. The set of operating parameters can be sent to data center cloud service equipment, data center energy management platform, and data center operation and maintenance management platform to ensure the compatibility of the system with the existing energy management system and operation and maintenance management platform of the data center. This enables data sharing and collaborative linkage between multiple platforms, effectively avoiding data silos and redundant construction, thereby improving the overall operational stability and operation and maintenance efficiency of the power supply system.

[0021] Please see Figure 3 The specific steps for obtaining a battery digital twin are as follows: S211: Extract the UPS battery voltage, UPS battery temperature, and UPS battery charging and discharging current from the set of operating parameters, construct a three-dimensional geometry, and input the UPS battery voltage, UPS battery temperature, and UPS battery charging and discharging current into the preset virtual UPS battery position of the three-dimensional geometry to obtain the initial attribute matrix of the three-dimensional geometry. The system extracts UPS battery voltage, UPS battery temperature, and UPS battery charging / discharging current from the operating parameter set. Multidimensional data records from the operating parameter set are read sequentially from the data processing terminal, and their respective continuous numerical sequences are separated. In virtual space, a spatial skeleton structure composed of a cuboid mesh is established according to the length, width, and height ratio of the physical UPS battery. Based on the actual physical distribution of the cells inside the physical UPS battery, multiple virtual UPS battery positions corresponding to three-dimensional spatial coordinates are divided within the cuboid mesh. According to the three-dimensional coordinate mapping relationship between each monitoring point of the physical UPS battery and the virtual UPS battery positions, the separated UPS battery voltage is assigned to the corresponding virtual positive and negative terminal nodes, the UPS battery temperature is assigned to the corresponding virtual casing node, and the UPS battery charging / discharging current is assigned to the corresponding virtual circuit node. Following the node numbering order of the three-dimensional geometry, the voltage, temperature, and current values ​​currently assigned to all virtual UPS battery positions are arranged according to row and column rules, forming a two-dimensional matrix format containing node spatial coordinates and corresponding physical parameter values, thus obtaining the initial attribute matrix of the three-dimensional geometry.

[0022] S212: Obtain the time sequence labels of UPS battery voltage, UPS battery temperature and UPS battery charging and discharging current parameters in the initial attribute matrix of the three-dimensional geometry, perform numerical replacement on each parameter according to the time sequence labels, and generate a dynamic numerical replacement sequence of nodes. The process iterates through each record in the initial attribute matrix of the 3D geometry, extracting the absolute timestamp information accompanying the record's generation. These timestamps are then arranged in ascending order, and each record is assigned an incrementing integer sequence number as a time sequence label within the initial attribute matrix. Following the ascending order of the time sequence labels, the values ​​of UPS battery voltage, UPS battery temperature, and UPS battery charging / discharging current corresponding to the next time label are extracted sequentially. The current value storage location of the corresponding virtual node in the initial attribute matrix is ​​then located, and the new value extracted with the next time label overwrites the old value. During each value replacement, the number of the replaced node, the timestamp of the replacement, and the change in parameter values ​​before and after the replacement are recorded. These replacement records are then appended sequentially in chronological order, forming a dynamic value replacement sequence for the nodes.

[0023] S213: Continuously input the dynamic value replacement sequence of nodes, establish the data association mapping structure between physical entities and virtual models in the cloud service device, and construct a digital twin of the battery; A data transmission channel is established between the virtual space and the cloud service device, sending the generated dynamic node value replacement sequence to the receiving port of the cloud service device. The cloud service device parses the received dynamic node value replacement sequence and extracts the change data. A hash mapping table is established with physical sensor identifiers as keys and virtual node coordinates and attribute parameters as values. When the sensor collects new data, the system performs key-value matching retrieval on the hash mapping table, accurately locating the virtual node that needs updating, and completing the synchronization and binding of virtual and physical data. After establishing the data association mapping structure, the visualization rendering engine is launched. Based on the updated 3D geometric attribute matrix and the dynamic node value replacement sequence, the 3D geometric model is driven to generate a 3D stereoscopic display of parameter color gradient changes and numerical fluctuations, ultimately forming a battery digital twin with the ability to map physical states.

[0024] Please see Figure 4 The specific steps for obtaining the quantitative results of the remaining lifespan of the UPS battery are as follows: S311: Collect UPS battery temperature and UPS battery charging and discharging current during historical time periods from the battery digital twin, and perform normalization processing. Calculate the product of the normalized UPS battery temperature and UPS battery charging and discharging current with the corresponding preset temperature attenuation coefficient and current attenuation coefficient to obtain the comprehensive physical attenuation rate. Calculate the product of the comprehensive physical attenuation rate with the preset future prediction time step sequence to obtain the cumulative capacity attenuation. Calculate the deviation between the preset rated initial capacity and the cumulative capacity attenuation to obtain the theoretical remaining capacity lower limit sequence. The temperature decay coefficient is set according to the temperature-accelerated aging characteristics of the internal electrochemical reaction of the UPS battery, and the current decay coefficient is set according to the cyclic damage decay characteristics of the UPS battery under various charge and discharge rates. The system collects UPS battery temperature and charging / discharging current data from a historical time period using a battery digital twin. It retrieves the attribute update logs of the battery digital twin from past historical periods and extracts the UPS battery temperature and charging / discharging current sequences for all nodes. The mean and standard deviation of the UPS battery temperature and charging / discharging current sequences are calculated separately and normalized using Z-score standardization. The normalized UPS battery temperature is obtained by subtracting the mean temperature from each original temperature value and dividing by the corresponding temperature standard deviation; similarly, the normalized UPS battery charging / discharging current is obtained by subtracting the mean current from each original current value and dividing by the corresponding current standard deviation. The temperature decay coefficient is set based on the temperature-accelerated aging characteristics of the internal electrochemical reaction of the UPS battery, and the current decay coefficient is set based on the cyclic damage decay characteristics of the UPS battery at various charge / discharge rates. The normalized UPS battery temperature and the normalized UPS battery charging / discharging current are multiplied by the preset temperature decay coefficient and the preset current decay coefficient, respectively. These two products are then summed to obtain the comprehensive physical decay rate. The preset future prediction time step sequence consists of consecutive integer months. The cumulative capacity decay is obtained by multiplying the overall physical decay rate by each value in the preset future prediction time step sequence. The theoretical remaining capacity lower limit sequence is then constructed by subtracting the cumulative capacity decay corresponding to each time step from the preset rated initial capacity.

[0025] S312: Extract UPS battery voltages from the same historical time period and normalize them. Input the normalized UPS battery temperature, UPS battery charging and discharging current, and UPS battery voltage parameters into the long short-term memory neural network model to extract the predicted remaining capacity sequence. Similarly, the UPS battery voltage sequence from past historical periods is retrieved, and its minimum and maximum values ​​are calculated. Using the range normalization method, the minimum voltage value is subtracted from each original voltage value, and then divided by the difference between the maximum and minimum voltage values ​​to obtain the normalized UPS battery voltage. A Long Short-Term Memory (LSTM) neural network model is constructed. The entire process is as follows: historical multi-dimensional normalized feature data and the corresponding actual capacity are obtained as the training label set. The network topology is established, with the input layer containing multiple neurons that receive normalized UPS battery temperature, UPS battery charging and discharging current, and UPS battery voltage, respectively. The hidden layer contains a multi-layer LSM unit structure, with each LSM unit containing a forget gate, an input gate, and an output gate. The gating mechanism uses the logistic activation function to calculate the control weights, and the candidate state generation uses the hyperbolic tangent activation function to handle nonlinear mappings. During the training phase, the system randomly initializes the weight matrix and bias vector, uses the mean squared error as the loss function to calculate the error between the model's output prediction value and the actual historical remaining capacity data, and uses an optimization algorithm to calculate the gradient and update the weights and bias terms of each gating unit. The model output layer receives the forward propagation features and outputs capacity prediction values ​​for multiple future time steps based on the processed temporal features, forming a predicted remaining capacity sequence.

[0026] S313: Replace the parameters in the predicted remaining capacity sequence that are less than the theoretical remaining capacity lower limit sequence with the corresponding lower limit values ​​to generate a limited remaining capacity evolution sequence. Extract the time points in the limited remaining capacity evolution sequence that correspond to the parameters reaching the preset lifespan end capacity threshold, calculate the difference between the corresponding time point and the current time point, and obtain the quantitative result of the remaining lifespan of the UPS battery. The predicted remaining capacity sequence is compared with the corresponding values ​​in the theoretical remaining capacity lower limit sequence step by step. If the predicted remaining capacity value is less than the theoretical remaining capacity lower limit value, the corresponding lower limit value is filled into the current position; if the predicted remaining capacity value is greater than or equal to the theoretical remaining capacity lower limit value, the output predicted remaining capacity value remains unchanged. After the replacement operation, a new limited remaining capacity evolution sequence is reassembled in ascending order of time. The limited remaining capacity evolution sequence is traversed from left to right. If the first element in the sequence whose value is less than or equal to the preset lifespan end capacity threshold is found, the time step label corresponding to the position of that element is extracted as the time point; if no element less than or equal to the threshold is found, the last time step label of the sequence is extracted and an extension mark is added as the time point. The extracted time point value is subtracted from the current time point value to obtain the quantitative result of the remaining lifespan of the UPS battery.

[0027] Please see Figure 5 The specific steps for obtaining the UPS battery health status early warning assessment level are as follows: S411: Obtain the numerical deviation of the UPS battery voltage at the current adjacent time point to obtain the relative voltage fluctuation, and the numerical deviation of the UPS battery charging and discharging current to obtain the relative current fluctuation. Calculate the ratio of the relative voltage fluctuation to the relative current fluctuation to obtain the internal resistance of the UPS battery. Input the internal resistance of the UPS battery and the UPS battery temperature into the gradient boosting tree model to classify the fault risk level labels. The relative voltage fluctuation is obtained by subtracting the UPS battery voltage from the previous sampling time from the current UPS battery voltage and taking the absolute value. The relative current fluctuation is obtained by subtracting the UPS battery charging / discharging current from the previous time from the current UPS battery charging / discharging current and taking the absolute value. The relative voltage fluctuation is then divided by the relative current fluctuation to obtain the UPS battery's internal resistance. The UPS battery's internal resistance and temperature are input into a gradient boosting tree model to classify fault risk levels. The entire model process involves: extracting samples with healthy or faulty states; constructing a model using a regression tree as the base learner; calculating the residual between the model's predicted values ​​and the negative gradient of the true labels; substituting the internal resistance and temperature features into a new decision tree to fit this residual; and scaling the new tree before adding it to the model. During application execution, the model judges the input feature values ​​and calculates the sum of the weights of all decision tree output leaf nodes. The summation result is then mapped to a probability interval using the sigmoid function. If the model's mapping probability is less than the first probability cutoff point, it is classified as a low-risk label; if the mapping probability is greater than or equal to the first probability cutoff point and less than or equal to the second probability cutoff point, it is classified as a medium-risk label; if the mapping probability is greater than the second probability cutoff point, it is classified as a high-risk label.

[0028] S412: Calculate the ratio of the internal resistance of the UPS battery to the preset standard internal resistance reference value to obtain the relative internal resistance ratio; calculate the ratio of the UPS battery temperature to the preset standard temperature reference value to obtain the relative temperature ratio; and combine the preset internal resistance weight component and temperature weight component to calculate the UPS battery health quantitative score. The internal resistance of the UPS battery is divided by a preset standard internal resistance benchmark value to calculate the relative internal resistance ratio. The UPS battery temperature is divided by a preset standard temperature benchmark value to calculate the relative temperature ratio. Combining preset weighted components for internal resistance and temperature, a quantitative health score for the UPS battery is calculated. The system then calls an improved weighted collaborative penalty evaluation algorithm, the calculation formula of which is: , in the formula, This represents a quantitative score for UPS battery health. The various indicators and weights are dimensionless parameters, calculated using the formula above. This represents the internal resistance weight component, which is dimensionless. It is set based on the statistical proportion of power outages caused by abnormal internal resistance in the long-term operation data of the data center. The higher the statistical proportion, the larger the weight component is set, which means that the internal resistance index has a stronger determinant effect on the overall health. The value range is 0.50 to 0.80. This represents the relative proportion of the internal resistance obtained above. It is dimensionless and is calculated by dividing the internal resistance of the UPS battery by a preset standard internal resistance reference value. This represents a dimensionless temperature weight component. It is set based on the historical probability assessment of charging and discharging failures caused by abnormal data center environments. The higher the historical failure probability, the larger the temperature weight component is set. The value range is 0.20 to 0.50. This represents the relative proportion of the aforementioned temperatures. It is dimensionless and is calculated by dividing the UPS battery temperature by a preset standard temperature reference value. The coefficient representing the synergistic degradation penalty is dimensionless and is set based on the accelerated aging multiplier calibration in UPS battery multiphysics coupling failure experiments. It characterizes the synergistic degradation effect where the risk of thermal runaway is nonlinearly amplified when the internal resistance of the UPS battery increases at high temperatures. A larger coefficient indicates a stronger penalty for the synergistic degradation state, with a value ranging from 0.05 to 0.20. Substitute the parameter settings into the formula for calculation: The calculation result serves as a quantitative indicator; the higher the score, the greater the deviation of the UPS battery from the factory health benchmark.

[0029] S413: Utilize the UPS battery health quantitative scoring to calibrate the health value distribution limits of the UPS battery, combine the health value distribution limits with the fault risk classification label, and determine the UPS battery health status early warning assessment level. The UPS battery health quantification score is used to calibrate the health value distribution boundaries of the UPS battery. Multiple distribution boundaries for the health quantification score are predefined. If the health quantification score is less than the first boundary, it is marked as healthy; if the health quantification score is greater than or equal to the first boundary and less than the second boundary, it is marked as sub-healthy; if the health quantification score is greater than or equal to the second boundary and less than or equal to the third boundary, it is marked as a warning boundary; if the health quantification score is greater than the third boundary, it is marked as a dangerous boundary. An assessment judgment matrix with two dimensions is established: the row dimension represents the health value distribution boundaries, and the column dimension represents the fault risk classification label. Based on the aforementioned risk labels and boundary positions, a row-column cross-referencing query is performed in the judgment matrix. If the distribution boundary is the healthy boundary and the risk label is low risk, it is mapped to a Level 5 early warning assessment level; if the distribution boundary is the sub-healthy boundary and the risk label is low risk, it is mapped to a Level 4 early warning assessment level; if the distribution boundary is the sub-healthy boundary and the risk label is medium risk, it is mapped to a Level 3 early warning assessment level; if the distribution boundary is the warning boundary and the risk label is medium risk, it is directly mapped to a Level 2 early warning assessment level; if the distribution boundary is the warning boundary and the risk label is high risk, or the distribution boundary is the danger boundary combined with any risk label, it is always mapped to a Level 1 early warning assessment level. The UPS battery health status early warning assessment level is obtained through cross-querying the decision matrix.

[0030] Please see Figure 6 The specific steps for obtaining UPS battery health prediction results are as follows: S511: Set matching simulated charge and discharge current parameters in the battery digital twin, control the battery digital twin to run simulated charge and discharge current parameters, extract the quadratic characteristic of simulated charge and discharge current parameters and combine it with the internal resistance of UPS battery to establish heat generation power evaluation quantity, and perform ratio normalization processing on heat generation power evaluation quantity based on equivalent convection heat dissipation coefficient to generate steady-state limit temperature rise amplitude reference value. At the same time, obtain the specific heat capacity coefficient of UPS battery material and the physical mass of UPS battery cell to construct battery total heat capacity characteristic benchmark, establish heat transfer dynamic response index evaluation item, use heat transfer dynamic response index evaluation item to construct corresponding exponential asymptotic decay factor, apply nonlinear weighted mapping correction to steady-state limit temperature rise amplitude reference value, generate actual heat generation temperature rise dynamic compensation quantity, and perform benchmark superposition and fusion calculation with the pre-acquired initial shell node temperature to obtain simulated operating temperature variable. Extract the projected future electricity load curve and calculate the discharge current required to handle the maximum load during a power outage. Multiply the required discharge current at the maximum load by a redundancy factor to obtain the simulated charge-discharge current parameters, which are then input into the twin system. Activate the thermodynamic simulation solver process in the twin cloud. Combining the previously obtained physical parameters, the system calls the Joule heat production coupled active cooling simulation algorithm based on the transient thermal equilibrium differential equation to calculate the dynamic heat accumulation. The calculation formula is as follows: ; In the formula, The simulated operating temperature variable represents the final predicted maximum shell temperature, expressed in degrees Celsius, and is calculated using the formula above. This represents the initial casing node temperature in degrees Celsius, and the actual value at the current moment is continuously obtained by a thermocouple temperature sensor attached to the surface of the UPS battery casing. The simulated charge and discharge current parameter is in amperes and is obtained by extracting the expected maximum load discharge current and multiplying it by the system redundancy factor. The value represents the internal resistance of the UPS battery obtained above, in ohms, and is calculated by dividing the relative voltage fluctuation by the relative current fluctuation. The equivalent convection heat dissipation coefficient is expressed in watts per degree Celsius. It is set based on the comprehensive heat exchange efficiency calibration experiment of the computer room air conditioning fan speed, ambient temperature difference and UPS battery exposed surface area. The larger the coefficient is set, the stronger the active convection cooling capacity of the computer room environment and the higher the heat dissipation efficiency. The value range is usually from 0.50 to 2.50. Representing the natural constant, it is a dimensionless mathematical constant approximately equal to 2.71828, and is used as the base of the exponential function in calculations; This represents the continuous discharge simulation time, in seconds, which is directly obtained and set based on the emergency power supply standard duration of the computer room backup power supply. The specific heat capacity coefficient represents the UPS battery material, and the unit is joules per kilogram of degree Celsius. It is obtained from the material physical property manual provided by the UPS battery manufacturer. This coefficient reflects the heat absorbed by a unit mass of material to increase the temperature by 1 degree Celsius. The larger the specific heat capacity coefficient, the slower the temperature rise of the UPS battery cell material under the same heat input. The value range is usually 800 to 1200. This represents the physical mass of a single UPS battery cell, expressed in kilograms, and can be obtained by consulting the manufacturer's equipment specifications.

[0031] For example, setting the initial temperature The simulated charging and discharging current is 38.5 degrees Celsius. 200.0 Amperes, internal resistance The equivalent convective heat dissipation coefficient is 0.005 ohms. It has a power output of 2.0 watts per degree Celsius and a discharge duration of [missing information]. 3600.0 seconds, specific heat capacity The physical mass is 1000.0 joules per kilogram per degree Celsius. The weight is 36.0 kg. Substituting this into the formula, we get: ; The calculated temperature variable for the simulated operation is 56.63 degrees Celsius.

[0032] S512: Compare the simulated operating temperature variable with the preset safe operating temperature assessment threshold. If the simulated operating temperature variable is less than the safe operating temperature assessment threshold, establish the temperature simulation safety status verification conclusion. The obtained simulated operating temperature variable is compared with the safe operating temperature assessment threshold. If the simulated operating temperature variable is strictly less than the safe operating temperature assessment threshold, a state conclusion of "parameter settings are reasonable and thermal control is safe" is generated as the temperature simulation safety state verification conclusion; if the simulated operating temperature variable is greater than or equal to the safe operating temperature assessment threshold, a danger warning conclusion of "parameter settings are unreasonable and there is a risk of overheating and runaway" is generated and output.

[0033] S513: Quantitative results of remaining lifespan of combined UPS batteries, UPS battery health status early warning assessment level and temperature simulation safety status verification conclusions, output UPS battery health prediction results; On the data processing side, the quantified result of the remaining service life of the UPS battery calculated above is first written into the time prediction field. Next, the UPS battery health status early warning assessment level is written into the risk level field. Finally, the temperature simulation safety status verification conclusion is written into the simulation verification field. After completing the field filling, a checksum is attached, and a UPS battery health prediction result report file is generated. This prediction result report file is then pushed to the display terminal of the data center operation and maintenance management platform.

[0034] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A data center battery health prediction method based on digital twins, characterized in that, Includes the following steps: S1: During the operation of the data center physical server room, acquire the voltage, temperature and charging / discharging current of individual UPS batteries, compare the corresponding thresholds, and extract the set of operating parameters. S2: Analyze the numerical change process of the UPS battery voltage, UPS battery temperature, and UPS battery charging and discharging current in the set of operating parameters within a preset three-dimensional geometry, and construct a battery digital twin; S3: Based on the UPS battery voltage, UPS battery temperature, and UPS battery charging and discharging current during the historical time period of the battery digital twin, construct a theoretical remaining capacity lower limit sequence and a time-series operation feature matrix. Input the time-series operation feature matrix into a long short-term memory neural network model to determine the predicted remaining capacity sequence. Based on the theoretical remaining capacity lower limit sequence and the predicted remaining capacity sequence, generate the quantitative result of the remaining service life of the UPS battery. S4: Obtain the relative voltage fluctuation of UPS battery voltage and UPS battery charging and discharging current, analyze the internal resistance of UPS battery, classify fault risk level labels based on the current UPS battery temperature, and determine the UPS battery health status early warning assessment level. S5: Based on the UPS battery health status early warning assessment level, set the simulated charging and discharging current parameters in the battery digital twin, determine the simulated operating temperature variable, and combine the quantification results of the remaining service life of the UPS battery to obtain the UPS battery health prediction result.

2. The data center battery health prediction method based on digital twins according to claim 1, characterized in that, The set of operating parameters includes UPS battery voltage extraction, UPS battery temperature extraction, and UPS battery charge / discharge current retention. The battery digital twin includes a three-dimensional geometric entity architecture, internal attributes of UPS battery nodes, and time-series value change records. The quantification result of the remaining service life of the UPS battery is specifically calculated by the time difference between the predicted capacity termination time and the current time. The UPS battery health status early warning assessment level is specifically calculated by the internal resistance conversion score and fault risk classification label of the equipment. The UPS battery health prediction result includes the simulation operating temperature comparison conclusion, the digital twin simulation verification status, and the equipment pre-intervention maintenance instructions.

3. The data center battery health prediction method based on digital twins according to claim 1, characterized in that, The specific steps for obtaining the set of operating parameters are as follows: S111: Obtain the voltage of a single UPS battery during the operation of the data center physical computer room, set a first voltage comparison threshold and a second voltage comparison threshold, and extract the UPS battery voltage between the first voltage comparison threshold and the second voltage comparison threshold; S112: Obtain the temperature of individual UPS batteries during the operation of the data center physical computer room, set the upper limit threshold for temperature safety, and extract the UPS battery temperatures that are lower than the upper limit threshold for temperature safety. S113: Synchronously acquire the charging and discharging current of a single UPS battery, integrate the UPS battery voltage, the UPS battery temperature, and the UPS battery charging and discharging current to generate a set of operating parameters.

4. The data center battery health prediction method based on digital twins according to claim 3, characterized in that, The specific steps for obtaining the battery digital twin are as follows: S211: Extract the UPS battery voltage, UPS battery temperature, and UPS battery charging and discharging current from the set of operating parameters, construct a three-dimensional geometry, and input the UPS battery voltage, UPS battery temperature, and UPS battery charging and discharging current into the preset virtual UPS battery position of the three-dimensional geometry to obtain the initial attribute matrix of the three-dimensional geometry; S212: Obtain the time sequence labels of the UPS battery voltage, UPS battery temperature and UPS battery charging and discharging current parameters in the initial attribute matrix of the three-dimensional geometry, perform numerical replacement on each parameter according to the time sequence labels, and generate a node dynamic numerical replacement sequence. S213: Continuously input the dynamic numerical replacement sequence of the node, establish the data association mapping structure between the physical entity and the virtual model in the cloud service device, and construct a battery digital twin.

5. The data center battery health prediction method based on digital twins according to claim 4, characterized in that, The specific steps for obtaining the quantitative result of the remaining lifespan of the UPS battery are as follows: S311: Collect the UPS battery temperature and UPS battery charging and discharging current during the historical time period of the battery digital twin, and perform normalization processing. Calculate the product of the normalized UPS battery temperature and UPS battery charging and discharging current with the corresponding preset temperature attenuation coefficient and current attenuation coefficient to obtain the comprehensive physical attenuation rate. Calculate the product of the comprehensive physical attenuation rate with the preset future prediction time step sequence to obtain the cumulative capacity attenuation. Calculate the deviation between the preset rated initial capacity and the cumulative capacity attenuation to obtain the theoretical remaining capacity lower limit sequence. S312: Extract UPS battery voltages from the same historical time period and normalize them. Input the normalized UPS battery temperature, UPS battery charging and discharging current, and UPS battery voltage parameters into the long short-term memory neural network model to extract the predicted remaining capacity sequence. S313: Replace the parameters in the predicted remaining capacity sequence that are less than the theoretical remaining capacity lower limit sequence with the corresponding lower limit values ​​to generate a limited remaining capacity evolution sequence. Extract the time points in the limited remaining capacity evolution sequence that correspond to the parameters reaching the preset lifespan end capacity threshold, calculate the difference between the corresponding time point and the current time point, and obtain the quantitative result of the remaining lifespan of the UPS battery.

6. The data center battery health prediction method based on digital twins according to claim 5, characterized in that, The specific steps for obtaining the UPS battery health status early warning assessment level are as follows: S411: Obtain the numerical deviation of the UPS battery voltage at the current adjacent time point to obtain the relative voltage fluctuation, and the numerical deviation of the UPS battery charging and discharging current to obtain the relative current fluctuation. Calculate the ratio of the relative voltage fluctuation to the relative current fluctuation to obtain the internal resistance of the UPS battery. Input the internal resistance of the UPS battery and the UPS battery temperature into the gradient boosting tree model to classify the fault risk level labels. S412: Calculate the ratio of the internal resistance of the UPS battery to the preset standard internal resistance reference value to obtain the relative internal resistance ratio; calculate the ratio of the UPS battery temperature to the preset standard temperature reference value to obtain the relative temperature ratio; and combine the preset internal resistance weight component and temperature weight component to calculate the UPS battery health quantitative score. S413: Using the UPS battery health quantification score, the health value distribution limit of the UPS battery is calibrated, and the health value distribution limit is combined with the fault risk classification label to determine the UPS battery health status early warning assessment level.

7. The data center battery health prediction method based on digital twins according to claim 6, characterized in that, The specific steps for obtaining the UPS battery health prediction results are as follows: S511: Set matching simulated charge and discharge current parameters in the battery digital twin, control the battery digital twin to run simulated charge and discharge current parameters, extract the quadratic characteristic of the simulated charge and discharge current parameters and combine it with the internal resistance of the UPS battery to establish a heat generation power evaluation quantity, and perform ratio normalization processing on the heat generation power evaluation quantity based on the equivalent convection heat dissipation coefficient to generate a steady-state limit temperature rise amplitude reference value. At the same time, obtain the specific heat capacity coefficient of the UPS battery material and the physical mass of the UPS battery cell, construct the battery total heat capacity characteristic benchmark, establish the heat transfer dynamic response index evaluation item, use the heat transfer dynamic response index evaluation item to construct the corresponding exponential asymptotic decay factor, apply nonlinear weighted mapping correction to the steady-state limit temperature rise amplitude reference value, generate the actual heat generation temperature rise dynamic compensation quantity, and perform benchmark superposition and fusion calculation with the pre-acquired initial shell node temperature to obtain the simulated operating temperature variable. S512: Compare the simulated operating temperature variable with the preset safe operating temperature assessment threshold, and establish a temperature simulation safety status verification conclusion when the simulated operating temperature variable is less than the safe operating temperature assessment threshold. S513: Combine the quantitative results of the remaining service life of the UPS battery, the early warning assessment level of the UPS battery health status, and the verification conclusion of the temperature simulation safety status to output the UPS battery health prediction result.

8. The data center battery health prediction method based on digital twins according to claim 3, characterized in that, The set of operating parameters can be sent to data center cloud service equipment, data center energy management platform, and data center operation and maintenance management platform.

9. The data center battery health prediction method based on digital twins according to claim 5, characterized in that, The temperature decay coefficient is set according to the temperature-accelerated aging characteristics of the internal electrochemical reaction of the UPS battery, and the current decay coefficient is set according to the cyclic damage decay characteristics of the UPS battery under various charge and discharge rates.

Images