Search

CN-121973635-A - Early warning method and system for thermal runaway of power battery of new energy automobile

CN121973635ACN 121973635 ACN121973635 ACN 121973635ACN-121973635-A

Abstract

The invention discloses a new energy automobile power battery thermal runaway early warning method and a system thereof, belonging to the technical field of new energy automobile battery safety monitoring, wherein the method comprises the following steps: the method comprises the steps of multi-dimensional parameter acquisition and feature cascade fusion, battery state time sequence prediction and residual error anomaly detection based on an attention-enhanced long-short-term memory network, thermal runaway multi-stage state migration estimation based on a hidden Markov model, model parameter dynamic update based on online self-adaptive learning, multi-stage early warning and emergency treatment grading linkage response, and accurate early warning of thermal runaway is achieved through deep coupling closed loop cooperation among the steps.

Inventors

  • YIN YUHAN
  • ZHAO BIN

Assignees

  • 长沙理工大学

Dates

Publication Date
20260505
Application Date
20260402

Claims (10)

  1. 1. The early warning method for the thermal runaway of the power battery of the new energy automobile is characterized by comprising the following steps of: Step S1, collecting and feature cascading fusion of multidimensional parameters, namely synchronously collecting original parameters of six dimensions including voltage, current, temperature, internal resistance, combustible gas concentration and shell strain through a distributed sensing array deployed on a battery module, preprocessing, intercepting through a sliding window, and cascading and splicing to form a multidimensional feature tensor; Step S2, based on battery state time sequence prediction and residual error abnormality detection of the attention-enhancing long-short-term memory network, inputting a multi-dimensional characteristic tensor into a pre-trained attention-enhancing long-short-term memory network model for time sequence prediction, calculating a multi-dimensional prediction residual error vector between a predicted value and an actual observed value, mapping the multi-dimensional prediction residual error vector into a scalar abnormality index by utilizing a Markov distance, generating an abnormality marking signal when the abnormality index exceeds a first threshold value, and outputting the scalar abnormality index at each reasoning moment as an abnormality index sequence according to time sequence; s3, performing thermal runaway multi-stage state migration estimation based on a hidden Markov model, namely dividing a hidden state space into four hidden states of normal, early warning, danger and runaway by taking the abnormality index sequence output in the step S2 as an observation sequence, calculating the posterior probability of each hidden state in real time through a forward algorithm, and estimating migration probability to the dangerous state and the runaway state; step S4, dynamically updating model parameters based on online self-adaptive learning, namely dynamically adjusting learning rate and forgetting gate bias parameters of the model in step S2 according to posterior probability distribution of each hidden state output in step S3, carrying out online incremental updating on parameters of the hidden Markov model in step S3, and feeding back the updated parameters to step S2 and step S3 to form closed loop correction; And S5, performing multistage early warning and emergency treatment grading linkage response, namely determining a risk grade according to the migration probability and the state posterior probability, triggering grading response strategies including reducing charge and discharge power, starting active liquid cooling circulation and emergency power off isolation, and feeding back a response result to the step S1 to dynamically adjust the acquisition frequency.
  2. 2. The early warning method for thermal runaway of a power battery of a new energy automobile according to claim 1, wherein in the step S1, the distributed sensing array comprises thermocouple temperature sensors, electrochemical gas sensors and flexible strain gauges, wherein the thermocouple temperature sensors are distributed at equal intervals along the longitudinal axis of a battery module, the electrochemical gas sensors are installed near an exhaust valve, the flexible strain gauges are attached to the surface of a battery shell, the acquisition frequency is set to 10 samples per second for a voltage channel and a current channel, 1 sample per second for a temperature channel and an internal resistance channel, 5 samples per second for a gas concentration and a strain channel, and the sliding window length is set to 60 seconds to 300 seconds.
  3. 3. The early warning method for thermal runaway of a power battery of a new energy automobile according to claim 1, wherein in step S1, the cascade splicing further comprises performing zero mean unit variance standardization on each dimension parameter, and resampling and aligning channels with different sampling frequencies by linear interpolation, so that all channels are unified to the same time resolution.
  4. 4. The early warning method for thermal runaway of a power battery of a new energy automobile according to claim 1, wherein in step S2, the attention-enhancing long-short-term memory network model includes an encoder and a decoder, the encoder is formed by stacking two layers of long-short-term memory network units, each layer of hidden dimension is set to 128, the hidden state of the last time step of the encoder is weighted by a multi-head self-attention mechanism and then is used as an initial state of the decoder, and the decoder outputs predicted values of parameters of each dimension of a preset step length in the future.
  5. 5. The early warning method for thermal runaway of a power battery of a new energy automobile according to claim 1, wherein in step S2, a covariance matrix updated based on an exponential weighted moving average is adopted when a mahalanobis distance mapping is performed on a multi-dimensional prediction residual vector, and the first threshold is determined by taking a 99 th percentile of cumulative distribution of abnormality indexes on normal working condition training data.
  6. 6. The early warning method for thermal runaway of a power battery of a new energy automobile according to claim 1, wherein in step S3, a state transition probability matrix of the hidden markov model is constrained to an upper triangular dominant matrix, wherein the normal state only allows transition to a warning state, the warning state allows transition to a normal state back or transition to a dangerous state, the dangerous state only allows transition to a runaway state, and the runaway state is an absorption state.
  7. 7. The early warning method for thermal runaway of a power battery of a new energy automobile according to claim 1, wherein in step S4, an online incremental update adopts a small-batch expectation maximization algorithm, each time an observation sequence within a latest preset time window is used for iterating an expectation step and a maximization step, and forgetting factors are introduced to exponentially attenuate historical statistics to adapt to parameter drift caused by battery aging.
  8. 8. The early warning method for thermal runaway of a power battery of a new energy automobile according to claim 1, wherein in the step S5, the risk level determining rule is that the first-level early warning is triggered when the posterior probability of the dangerous state is greater than a second threshold value and the migration probability is greater than a third threshold value, the second-level early warning is triggered when the posterior probability of the dangerous state is greater than a fourth threshold value, and the third-level early warning is triggered when the posterior probability of the runaway state is greater than a fifth threshold value.
  9. 9. The early warning method for thermal runaway of the power battery of the new energy automobile according to claim 1 is characterized in that step S5 further comprises the step of uploading all state parameters and warning information of a current battery module to a cloud server through a vehicle-mounted communication module after triggering any stage of warning, and the cloud server performs cluster-level statistical analysis on the received warning information and pushes a collaborative treatment suggestion to a vehicle.
  10. 10. The early warning system for thermal runaway of a power battery of a new energy automobile, which is used for realizing the early warning method for thermal runaway of a power battery of a new energy automobile according to any one of claims 1 to 9, is characterized by comprising: The multi-dimensional parameter acquisition and characteristic cascading fusion module is configured to synchronously acquire original parameters of six dimensions including voltage, current, temperature, internal resistance, combustible gas concentration and shell strain through a distributed sensing array deployed on the battery module, and perform cascading and splicing after preprocessing and sliding window interception operation on the original parameters to form a multi-dimensional characteristic tensor; The attention-enhancing long-short-term memory network prediction and residual error anomaly detection module is configured to receive the multidimensional feature tensor, output predicted values of all dimension parameters through the attention-enhancing long-short-term memory network model, calculate multidimensional prediction residual error vectors, map the multidimensional prediction residual error vectors into scalar anomaly indexes by utilizing the mahalanobis distance, generate anomaly marking signals when the anomaly indexes exceed a first threshold, and output the scalar anomaly indexes at all reasoning moments as an anomaly index sequence in a time sequence; The hidden Markov state transition estimation module is configured to take the abnormality index sequence output by the attention-enhanced long-short-term memory network prediction and residual error abnormality detection module as an observation sequence, and calculate the posterior probability of each hidden state and the transition probability to a dangerous state and an out-of-control state in real time based on the hidden Markov model of four discrete hidden states; The on-line self-adaptive learning and model updating module is configured to dynamically adjust the parameters of the attention-enhancing long-short-term memory network model and the hidden Markov model according to the posterior probability distribution of each hidden state output by the hidden Markov state migration estimation module and feed back the updated parameters to the corresponding modules to form closed loop correction; The multi-stage early warning and emergency treatment linkage module is configured to determine a risk level according to the migration probability and the state posterior probability and trigger a stage response strategy comprising reducing charge and discharge power, starting active liquid cooling circulation and emergency power off isolation.

Description

Early warning method and system for thermal runaway of power battery of new energy automobile Technical Field The invention relates to the technical field of safety monitoring of a new energy automobile battery, in particular to a new energy automobile power battery thermal runaway early warning method based on multi-parameter fusion and machine learning and a corresponding early warning system. Background With the rapid development of new energy automobile industry, lithium ion power batteries are widely used as core energy storage devices for pure electric automobiles and plug-in hybrid electric automobiles. However, lithium ion batteries are prone to thermal runaway under abnormal conditions such as overcharging, internal short-circuiting, mechanical abuse and the like, so that the internal temperature of the battery is rapidly increased and the internal temperature is released along with combustible gas, and serious safety accidents such as fire and severe explosion are further caused. According to statistics, more than half of new energy automobile safety accidents are directly related to battery thermal runaway in recent years, and life safety of drivers and surrounding pedestrians is seriously threatened, so that development of an efficient and reliable early warning technology for the thermal runaway is needed. From the evolution mechanism of thermal runaway, the internal side reactions of the battery generally undergo multiple progressive stages of solid electrolyte interfacial film decomposition, separator shrinkage, and positive electrode material decomposition, with time windows ranging from minutes to tens of minutes between the initial minute gas release and slight temperature rise to the final irreversible thermal runaway. How to accurately identify early signs of thermal runaway and make correct decisions within this limited time window and to take appropriate step-wise treatment measures according to the degree of risk evolution is a key technical challenge facing the current power battery safety field. In addition, the capacity attenuation and the internal resistance increase of the battery inevitably occur in the long-term use process, so that the thermal runaway precursor characteristics of the battery in different aging stages are obviously different, and the thermal stability and the side reaction paths among different battery cell chemical systems are different, so that the technical difficulty of early warning is further increased. The existing battery thermal runaway detection method mainly depends on threshold judgment of single or small quantity of parameters such as temperature, voltage and the like. For example, chinese patent application CN119291535A discloses a battery thermal runaway detection system based on an early warning function, where the system collects battery voltage, temperature, current and pressure parameters through a monitoring device, establishes an early warning model by using multiple learning algorithms such as a long-short-term memory network, a time convolution network and a gate control circulation unit, and sets upper and lower limits of a thermal runaway threshold for judgment. The scheme realizes the early warning function of thermal runaway of the battery to a certain extent, but the following defects are found after deep technical analysis. Firstly, the early warning mechanism of the scheme still depends on static threshold comparison basically, early warning is triggered when the predicted parameter data is not in a threshold interval, the gradual change characteristic that the thermal runaway gradually evolves from a normal state to a dangerous state through an early warning stage is difficult to capture by the binary judgment mode, and sufficient early warning margin and grading treatment time cannot be provided for drivers and passengers. Secondly, the scheme only collects four parameters of voltage, temperature, current and pressure, and does not cover key information closely related to early signs of thermal runaway such as concentration of combustible gas, deformation of a shell and the like, so that weak signal perception capability of the battery in an initial stage of side reaction is insufficient. Thirdly, parameters of the early warning model in the scheme are fixed after training is finished, self-adaptive updating capability aiming at different battery cell types and aging degrees is lacked, and when performance degradation occurs in a long-term use process of the battery, early warning precision is gradually reduced. Fourth, the early warning output of this scheme only includes disconnection circuit and two kinds of measures of buzzing warning, lacks the ability of carrying out hierarchical response according to risk severity, can't realize the orderly linkage of multistage treatment strategies such as power reduction, initiative cooling and urgent outage. Therefore, it is necessary to provide a power battery thermal runaway early war