An electrochemical energy storage tank toxic gas risk prediction method based on multi-model fusion

Abstract

The present disclosure relates to the technical field of electrochemical energy storage tank toxic gas risk prediction, and particularly relates to an electrochemical energy storage tank toxic gas risk prediction method based on multi-model fusion, comprising: searching for a preset number of similar transportation scenarios in a preset weight scenario library, and obtaining preset optimal weights of a tank model, a Gaussian plume model and a machine learning model under the preset number of similar transportation scenarios; performing interpolation calculation based on the preset optimal weights and a similarity corresponding to the similar transportation scenarios to obtain target weights of the tank model, the Gaussian plume model and the machine learning model, performing weighted summation based on the target weights to obtain target diffusion data; and obtaining a toxic gas risk prediction result of the electrochemical energy storage tank based on at least the target diffusion data. Thus, by fusing the tank model, the Gaussian plume model and the machine learning model, the accuracy and reliability of the electrochemical energy storage tank toxic gas risk prediction are improved.

Description

Technical Field

[0001] This disclosure relates to the field of toxic gas risk prediction technology for electrochemical energy storage boxes, specifically to a method for toxic gas risk prediction for electrochemical energy storage boxes based on multi-model fusion. Background Technology

[0002] With the rapid development of the electrochemical energy storage industry, the large-scale application of lithium-ion battery energy storage systems has brought new challenges to their safety risk management. Electrochemical energy storage tanks can release large amounts of toxic gases (such as HF, CO, and SO2) under thermal runaway conditions. These gases are highly corrosive and extremely toxic. If leaks occur during transportation or use, they can not only cause poisoning and injury to personnel but also potentially trigger secondary environmental pollution accidents. Especially in the complex environment of multimodal transport, the risk of toxic gas leaks from energy storage tanks increases significantly due to external factors such as transport vibration and temperature and humidity changes.

[0003] However, the industry currently lacks a systematic measurement method for assessing the risk of toxic gas diffusion in electrochemical energy storage boxes. Existing models struggle to accurately quantify the diffusion range and severity of hazards under different scenarios, severely hindering the improvement of safe transportation and emergency management capabilities for energy storage systems. Therefore, establishing a scientific and accurate risk measurement model for toxic gas diffusion in electrochemical energy storage boxes has become a key technological requirement for ensuring the safe development of the electrochemical energy storage industry, and is of significant practical importance for improving industry safety standards and enhancing accident prevention capabilities. Summary of the Invention

[0004] To overcome the problems existing in related technologies, this disclosure provides a method for predicting the risk of toxic gases in electrochemical energy storage boxes based on multi-model fusion, in order to solve the defects in related technologies.

[0005] According to a first aspect of the present disclosure, a method for predicting the risk of toxic gases in an electrochemical energy storage box based on multi-model fusion is provided, comprising:

[0006] The first diffusion data of the toxic gas in the electrochemical energy storage box was obtained by using a box model, the second diffusion data of the toxic gas was obtained by using a Gaussian plume model, and the third diffusion data of the toxic gas was obtained by using a machine learning model.

[0007] Based on the environmental information around the electrochemical energy storage box, a preset number of similar transportation scenarios similar to the current transportation scenario of the electrochemical energy storage box are searched in a preset weighted scenario library, and the preset optimal weights corresponding to the box model, the Gaussian plume model and the machine learning model under the preset number of similar transportation scenarios are obtained respectively.

[0008] Based on the preset optimal weights and the similarity between the current transportation scenario and the similar transportation scenario, interpolation calculations are performed to obtain the target weights corresponding to the box model, the Gaussian plume model and the machine learning model, respectively. Then, the first diffusion data, the second diffusion data and the third diffusion data are weighted and summed based on the target weights to obtain the target diffusion data.

[0009] Based at least on the target diffusion data, the toxic gas risk prediction results of the electrochemical energy storage box are obtained.

[0010] In one embodiment, the step of interpolating based on the preset optimal weights and the similarity between the current transportation scenario and the similar transportation scenario to obtain the target weights corresponding to the box model, the Gaussian plume model, and the machine learning model includes:

[0011] The sum of similarities between the current transportation scenario and each of the similar transportation scenarios is used as a normalization parameter. For each of the similar transportation scenarios, the result of dividing the similarity between the current transportation scenario and the similar transportation scenario by the normalization parameter is used as the normalized similarity weight corresponding to the similar transportation scenario.

[0012] Based on the normalized similarity weights, the preset optimal weights corresponding to the box-type models in each of the similar transportation scenarios are weighted and summed to obtain the target weights corresponding to the box-type models. Similarly, based on the normalized similarity weights, the preset optimal weights corresponding to the Gaussian plume models in each of the similar transportation scenarios are weighted and summed to obtain the target weights corresponding to the Gaussian plume models. Finally, based on the normalized similarity weights, the preset optimal weights corresponding to the machine learning models in each of the similar transportation scenarios are weighted and summed to obtain the target weights corresponding to the machine learning models.

[0013] In one embodiment, the weighted summation of the first diffusion data, the second diffusion data, and the third diffusion data based on the target weight includes:

[0014] The confidence scores of the box model, Gaussian plume model, and machine learning model are calculated using the Bayesian confidence assessment method, based on the historical accuracy of the box model, Gaussian plume model, and machine learning model, as well as the deviation between the current results and the actual observations.

[0015] If there is a target result confidence level in the result confidence level that is less than a preset confidence level threshold, then according to a preset mapping relationship, the target weight of the model corresponding to the target result confidence level is reduced. The preset mapping relationship indicates that for every first decrease in the target result confidence level compared to the preset initial confidence level, the target weight is reduced by a second value, and the first value is different from the second value.

[0016] The first diffusion data, the second diffusion data, and the third diffusion data are weighted and summed based on the reduced target weights.

[0017] In one embodiment, the weighted summation of the first diffusion data, the second diffusion data, and the third diffusion data based on the target weight includes:

[0018] Based on the characteristics of the packaging material of the electrochemical energy storage box, the protection level of the electrochemical energy storage box is determined, and the monitoring parameters of the protection device of the electrochemical energy storage box are obtained.

[0019] If the protection level is higher than or equal to a preset level threshold and the difference between the monitoring parameter and the first parameter threshold is less than a preset difference threshold, then the target weight corresponding to the machine learning model is increased by a third value, and the target weight corresponding to the box model is decreased by the third value.

[0020] If the protection level is lower than the preset level threshold and the difference between the monitoring parameter and the second parameter threshold is less than the preset difference threshold, then the target weight corresponding to the machine learning model is increased by a fourth value, and the target weight corresponding to the box model is decreased by the fourth value, wherein the third value is greater than the fourth value;

[0021] The first diffusion data, the second diffusion data, and the third diffusion data are weighted and summed based on the adjusted target weights.

[0022] In one embodiment, the weighted summation of the first diffusion data, the second diffusion data, and the third diffusion data based on the target weight includes:

[0023] Obtain the transportation route planning information corresponding to the electrochemical energy storage box, and predict the route congestion level during the transportation of the electrochemical energy storage box based on the transportation route planning information;

[0024] If the congestion level is higher than a preset threshold, the target weight corresponding to the machine learning model is increased, and the target weight corresponding to the box model is decreased. The first diffusion data, the second diffusion data, and the third diffusion data are then weighted and summed based on the adjusted target weights.

[0025] In one embodiment, it also includes:

[0026] The release rate and composition changes of the toxic gas were simulated using a thermal runaway kinetic model based on the electrochemical reaction mechanism, battery material characteristics, and thermodynamic parameters within the electrochemical energy storage tank.

[0027] The process of obtaining first diffusion data of the toxic gas in the electrochemical energy storage box using a box model, second diffusion data of the toxic gas using a Gaussian plume model, and third diffusion data of the toxic gas using a machine learning model includes:

[0028] Based on the release rate and the composition change information, the vapor cloud density and initial volume parameters are adjusted in real time using the box model. Based on the adjusted vapor cloud density and initial volume parameters, the first diffusion data of the toxic gas in the electrochemical energy storage box is obtained.

[0029] The leakage source intensity and diffusion coefficient are adjusted in real time based on the release rate and the composition change information using the Gaussian plume model, and the second diffusion data of the toxic gas is obtained based on the adjusted leakage source intensity and diffusion coefficient.

[0030] The machine learning model captures the time-series characteristics of the thermal runaway process, and based on these time-series characteristics, predicts the third diffusion data of the toxic gas within a predetermined time period in the future.

[0031] In one embodiment, obtaining the toxic gas risk prediction result of the electrochemical energy storage tank based at least on the target diffusion data includes:

[0032] Based on the target diffusion data, the diffusion range of the toxic gas is determined, and the population and the number of sensitive areas within the diffusion range are determined. The risk exposure value is obtained by adding the product of the population and the preset exposure coefficient to the product of the number of sensitive areas and the preset sensitivity coefficient for each sensitive area. The preset exposure coefficient is set according to the diffusion concentration level, and the preset sensitivity coefficient is set according to the sensitive area type.

[0033] Based on geographic information system data, the interaction between the diffusion range and surrounding terrain and buildings is analyzed to obtain the risk impact range. Based on the functional importance and damage threshold of surrounding facilities, combined with the target diffusion data, the degree of impact of the toxic gas on the surrounding facilities is determined.

[0034] The risk exposure value, the risk impact range, and the facility impact degree are combined to form the toxic gas risk prediction result of the electrochemical energy storage box.

[0035] In one embodiment, obtaining the toxic gas risk prediction result of the electrochemical energy storage tank based at least on the target diffusion data includes:

[0036] The target diffusion data, the first diffusion data, the second diffusion data, the third diffusion data, the environmental information, the attribute information of the electrochemical energy storage box, and the historical leakage data of the electrochemical energy storage box are input into a large model to obtain the toxic gas risk prediction result of the electrochemical energy storage box output by the large model. The toxic gas risk prediction result includes at least one of the diffusion range, concentration distribution, and risk level of the toxic gas.

[0037] In one embodiment, the machine learning model includes a long short-term memory network, a convolutional neural network, and a Transformer model. The long short-term memory network is used to process time-series data during the transportation of the electrochemical energy storage box to capture the dynamic changes in the diffusion of the toxic gas. The convolutional neural network is used to extract spatial features during the transportation of the electrochemical energy storage box to analyze the spatial distribution pattern of the diffusion of the toxic gas. The Transformer model is used to learn the complex relationships in the multi-dimensional data during the transportation of the electrochemical energy storage box.

[0038] The technical solutions provided by the embodiments of this disclosure may include the following beneficial effects:

[0039] The method for predicting the risk of toxic gases in electrochemical energy storage boxes provided in this disclosure can fully leverage the advantages of each model by integrating a box model, a Gaussian plume model, and a machine learning model. This overcomes the limitation of a single model's predictive ability in complex scenarios, thereby improving the accuracy and reliability of toxic gas risk prediction for electrochemical energy storage boxes. Furthermore, considering the continuous changing characteristics of the environment during the transportation of electrochemical energy storage boxes (such as gradually moving from open suburbs to densely populated urban areas), the method extracts weight distributions for similar scenarios from a pre-set weighted scenario library and performs smooth interpolation. This ensures a smooth transition of model weights when switching between different scenarios, avoiding instability in prediction results due to sudden weight changes. This further improves the accuracy of toxic gas risk prediction for electrochemical energy storage boxes under different scenarios, providing a more accurate scientific basis for the safe transportation of electrochemical energy storage boxes. Attached Figure Description

[0040] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0041] Figure 1 This is a flowchart illustrating an exemplary embodiment of the present disclosure of a method for predicting the risk of toxic gases in an electrochemical energy storage box based on multi-model fusion. Detailed Implementation

[0042] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure.

[0043] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The singular forms “a,” “the,” and “the” as used herein are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0044] It should be understood that although the terms first, second, third, etc., may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are used only to distinguish information of the same type from one another. For example, without departing from the scope of this disclosure, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word “if” as used herein may be interpreted as “when”, “when…”, or “in response to determination.”

[0045] At least one embodiment of this disclosure provides a method for predicting the risk of toxic gases in electrochemical energy storage boxes based on multi-model fusion. Please refer to the appendix. Figure 1 It illustrates the flow of the method, including steps S101 to S104.

[0046] In step S101, the first diffusion data of toxic gas in the electrochemical energy storage box is obtained through the box model, the second diffusion data of toxic gas is obtained through the Gaussian plume model, and the third diffusion data of toxic gas is obtained through the machine learning model.

[0047] In step S102, based on the environmental information around the electrochemical energy storage box, a preset number of similar transportation scenarios similar to the current transportation scenario of the electrochemical energy storage box are searched in the preset weight scenario library, and the preset optimal weights corresponding to the box model, Gaussian plume model and machine learning model under the preset number of similar transportation scenarios are obtained respectively.

[0048] In step S103, interpolation calculation is performed based on the preset optimal weights and the similarity between the current transportation scenario and similar transportation scenarios to obtain the target weights corresponding to the box model, Gaussian plume model and machine learning model, respectively. The first diffusion data, the second diffusion data and the third diffusion data are then weighted and summed based on the target weights to obtain the target diffusion data.

[0049] In step S104, the toxic gas risk prediction result of the electrochemical energy storage box is obtained based at least on the target diffusion data.

[0050] Therefore, by integrating the box model, Gaussian plume model, and machine learning model, the advantages of each model can be fully utilized to overcome the limited predictive ability of a single model in complex scenarios, thereby improving the accuracy and reliability of toxic gas risk prediction for electrochemical energy storage boxes. Furthermore, considering the continuous changing characteristics of the scenarios during the transportation of electrochemical energy storage boxes (such as gradually moving from open suburbs to densely populated urban areas), the weight distributions of similar scenarios are extracted from a pre-set weight scenario library and smoothly interpolated. This ensures a smooth transition of model weights when switching between different scenarios, avoiding instability in prediction results caused by sudden weight changes. This, in turn, improves the accuracy of toxic gas risk prediction for electrochemical energy storage boxes under different scenarios, providing a more accurate scientific basis for the safe transportation of electrochemical energy storage boxes.

[0051] To facilitate understanding, the steps described above will be further explained below.

[0052] For example, before step S101, relevant data about the electrochemical energy storage box can be collected, including basic information such as dimensions, total mass, total lithium battery energy, composition and proportion of harmful substances, as well as environmental data along the transportation route, such as meteorological conditions (wind speed, wind direction, temperature, humidity, etc.), surrounding population density, and topography. The collected data can then be preprocessed, such as cleansing and normalization, to eliminate noise and outliers and improve data quality.

[0053] For example, in step S101, it can be assumed that after the electrochemical energy storage box explodes, the toxic gas rapidly and uniformly distributes throughout the space and the volume of the diffusion space remains unchanged. Therefore, the initial diffusion radius of the toxic gas can be calculated according to the box model formula. Accordingly, the first diffusion data includes the diffusion radius of the toxic gas.

[0054] The processing steps for the box model include: first, determining the geometric center of the electrochemical energy storage box as the diffusion center based on its size and geometry; then, calculating the vapor cloud density by considering the composition and proportion of harmful substances within the electrochemical energy storage box; finally, using the following box model formula, considering factors such as gravitational acceleration, vapor cloud density, ambient air density, initial radius of the heavy gas cloud, initial volume of the heavy gas cloud formed by the leak, and diffusion time, calculating the diffusion radius of the toxic gas:

[0055] (1)

[0056] in, The diffusion radius of the toxic gas. For gravitational acceleration, , These are vapor cloud density and ambient air density, respectively, in units of... kg / m , The initial radius of the heavy gas cloud, in units of . m , The initial volume unit for the formation of a heavy gas cloud from the leak is... m 3 , The time taken for the initial cloud to disperse after the leak ended, in units of s .

[0057] It should be understood that the density of a vapor cloud can be calculated using the density formula. Calculate ambient air density using the ideal gas law. The initial radius R0 can be calculated based on the standard dimensions of the electrochemical energy storage box. The electrochemical energy storage box is equivalent to a standardized rectangular box. After the explosion, the toxic gas first accumulates inside the box and then diffuses. Therefore, the initial cloud boundary is strongly correlated with the geometric boundary of the box.

[0058] For example, in step S101, it can be assumed that the harmful gas is leaking continuously, and the diffusion radius of the harmful gas can be calculated using the Gaussian plume model formula based on the concentration of the harmful substance components. Accordingly, the second diffusion data includes the diffusion radius of the toxic gas.

[0059] The processing steps of the Gaussian plume model include: establishing a three-dimensional coordinate system by determining the projection of the leak source onto the ground as the origin; determining the concentration of harmful gases using the following Gaussian plume model formula by combining parameters such as the concentration of harmful substances, the intensity of the leak source, the height of the leak source, the average wind speed, and the diffusion coefficient; and then calculating the diffusion radius of the harmful gases based on this concentration.

[0060] (2)

[0061] in, Let represent the concentration of harmful gases at the leak source, and take the projection of the leak source onto the ground as the origin of the coordinate system. x The axis points in the wind direction. y The axis represents the direction perpendicular to the wind direction in the horizontal plane. z The axis points in a direction perpendicular to the horizontal plane. x Downwind distance y Crosswind distance, z Height above the ground , , for x axis, y shaft and z Diffusion coefficient on the axis, The strength of the leak source, H The height of the leakage source, i.e., the height of the vehicle transporting the electrochemical energy storage box. u The average wind speed at the leakage height indicates that the leakage source is the electrochemical energy storage tank where the explosion occurred.

[0062] It should be understood that the proportions of toxic gas components vary at different stages of thermal runaway. For example, in the early stages of thermal runaway, HF accounts for 15% and CO for 10%, while in the middle stages, HF rises to 25% and CO to 18%. Therefore, for the hazardous gas concentrations in the above calculation formula, the concentrations of each hazardous gas can be estimated first based on the size of the electrochemical energy storage tank and the proportion of hazardous gases contained within it, and then the concentration of the hazardous gas with the highest proportion can be taken. The leakage source intensity can be calculated based on the leakage duration and the total mass or volume of the leaked material.

[0063] It should also be understood that the electrochemical energy storage box is fixedly installed on the cargo platform of the transport vehicle. In the event of thermal runaway leakage, the gas released from the box first diffuses to the height of the cargo platform and then diffuses with the wind. In this embodiment, the height of the leakage source is fixed to the height of the vehicle transporting the electrochemical energy storage box. For example, in the case of highway transportation, the default height is 2.5-4m, with 2.5m for light trucks and 4m for heavy trucks. The value can be selected according to the actual situation to ensure that it matches the actual transport carrier and avoid the diffusion radius error caused by the height assumption deviation in the general model. This can improve the accuracy of the final prediction of the risk of toxic gas from the electrochemical energy storage box.

[0064] In some embodiments, considering the safety risks of electrochemical energy storage boxes in road transport scenarios, and given that the energy storage boxes are fixedly installed on transport vehicles and have a specific ground clearance, the toxic gas leakage source has a defined initial height characteristic in the event of a sudden thermal runaway accident. Based on this, an improved Gaussian plume diffusion model can be used to derive the three-dimensional concentration distribution of typical harmful gases such as HF and CO under downwind conditions. Specifically, the standardized Gaussian plume model can be transformed into the following form: .

[0065] For example, before step S101, a large amount of data on toxic gas leak accidents and simulations in electrochemical energy storage boxes can be collected as training datasets for the machine learning model. The training dataset is then input into the machine learning model for training. By adjusting the model's hyperparameters and optimizing the algorithm, the model's prediction accuracy and generalization ability can be improved. Furthermore, techniques such as cross-validation and early stopping can be used to prevent overfitting and ensure the predictive performance of the machine learning model on unknown data.

[0066] In some embodiments, the machine learning model may include a long short-term memory network, a convolutional neural network, and a Transformer model. The long short-term memory network is used to process time-series data during the transportation of the electrochemical energy storage box to capture the dynamic changes in the diffusion of toxic gases. The convolutional neural network is used to extract spatial features during the transportation of the electrochemical energy storage box to analyze the spatial distribution patterns of the diffusion of toxic gases. The Transformer model is used to learn the complex relationships in the multi-dimensional data during the transportation of the electrochemical energy storage box.

[0067] For example, time-series data could include temperature and wind speed changes of the electrochemical energy storage tank over the past 10 minutes. Spatial features could include the location characteristics of sensitive areas such as schools and hospitals along the transportation route of the electrochemical energy storage tank. Multi-dimensional data could include the aforementioned time-series data and spatial features.

[0068] For example, the input to a machine learning model can include features related to the electrochemical energy storage tank, such as its size, volume, and temperature; environmental features, such as meteorological features like wind speed and temperature, and / or topographical features like road layout and the density of surrounding buildings; accident features, such as the intensity and duration of the leak source; time-series features; and spatial features. Thus, through detailed input features, the machine learning model can comprehensively understand the accident scenario and environmental conditions, thereby more accurately predicting the diffusion range of toxic gases, concentration changes at different time points, and risk assessment results. Furthermore, these input features provide the machine learning model with rich information, enabling it to learn the complex patterns and laws of toxic gas diffusion.

[0069] For example, the machine learning model, including Long Short-Term Memory (LSTM) networks, Convolutional Neural Networks (CNNs), and Transformer models, can be configured in parallel. Thus, the LTM network can capture the dynamic changes in toxic gas diffusion based on time-series data. Simultaneously, the CNN can analyze the spatial distribution patterns of toxic gas diffusion based on spatial features, and the Transformer model can learn the complex relationships in multi-dimensional data during the transportation of electrochemical energy storage boxes. Then, the machine learning model fuses the feature data from the LTM network, CNN, and Transformer model through a fusion layer, ultimately outputting a single result.

[0070] For example, the output of a machine learning model may include the diffusion radius of a toxic gas; predictions of toxic gas concentrations at different time points, such as an increase in concentration at a location from 0.5 ppm to 1.0 ppm within 30 minutes of a leak; and risk level results, such as high risk. Accordingly, third-party diffusion data includes the diffusion radius, concentration predictions, and risk level of the toxic gas.

[0071] In some embodiments, a thermal runaway kinetic model can be used to simulate the release rate and compositional changes of toxic gases based on the electrochemical reaction mechanism, battery material characteristics, and thermodynamic parameters within the electrochemical energy storage tank. Accordingly, in step S101, a box model can be used to adjust the vapor cloud density and initial volume parameters in real time based on the release rate and compositional changes, and the first diffusion data of the toxic gases in the electrochemical energy storage tank can be obtained based on the adjusted vapor cloud density and initial volume parameters; a Gaussian plume model can be used to adjust the leakage source intensity and diffusion coefficient in real time based on the release rate and compositional changes, and the second diffusion data of the toxic gases can be obtained based on the adjusted leakage source intensity and diffusion coefficient; a machine learning model can be used to capture the time-series characteristics of the thermal runaway process, and based on these time-series characteristics, the third diffusion data of the toxic gases within a predetermined time period can be predicted.

[0072] It should be understood that during thermal runaway, the internal temperature and pressure of an electrochemical energy storage tank rise sharply, causing the release rate and composition of harmful gases to change over time. Introducing a thermal runaway kinetic model to simulate the harmful gas release process in real time, and feeding back the dynamic information of the release rate and composition to the tank model and Gaussian plume model respectively, while simultaneously using a machine learning model to capture the time-series characteristics of the thermal runaway process, can improve the accuracy of predicting the risk of toxic gas diffusion throughout the entire thermal runaway process.

[0073] For example, before step S102, a large amount of data from past transportation of electrochemical energy storage boxes under different scenarios can be collected, including various transitional scenarios from open suburbs to densely populated urban areas. Cluster analysis is performed on this data to group scenarios with similar characteristics into one category, forming several typical training scenarios. The optimal model weight combination is then assigned to each typical training scenario. For instance, 10 typical training scenarios are obtained through cluster analysis, each with a corresponding weight distribution. For example, scenario A corresponds to a box model weight of 0.4, a Gaussian plume model weight of 0.3, and a machine learning model weight of 0.3; scenario B corresponds to a box model weight of 0.35, a Gaussian plume model weight of 0.25, and a machine learning model weight of 0.4, etc.

[0074] For example, in step S102, cosine similarity can be used to calculate the similarity between the current transportation scenario and each scenario in the typical training scenario library. Then, based on the similarity magnitude, the weight distributions corresponding to the preset number of similar transportation scenarios with the highest similarity are extracted from the typical training scenario library. The preset number can be set according to actual conditions, such as 3, but this embodiment does not limit this setting.

[0075] In some embodiments, in step S103, the sum of similarities between the current transportation scenario and each similar transportation scenario can be used as a normalization parameter. For each similar transportation scenario, the result of dividing the similarity between the current transportation scenario and the similar transportation scenario by the normalization parameter is used as the normalized similarity weight corresponding to the similar transportation scenario. Based on the normalized similarity weight, the preset optimal weights corresponding to the box model under each similar transportation scenario are weighted and summed to obtain the target weight corresponding to the box model. Based on the normalized similarity weight, the preset optimal weights corresponding to the Gaussian plume model under each similar transportation scenario are weighted and summed to obtain the target weight corresponding to the Gaussian plume model. Based on the normalized similarity weight, the preset optimal weights corresponding to the machine learning model under each similar transportation scenario are weighted and summed to obtain the target weight corresponding to the machine learning model.

[0076] For example, the current transportation scenario has the highest similarity of 0.8 with a typical open suburban scenario, followed by a similarity of 0.7 with transitional scenario C, and then a similarity of 0.6 with transitional scenario D. Therefore, the corresponding preset optimal weights are extracted from these three similar transportation scenarios. Linear interpolation is then performed on the extracted preset optimal weights, or more complex interpolation methods, such as spline interpolation, can also be used; this disclosure does not limit the specific methods used.

[0077] Taking linear interpolation as an example, the box model weight corresponding to the typical open suburban scenario is 0.4, the box model weight corresponding to transition scenario C is 0.38, and the box model weight corresponding to transition scenario D is 0.35. The similarity of the three similar transportation scenarios is 0.8, 0.7, and 0.6, respectively, with a total of 2.1, meaning the normalization parameter is 2.1. Correspondingly, the normalized similarity weights of each similar transportation scenario are: 0.8 / 2.1≈0.381, 0.7 / 2.1≈0.333, and 0.6 / 2.1≈0.286. Therefore, the target weight of the interpolated box model is: 0.381×0.4+0.333×0.38+0.286×0.35≈0.379. Similarly, similar interpolation calculations are performed on the target weights of the Gaussian plume model and the machine learning model to obtain the weights of each model after a smooth transition in the current transportation scenario.

[0078] This enables a smooth transition of model weights when switching between different scenarios, avoiding instability in prediction results caused by sudden changes in weights. Consequently, it improves the accuracy of predicting the risk of toxic gases in electrochemical energy storage boxes under different scenarios, providing a more accurate scientific basis for the safe transportation of electrochemical energy storage boxes.

[0079] For example, in step S103, the diffusion radii included in the first diffusion data, the second diffusion data, and the third diffusion data can be weighted and summed based on the target weight to obtain the target diffusion radius. Accordingly, in step S104, the toxic gas risk prediction result of the electrochemical energy storage box can be obtained at least based on the target diffusion radius.

[0080] In some embodiments, in step S103, the confidence scores of the box model, Gaussian plume model, and machine learning model can be calculated using the Bayesian confidence assessment method, based on the historical accuracy of the box model, Gaussian plume model, and machine learning model, as well as the deviation between the current results and the actual observations. If there is a target result confidence score that is less than a preset confidence threshold, the target weight of the model corresponding to the target result confidence score is reduced according to a preset mapping relationship. The preset mapping relationship indicates that for every first decrease in the target result confidence score compared to the preset initial confidence score, the target weight is reduced by a second value, and the first value is different from the second value. The first diffusion data, the second diffusion data, and the third diffusion data are weighted and summed based on the reduced target weight.

[0081] For example, the accuracy of historical results can be determined based on the historical prediction results of each model.

[0082] For example, the preset initial confidence level can be determined based on the average error of each model in the historical prediction process, or it can be set based on human experience; this disclosure does not limit this. Taking the average error as an example, the average error of the box model in the past 10 predictions is 10%, the average error of the Gaussian plume model in the past 10 predictions is 8%, and the average error of the machine learning model in the past 10 predictions is 12%. Based on the magnitude of the prediction error, an initial confidence level can be assigned to each model: 0.4 for the box model (moderate prediction error), 0.5 for the Gaussian plume model (small prediction error), and 0.3 for the machine learning model (large prediction error). In addition, to ensure that the sum of the confidence levels is 1, the initial confidence levels can also be normalized. Thus, the initial confidence level will serve as the prior probability for Bayesian confidence evaluation, used for subsequent confidence level calculations and model weight adjustments.

[0083] For example, actual observations can be gas diffusion parameters such as gas concentration obtained through gas sensors. It should be understood that actual observations only provide information on the diffusion of toxic gases at a specific time and location, and cannot comprehensively reflect the entire diffusion range and future trends. Furthermore, the acquisition of observations may be limited by various factors, such as sensor cost, deployment difficulty, and environmental severity. Therefore, this disclosure provides a method for predicting the risk of toxic gases in electrochemical energy storage boxes based on multi-model fusion, using actual observations as real-time feedback to correct and optimize the results of multi-model fusion in real time, thereby improving the accuracy of predicting the risk of toxic gases in electrochemical energy storage boxes.

[0084] For example, the preset mapping relationship can be obtained by fitting a large amount of historical data. For instance, the weight can be reduced by 0.05 for every 0.1 decrease in confidence. If the confidence decrease is less than 0.1, the weight can be left unchanged. If the confidence decrease is greater than 0.1 but less than 0.2, the weight can be reduced by 0.05, and so on. This disclosure does not limit the specific implementation of the embodiments.

[0085] For example, continuing with the above example, the target weight of the box model is 0.379. The confidence level obtained based on the Bayesian confidence assessment method is 0.3, which is 0.1 lower than the initial confidence level of 0.4. The target weight is reduced by 0.05, so the new target weight is 0.379 - 0.05 = 0.329.

[0086] Therefore, the confidence level of each model's prediction results can be quantified using the Bayesian confidence assessment method. If the prediction confidence of each model is high, there is no need to adjust the target weights of each model. If the prediction confidence of a certain model is low, its weight contribution in the fusion result can be adaptively reduced according to a preset mapping relationship, thereby improving the accuracy of predicting the risk of toxic gas diffusion in electrochemical energy storage boxes.

[0087] In some embodiments, in step S103, the protection level of the electrochemical energy storage box can be determined based on the characteristics of the packaging material of the electrochemical energy storage box, and the monitoring parameters of the protection device of the electrochemical energy storage box can be obtained; if the protection level is higher than or equal to a preset level threshold and the difference between the monitoring parameter and the first parameter threshold is less than a preset difference threshold, the target weight corresponding to the machine learning model is increased by a third value, and the target weight corresponding to the box model is decreased by a third value, and the first diffusion data, the second diffusion data, and the third diffusion data are weighted and summed based on the adjusted target weight; if the protection level is lower than a preset level threshold and the difference between the monitoring parameter and the second parameter threshold is less than a preset difference threshold, the target weight corresponding to the machine learning model is increased by a fourth value, and the target weight corresponding to the box model is decreased by a fourth value, and the first diffusion data, the second diffusion data, and the third diffusion data are weighted and summed based on the adjusted target weight, wherein the third value is greater than the fourth value.

[0088] It should be understood that electrochemical energy storage boxes of different specifications and protection levels have varying capabilities in suppressing harmful gas leakage during transportation. This disclosure uses the characteristics of the packaging materials and monitoring data of the protective devices of the electrochemical energy storage boxes as one of the bases for adjusting the weights of each model. This allows for a more accurate reflection of the harmful gas leakage and diffusion situation of electrochemical energy storage boxes of different specifications and protection levels during transportation, thereby improving the accuracy and practicality of harmful gas risk prediction for electrochemical energy storage boxes.

[0089] For example, the characteristics of packaging materials can be leakage prevention performance, pressure resistance, etc. For instance, the protection level of the electrochemical energy storage box can be predicted based on the characteristics of packaging materials using a large model, or the correspondence between packaging material characteristics and protection levels can be pre-configured, and then the protection level of the electrochemical energy storage box can be determined based on the correspondence. This disclosure does not limit this aspect.

[0090] For example, the monitoring parameters of the protective device may be the opening pressure of the pressure relief valve, the sealing level, etc.

[0091] For example, the preset level threshold, the first parameter threshold, the preset difference threshold, the first value, and the second value can be set according to requirements, and this embodiment does not limit them.

[0092] For example, for a high-protection-level energy storage tank, the target weight for the machine learning model is 0.2, for the tank-type model it is 0.5, and for the Gaussian plume model it is 0.3. The monitoring system indicates that the internal pressure of the energy storage tank is gradually approaching the pressure relief valve opening pressure (e.g., 1.8 MPa), increasing the risk of protective device failure. Based on the monitoring parameters and the pre-set weight adjustment strategy, the weight of the machine learning model is gradually increased to 0.4, the weight of the tank-type model is decreased to 0.3, and the weight of the Gaussian plume model remains at 0.3. Thus, the machine learning model begins to consider more of the complexity of leakage diffusion, such as the impact of pressure changes on the leakage rate, to more accurately reflect the actual leakage diffusion situation.

[0093] For example, for a low-protection-level energy storage tank, the target weight for the machine learning model is 0.4, for the box-type model it is 0.3, and for the Gaussian plume model it is 0.3. The monitoring system shows that the internal pressure is approaching the pressure relief valve opening pressure (e.g., 0.6 MPa), significantly increasing the risk of protective device failure. Based on the monitoring parameters and a pre-set weight adjustment strategy, the weight of the machine learning model is increased to 0.5, the weight of the box-type model is reduced to 0.2, and the weight of the Gaussian plume model remains at 0.3. Thus, the machine learning model begins to consider more of the complexity of leakage diffusion, such as the impact of pressure changes on the leakage rate, to more accurately reflect the actual leakage diffusion situation.

[0094] It should be understood that machine learning models are primarily used to capture complex leakage and diffusion patterns, while box-type models mainly predict leakage based on the ideal sealing state of the energy storage box. Therefore, for energy storage boxes with high protection levels, due to their excellent protective performance, the risk of leakage is low in the initial stages of transportation, and the prediction results of box-type models are more optimistic, thus giving them a higher initial weight. For energy storage boxes with low protection levels, due to their poor protective performance, the risk of leakage is higher in the initial stages, thus giving machine learning models a higher initial weight.

[0095] In some embodiments, in step S103, the transportation route planning information corresponding to the electrochemical energy storage box can be obtained, and the degree of route congestion during the transportation of the electrochemical energy storage box can be predicted based on the transportation route planning information; if the degree of congestion is higher than a preset degree threshold, the target weight corresponding to the machine learning model is increased, and the target weight corresponding to the box model is decreased, and the first diffusion data, the second diffusion data and the third diffusion data are weighted and summed based on the adjusted target weight.

[0096] For example, route planning information includes road condition information and traffic flow of the planned route. For instance, a large model can be used to predict the degree of route congestion during the transportation of electrochemical energy storage boxes based on the transportation route planning information, or the degree of route congestion during the transportation of electrochemical energy storage boxes can be obtained from a map application based on the transportation route planning information. This disclosure does not limit the scope of the embodiments.

[0097] For example, the preset threshold can be set according to requirements, and this disclosure does not limit it.

[0098] It should be understood that the transportation route of electrochemical energy storage boxes is usually dynamically adjusted based on factors such as road conditions and traffic flow. Therefore, in the process of predicting the risk of toxic gas diffusion, route planning information can be used as one of the bases for weight adjustment. For example, in congested road sections, the possibility of accumulating risk of harmful gas diffusion increases due to increased vehicle dwell time. In this case, the weight of the machine learning model can be increased to enhance its predictive ability of diffusion pattern changes under congestion scenarios, while the weight of the box model based on the ideal leakage assumption can be reduced. In addition, the diffusion coefficient of the Gaussian plume model can be adjusted according to the wind field disturbance caused by congestion to better adapt to the impact of local wind field changes on the diffusion of toxic gases.

[0099] Therefore, by interacting with the transportation route planning system in real time, the weights of each model can be dynamically adapted to changes in the route, providing accurate risk warnings for the safe transportation of electrochemical energy storage boxes throughout the entire process.

[0100] In some embodiments, in step S104, the diffusion range of toxic gas can be determined based on target diffusion data, and the population and number of sensitive areas within the diffusion range can be determined. The risk exposure value is obtained by multiplying the population by a preset exposure coefficient and adding the product of the number of sensitive areas and the preset sensitivity coefficient for each sensitive area. The preset exposure coefficient is set according to the diffusion concentration level, and the preset sensitivity coefficient is set according to the sensitive area type. Based on geographic information system data, the interaction between the diffusion range and surrounding terrain and buildings is analyzed to obtain the risk impact range. Based on the functional importance and damage threshold of surrounding facilities, combined with the target diffusion data, the degree of facility impact of toxic gas on surrounding facilities is determined. The risk exposure value, risk impact range, and facility impact degree are combined to form the toxic gas risk prediction result of the electrochemical energy storage box.

[0101] For example, the spread range of toxic gases can be predicted using large models based on target diffusion data.

[0102] For example, the area surrounding the transportation route can be divided into grids, and the population within each grid can be counted. Correspondingly, a preset exposure coefficient can be assigned to each grid. For instance, with three grid areas, the first grid area has a toxic gas concentration below the safety threshold, with a preset exposure coefficient of 0.1; the second grid area has a toxic gas concentration between the safety and danger thresholds, with a preset exposure coefficient of 0.6; and the third grid area has a toxic gas concentration above the danger threshold, with a preset exposure coefficient of 1.2. The preset exposure coefficient can be set according to actual conditions, and this embodiment does not limit this.

[0103] For example, sensitive areas may include schools, hospitals, residential areas, etc. Different types of sensitive areas correspond to different preset sensitivity coefficients. For instance, the preset sensitivity coefficient for schools is 2.5, for hospitals it is 3.0, and for residential areas it is 2. The preset sensitivity coefficients can be set according to actual conditions, and this embodiment does not limit this.

[0104] For example, the risk exposure value represents the number of people exposed and the number of sensitive areas within the range of toxic gas diffusion. It can be obtained as follows: multiply the number of people corresponding to each grid area by a preset exposure coefficient and sum them to obtain a first sum; multiply the number of sensitive areas corresponding to each sensitive area by a preset sensitivity coefficient and sum them to obtain a second sum; finally, add the first sum and the second sum to obtain the risk exposure value.

[0105] For example, the risk exposure value, risk impact range, and facility impact degree can be directly combined into a toxic gas risk prediction result for the electrochemical energy storage box for output; or, the risk exposure value, risk impact range, and facility impact degree can be input into a large model to obtain the toxic gas risk prediction result for the electrochemical energy storage box output by the large model.

[0106] In some embodiments, in step S104, target diffusion data, first diffusion data, second diffusion data, third diffusion data, environmental information, attribute information of the electrochemical energy storage box, and historical leakage data of the electrochemical energy storage box can be input into the large model to obtain the toxic gas risk prediction result of the electrochemical energy storage box output by the large model. The toxic gas risk prediction result includes at least one of the diffusion range, concentration distribution, and risk level of the toxic gas.

[0107] Therefore, the complex nonlinear relationship between the outputs of the three models can be captured by a large model, which can improve the accuracy and reliability of predicting the risk of toxic gases in electrochemical energy storage boxes and provide stronger protection for the transportation safety of electrochemical energy storage boxes.

[0108] The preferred embodiments of this disclosure have been described in detail above with reference to the accompanying drawings. However, this disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this disclosure, various simple modifications can be made to the technical solutions of this disclosure, and these simple modifications all fall within the protection scope of this disclosure.

[0109] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, this disclosure will not describe the various possible combinations separately.

[0110] Furthermore, various different embodiments of this disclosure can be combined in any way, as long as they do not violate the spirit of this disclosure, they should also be regarded as the content disclosed in this disclosure.

Claims

1. A method for predicting the risk of toxic gases in an electrochemical energy storage box based on multi-model fusion, characterized in that, include: The first diffusion data of the toxic gas in the electrochemical energy storage box was obtained by using a box model, the second diffusion data of the toxic gas was obtained by using a Gaussian plume model, and the third diffusion data of the toxic gas was obtained by using a machine learning model. Based on the environmental information around the electrochemical energy storage box, a preset number of similar transportation scenarios similar to the current transportation scenario of the electrochemical energy storage box are searched in a preset weighted scenario library, and the preset optimal weights corresponding to the box model, the Gaussian plume model and the machine learning model under the preset number of similar transportation scenarios are obtained respectively. Based on the preset optimal weights and the similarity between the current transportation scenario and the similar transportation scenario, interpolation calculations are performed to obtain the target weights corresponding to the box model, the Gaussian plume model and the machine learning model, respectively. Then, the first diffusion data, the second diffusion data and the third diffusion data are weighted and summed based on the target weights to obtain the target diffusion data. Based at least on the target diffusion data, the toxic gas risk prediction results of the electrochemical energy storage box are obtained.

2. The method for predicting the risk of toxic gases in electrochemical energy storage boxes based on multi-model fusion as described in claim 1, characterized in that, The interpolation calculation based on the preset optimal weights and the similarity between the current transportation scenario and the similar transportation scenarios yields the target weights corresponding to the box model, the Gaussian plume model, and the machine learning model, respectively, including: The sum of similarities between the current transportation scenario and each of the similar transportation scenarios is used as a normalization parameter. For each of the similar transportation scenarios, the result of dividing the similarity between the current transportation scenario and the similar transportation scenario by the normalization parameter is used as the normalized similarity weight corresponding to the similar transportation scenario. Based on the normalized similarity weights, the preset optimal weights corresponding to the box-type models in each of the similar transportation scenarios are weighted and summed to obtain the target weights corresponding to the box-type models. Similarly, based on the normalized similarity weights, the preset optimal weights corresponding to the Gaussian plume models in each of the similar transportation scenarios are weighted and summed to obtain the target weights corresponding to the Gaussian plume models. Finally, based on the normalized similarity weights, the preset optimal weights corresponding to the machine learning models in each of the similar transportation scenarios are weighted and summed to obtain the target weights corresponding to the machine learning models.

3. The method for predicting the risk of toxic gases in electrochemical energy storage boxes based on multi-model fusion as described in claim 1, characterized in that, The weighted summation of the first diffusion data, the second diffusion data, and the third diffusion data based on the target weight includes: The confidence scores of the box model, Gaussian plume model, and machine learning model are calculated using the Bayesian confidence assessment method, based on the historical accuracy of the box model, Gaussian plume model, and machine learning model, as well as the deviation between the current results and the actual observations. If there is a target result confidence level in the result confidence level that is less than a preset confidence level threshold, then according to a preset mapping relationship, the target weight of the model corresponding to the target result confidence level is reduced. The preset mapping relationship indicates that for every first decrease in the target result confidence level compared to the preset initial confidence level, the target weight is reduced by a second value, and the first value is different from the second value. The first diffusion data, the second diffusion data, and the third diffusion data are weighted and summed based on the reduced target weights.

4. The method for predicting the risk of toxic gases in electrochemical energy storage boxes based on multi-model fusion as described in claim 1, characterized in that, The weighted summation of the first diffusion data, the second diffusion data, and the third diffusion data based on the target weight includes: Based on the characteristics of the packaging material of the electrochemical energy storage box, the protection level of the electrochemical energy storage box is determined, and the monitoring parameters of the protection device of the electrochemical energy storage box are obtained. If the protection level is higher than or equal to a preset level threshold and the difference between the monitoring parameter and the first parameter threshold is less than a preset difference threshold, then the target weight corresponding to the machine learning model is increased by a third value, and the target weight corresponding to the box model is decreased by the third value. If the protection level is lower than the preset level threshold and the difference between the monitoring parameter and the second parameter threshold is less than the preset difference threshold, then the target weight corresponding to the machine learning model is increased by a fourth value, and the target weight corresponding to the box model is decreased by the fourth value, wherein the third value is greater than the fourth value; The first diffusion data, the second diffusion data, and the third diffusion data are weighted and summed based on the adjusted target weights.

5. The method for predicting the risk of toxic gases in electrochemical energy storage boxes based on multi-model fusion according to claim 1, characterized in that, The weighted summation of the first diffusion data, the second diffusion data, and the third diffusion data based on the target weight includes: Obtain the transportation route planning information corresponding to the electrochemical energy storage box, and predict the route congestion level during the transportation of the electrochemical energy storage box based on the transportation route planning information; If the congestion level is higher than a preset threshold, the target weight corresponding to the machine learning model is increased, and the target weight corresponding to the box model is decreased. The first diffusion data, the second diffusion data, and the third diffusion data are weighted and summed based on the adjusted target weights.

6. The method for predicting the risk of toxic gases in electrochemical energy storage boxes based on multi-model fusion according to any one of claims 1-5, characterized in that, Also includes: The release rate and composition changes of the toxic gas were simulated using a thermal runaway kinetic model based on the electrochemical reaction mechanism, battery material characteristics, and thermodynamic parameters within the electrochemical energy storage tank. The process of obtaining first diffusion data of the toxic gas in the electrochemical energy storage box using a box model, second diffusion data of the toxic gas using a Gaussian plume model, and third diffusion data of the toxic gas using a machine learning model includes: Based on the release rate and the composition change information, the vapor cloud density and initial volume parameters are adjusted in real time using the box model. Based on the adjusted vapor cloud density and initial volume parameters, the first diffusion data of the toxic gas in the electrochemical energy storage box is obtained. The leakage source intensity and diffusion coefficient are adjusted in real time based on the release rate and the composition change information using the Gaussian plume model, and the second diffusion data of the toxic gas is obtained based on the adjusted leakage source intensity and diffusion coefficient. The machine learning model captures the time-series characteristics of the thermal runaway process, and based on these time-series characteristics, predicts the third diffusion data of the toxic gas within a predetermined time period in the future.

7. The method for predicting the risk of toxic gases in electrochemical energy storage boxes based on multi-model fusion according to any one of claims 1-5, characterized in that, The process of obtaining the toxic gas risk prediction result of the electrochemical energy storage box based at least on the target diffusion data includes: Based on the target diffusion data, the diffusion range of the toxic gas is determined, and the population and the number of sensitive areas within the diffusion range are determined. The risk exposure value is obtained by adding the product of the population and the preset exposure coefficient to the product of the number of sensitive areas and the preset sensitivity coefficient for each sensitive area. The preset exposure coefficient is set according to the diffusion concentration level, and the preset sensitivity coefficient is set according to the sensitive area type. Based on geographic information system data, the interaction between the diffusion range and surrounding terrain and buildings is analyzed to obtain the risk impact range. Based on the functional importance and damage threshold of surrounding facilities, combined with the target diffusion data, the degree of impact of the toxic gas on the surrounding facilities is determined. The risk exposure value, the risk impact range, and the facility impact degree are combined to form the toxic gas risk prediction result of the electrochemical energy storage box.

8. The method for predicting the risk of toxic gases in an electrochemical energy storage box based on multi-model fusion according to any one of claims 1-5, characterized in that, The process of obtaining the toxic gas risk prediction result of the electrochemical energy storage box based at least on the target diffusion data includes: The target diffusion data, the first diffusion data, the second diffusion data, the third diffusion data, the environmental information, the attribute information of the electrochemical energy storage box, and the historical leakage data of the electrochemical energy storage box are input into a large model to obtain the toxic gas risk prediction result of the electrochemical energy storage box output by the large model. The toxic gas risk prediction result includes at least one of the diffusion range, concentration distribution, and risk level of the toxic gas.

9. The method for predicting the risk of toxic gases in electrochemical energy storage boxes based on multi-model fusion according to any one of claims 1-5, characterized in that, The machine learning model includes a long short-term memory network, a convolutional neural network, and a Transformer model. The long short-term memory network is used to process time-series data during the transportation of the electrochemical energy storage box to capture the dynamic changes in the diffusion of the toxic gas. The convolutional neural network is used to extract spatial features during the transportation of the electrochemical energy storage box to analyze the spatial distribution pattern of the diffusion of the toxic gas. The Transformer model is used to learn the complex relationships in the multi-dimensional data during the transportation of the electrochemical energy storage box.

Images