Medical simulator battery life prediction method based on dynamic weight and physical constraint

Abstract

The invention discloses a medical simulator battery life prediction method based on dynamic weight and physical constraint, and relates to the field of medical equipment battery health management, and the method comprises the steps: building a dynamic fusion weight, a battery capacity physical prediction model and a battery capacity LSTM prediction model according to the physical characteristics of a medical simulator battery; the battery capacity physical prediction model and the LSTM battery capacity prediction model are fused through dynamic fusion weights, a physical constraint loss function, an LSTM loss function and a joint loss function are established, parameters of the prediction models are updated through the loss functions, and finally a medical simulator battery life prediction result is obtained. Through fusion of the battery capacity physical prediction model and the LSTM battery capacity prediction model, the problems that time sequence characteristics of a traditional LSTM model are insufficient to capture and a pure data driving method neglects a battery physical mechanism are solved, and the accuracy and reliability of medical simulator battery life prediction are improved.

Description

Technical Field

[0001] The present invention relates to the field of medical device battery health management, and more specifically, to a method for predicting the battery life of a medical simulator based on dynamic weights and physical constraints. Background Art

[0002] Medical simulators are devices used to train medical professionals. They can simulate real medical scenarios, enabling medical service providers to learn, practice, and evaluate medical skills in a safe environment, which helps to limit the risk of accidents when medical staff are not fully trained. At the same time, the battery operating temperature range of medical simulators is wide, generally 10 - 40°C, and it needs to maintain stability under simulated extreme test conditions; the charge and discharge mode is high-frequency intermittent, with 3 - 5 charge and discharge cycles per test period, which is significantly different from the daily usage scenario of the battery of real medical devices. It needs to support long-term continuous testing, and the battery life directly affects the test accuracy of the simulator. Accurately predicting the battery life of medical simulators helps to arrange battery maintenance and replacement in advance, reduce equipment downtime, ensure the smooth progress of medical training and testing processes, and also improve the safety and performance of medical simulators.

[0003] There are many deficiencies in the existing technologies in the field of predicting the battery life of medical simulators. The traditional threshold method is too simple and can only identify obvious decreases in battery capacity, unable to capture the slight capacity attenuation of simulator batteries; the industrial hybrid model has insufficient capture ability when facing the high-frequency charge and discharge of medical simulators, with a high false alarm rate and low accuracy of prediction results; pure physical models, such as the Arrhenius equation, are based on theoretical assumptions and ignore the non-linear characteristics of the actual test data of medical simulators, resulting in limited prediction accuracy. In addition, the traditional LSTM model does not fully capture the temporal characteristics of the high-frequency charge and discharge of medical simulators and cannot accurately fit the actual working conditions of simulator batteries. At the same time, pure data-driven methods ignore the non-linear impact of temperature gradients on the thickening rate of the battery SEI film, weakening the accuracy and reliability of prediction. Summary of the Invention

[0004] The present invention provides a method for predicting the battery life of a medical simulator based on dynamic weights and physical constraints. By fusing a physical prediction model of battery capacity and an LSTM prediction model of battery capacity, it solves the problems of insufficient capture of temporal characteristics by the traditional LSTM model and the neglect of battery physical mechanisms by pure data-driven methods, and improves the accuracy and reliability of predicting the battery life of medical simulators.

[0005] To achieve the above object, the present invention provides a method for predicting the battery life of a medical simulator based on dynamic weights and physical constraints, including: According to the physical characteristics of the battery of the medical simulator, establish a dynamic fusion weight, a physical prediction model of battery capacity, and an LSTM prediction model of battery capacity; According to the dynamic fusion weight, fuse the physical prediction capacity of the battery capacity physical prediction model and the LSTM prediction capacity of the battery capacity LSTM prediction model to obtain the final battery capacity prediction result; According to the final battery capacity prediction result, establish the physical constraint loss function of the battery capacity physical prediction model and the LSTM loss function of the battery capacity LSTM prediction model, and establish a joint loss function according to the physical constraint loss function and the LSTM loss function; Update the parameters of the battery capacity physical prediction model according to the joint loss function, and update the parameters of the battery capacity LSTM prediction model according to the joint loss function to obtain the updated battery capacity physical prediction model and the updated battery capacity LSTM prediction model; Use the updated battery capacity physical prediction model and the updated battery capacity LSTM prediction model to process the medical simulator battery to obtain the medical simulator battery life prediction result.

[0006] In view of the characteristics of the medical simulator battery with a wide operating temperature range, high-frequency intermittent charging and discharging, and long continuous working time, the present invention proposes a medical simulator battery life prediction method based on dynamic weight and physical constraint. Through the fusion of the battery capacity physical prediction model and the LSTM battery capacity prediction model, the medical simulator battery life is predicted; the battery capacity physical prediction model can predict the medical simulator battery capacity within the extreme operating temperature, so that the battery physical prediction capacity conforms to the actual aging rate of the medical simulator at different temperatures; the LSTM battery capacity prediction model can capture the time-series characteristics of the battery capacity change and obtain the accurate battery LSTM prediction capacity; based on the high-frequency intermittent charging and discharging unique to the medical simulator, design a dynamic fusion weight related to the charging and discharging interval and average current of the medical simulator battery, and fuse the battery physical prediction capacity and the battery LSTM prediction capacity through the dynamic fusion weight, reduce the battery capacity prediction error of the medical simulator under high-frequency charging and discharging and long working time, and make the prediction result conform to the battery physical and chemical change process within the operating temperature of the medical simulator, improving the accuracy and reliability of the medical simulator battery life prediction.

[0007] Furthermore: The expression of the dynamic fusion weight is as follows:

[0008] ;

[0009] Where is the dynamic fusion weight, is the Sigmoid function, is the weight matrix of the dynamic fusion weight, is the bias term of the dynamic fusion weight, is the average current of the medical simulator battery, is the charge and discharge interval of the medical simulator battery, is a vector composed of the average current and charge and discharge interval of the medical simulator battery.

[0010] The present invention takes the average current and charge and discharge interval of the medical simulator battery as the input vector, and uses the Sigmoid function, weight matrix and bias term to calculate the dynamic fusion weight, so as to solve the problem that the traditional LSTM model in the prior art captures insufficient high-frequency charge and discharge timing characteristics of the medical simulator and does not consider the battery attenuation rate, resulting in inaccurate battery life prediction results; the dynamic fusion weight can be dynamically adjusted according to different situations of the medical simulator battery, dynamically balance the contributions of the physical prediction model of battery capacity and the LSTM prediction model of battery capacity, improve the adaptability of the model to the high-frequency intermittent charge and discharge scenario, and improve the accuracy of battery life prediction.

[0011] Furthermore: The expression of the physical prediction model of battery capacity is as follows:

[0012] ;

[0013] ;

[0014] ;

[0015] Among them, is the physical prediction capacity of the medical simulator battery for the th charge and discharge cycle, is the initial capacity of the medical simulator battery, is the natural constant, is the number of charge and discharge cycles of the medical simulator, is the temperature-dependent aging rate, is the frequency factor, is the activation energy of the medical simulator battery for the th charge and discharge cycle, is the temperature change compensation coefficient of the medical simulator battery for the th charge and discharge cycle, is the temperature fluctuation amplitude of a single charge cycle period, is the gas constant, is the battery temperature, is the maximum temperature of a single charge cycle period, is the minimum temperature of a single charge cycle period.

[0016] The present invention modifies the traditional Arrhenius equation by introducing a temperature fluctuation compensation coefficient, correlates the battery capacity attenuation with the battery temperature, activation energy, and temperature fluctuation amplitude, reflects the aging rate of the medical simulator battery at different temperatures, and accurately captures the impact of temperature changes on the battery capacity attenuation, solving the problem that the traditional method cannot accurately describe the battery capacity attenuation characteristics of the medical simulator battery under complex temperature environments and high-frequency charge and discharge conditions, improving the prediction accuracy of the medical simulator battery under extreme working temperature conditions, and making the battery capacity prediction result more conform to the battery aging law in the actual use scenario.

[0017] Further: The expression of the battery capacity LSTM prediction model is as follows:

[0018] ;

[0019] ;

[0020] ;

[0021] ;

[0022] Wherein, is the LSTM predicted capacity of the medical simulator battery for the th charge and discharge cycle, is the trainable weight, is the bias, is the LSTM hidden state of the th charge and discharge cycle, is the LSTM hidden state of the th charge and discharge cycle, is the LSTM network model function, is the capacity sequence, is the temperature, is the charge and discharge current, is the reset gate, is the weight of the reset gate, is the bias of the reset gate, is the Sigmoid activation function, is the candidate gate, is the weight of the candidate gate, is the bias of the candidate gate, is the hyperbolic tangent function.

[0023] The battery capacity LSTM prediction model of the present invention uses data such as capacity sequences, temperature, and charge and discharge currents, and utilizes the gating mechanism of the LSTM network to capture temporal dependencies, learning the complex dynamic characteristics of the medical simulator battery during high-frequency intermittent charge and discharge processes. It can solve the problem that pure data-driven methods cannot fully utilize the physical and chemical mechanism information of the battery, and at the same time overcome the defect of the prior art in insufficiently capturing the high-frequency charge and discharge temporal characteristics when dealing with the specific working conditions of medical simulators, improve the prediction accuracy of the battery capacity change trend, and adapt to the actual use conditions of the medical simulator battery in high-frequency intermittent charge and discharge scenarios.

[0024] Furthermore: The expression of the final battery capacity prediction result is as follows:

[0025] ;

[0026] Wherein, is the final battery predicted capacity of the medical simulator battery for the th charge and discharge cycle, is the dynamic fusion weight, is the physical predicted capacity of the medical simulator battery for the th charge and discharge cycle, is the LSTM predicted capacity of the medical simulator battery for the th charge and discharge cycle.

[0027] The expression of the final battery capacity prediction result of the present invention performs a weighted sum of the physical model predicted capacity and the LSTM predicted capacity through the dynamic fusion weight, realizing the combination of the battery capacity physical prediction model and the battery capacity LSTM prediction model. Utilizing the in-depth analysis of the battery aging mechanism by the battery capacity physical prediction model and the fitting ability of the battery capacity LSTM prediction model for the battery's complex temporal data, it solves the problem that traditional methods are difficult to balance the battery's physical laws and data characteristics under different medical simulator working conditions. By dynamically adjusting the weight, the medical simulator can better adapt to the actual use scenario under complex working conditions such as high-frequency intermittent charge and discharge conditions and extreme working temperatures, improving the accuracy and reliability of battery capacity prediction.

[0028] Furthermore: The expression of the physical constraint loss function is as follows:

[0029] ;

[0030] ;

[0031] ;

[0032] ;

[0033] Wherein, is the physical constraint loss function, is the first weight parameter, is the predicted battery capacity attenuation gradient, is the actual battery capacity attenuation gradient.

[0034] In view of the problem of insufficient prediction accuracy caused by the pure data-driven method in the prior art when ignoring the physical and chemical processes of the battery, the present invention designs a physical constraint loss function for the physical prediction model of the battery capacity. By introducing a physical constraint term, the difference between the predicted battery capacity attenuation gradient and the actual battery capacity attenuation gradient is incorporated into the calculation, so that the modified Arrhenius equation fits the actual battery aging rate under extreme temperatures, improves the prediction accuracy of the physical prediction model of the battery capacity, enhances the reliability of the technical solution of the present invention under complex working conditions, and solves the problem that the traditional method cannot effectively utilize the physical laws of the battery to constrain the model training to improve the prediction accuracy.

[0035] Further: The expression of the LSTM loss function is as follows:

[0036] ;

[0037] ;

[0038] Among them, is the LSTM loss function, is the gradient of each gating unit of the LSTM, are the LSTM network parameters, is the second weight parameter, is the final battery capacity prediction result, is the actual battery capacity.

[0039] In view of the problem that the traditional LSTM model in the prior art is difficult to adapt to the complex working conditions of high-frequency charge and discharge of medical simulators, the present invention quantifies the mean square error between the final battery capacity prediction result and the actual battery capacity through the LSTM loss function, and calculates the gradient of the error with respect to the LSTM network parameters; solves the deficiency of the traditional method in capturing time series features, can promote the adjustment of the LSTM network parameters, enables the LSTM prediction model of the battery capacity to better learn the time series features of the battery capacity change, and improves the prediction accuracy and adaptability of the LSTM prediction model of the battery capacity under high-frequency intermittent charge and discharge working conditions.

[0040] Further: The expression of the joint loss function is as follows:

[0041] ;

[0042] Among them, is the joint loss function, is the third weight parameter, is the predicted battery capacity of the LSTM, is the actual battery capacity, is the predicted battery capacity attenuation gradient, is the actual battery capacity attenuation gradient.

[0043] The present invention combines the physical prediction of battery capacity and the LSTM prediction of battery capacity through a joint loss function, taking into account both the deviation between the predicted battery capacity and the actual battery capacity and the difference between the predicted capacity attenuation gradient and the actual capacity attenuation gradient, solving the problem of inaccurate prediction by a single model in the prior art and the problem that it is difficult for traditional methods to balance the battery data characteristics and the battery physical change law; through the joint loss function, it is possible to optimize the parameters of the battery capacity physical prediction model and the parameters of the battery capacity LSTM prediction model simultaneously, while adapting to the complex working conditions of high-frequency intermittent charge and discharge, ensuring that the prediction results conform to the actual aging law of the battery, and improving the accuracy and reliability of battery life prediction.

[0044] Further: updating the parameters of the battery capacity physical prediction model according to the joint loss function specifically includes:

[0045] Updating the temperature change compensation coefficient of the battery capacity physical prediction model, and the expression is as follows:

[0046] ;

[0047] Wherein, is the temperature change compensation coefficient of the th charge and discharge cycle, is the temperature change compensation coefficient of the th charge and discharge cycle, is the learning rate of the temperature change compensation coefficient, is the joint loss function;

[0048] Updating the activation energy of the battery capacity physical prediction model, and the expression is as follows:

[0049] ;

[0050] Wherein, is the activation energy updated based on the physical constraint loss function of the th charge and discharge cycle, is the activation energy updated based on the physical constraint loss function of the th charge and discharge cycle, is the learning rate of the activation energy.

[0051] Through the gradient descent method, the present invention dynamically adjusts the temperature change compensation coefficient and activation energy of the physical prediction model of battery capacity by using the joint loss function, solves the problems in the prior art that the physical model parameters cannot adapt to the actual working conditions of the medical simulator battery, and it is difficult to correct the physical model parameters at different temperatures and charge-discharge modes; through the update of the temperature change compensation coefficient and activation energy, the physical prediction model of battery capacity can better reflect the aging characteristics of the actual battery, and has higher prediction accuracy and adaptability under extreme working temperatures and high-frequency charge-discharge conditions.

[0052] Further: The parameters of the LSTM prediction model of battery capacity are updated according to the joint loss function, and the expression is as follows:

[0053] ;

[0054] ;

[0055] ;

[0056] Among them, is the updated LSTM network parameter, is the LSTM network parameter before update, is the learning rate of the LSTM prediction model of battery capacity, is the first moment of the gradient of the th charge-discharge cycle, is the first moment of the th charge-discharge cycle, is the second moment of the th charge-discharge cycle, is the second moment of the th charge-discharge cycle, is a constant, is the first moment decay rate, is the second moment decay rate, is the gradient of the LSTM network parameter of the th charge-discharge cycle, is the element-wise square of the gradient of the LSTM network parameter of the

[0057] According to the combined loss function, the present invention updates the parameters of the battery capacity LSTM prediction model by using the adaptive learning rate method of the Adam optimizer, and adjusts the LSTM network parameters through the estimation of the first moment and the second moment of the gradient, so that the LSTM model can simultaneously consider data fitting and physical law constraints, improve the prediction accuracy of the battery capacity LSTM prediction model for the change trend of the battery capacity, and solve the problem that the traditional LSTM model in the prior art is difficult to capture the time series characteristics and adapt to high-noise data when dealing with the complex working conditions of high-frequency intermittent charging and discharging of medical simulator batteries, resulting in insufficient prediction accuracy.

[0058] One or more technical solutions provided by the present invention have at least the following technical effects or advantages:

[0059] Based on the charging and discharging intervals and average currents of medical simulator batteries, the present invention designs dynamic fusion weights, fuses the battery capacity physical prediction model with the battery capacity LSTM prediction model through the dynamic fusion weights, constructs a hybrid battery life prediction model architecture, introduces a dynamic weight adjustment mechanism and a combined loss function, enables the hybrid battery life prediction model architecture to adapt to the high-frequency intermittent charging and discharging characteristics of medical simulator batteries, accurately capture time series characteristics, and at the same time combines physical constraints to improve the prediction accuracy and reliability, so that the battery capacity prediction results conform to the battery aging law of medical simulator batteries in actual use scenarios, and improve the accuracy and reliability of the prediction of the battery life of medical simulator batteries. Description of the Drawings

[0060] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, form a part of the present invention, and do not constitute a limitation on the embodiments of the present invention;

[0061] Figure 1 is a flowchart of the method for predicting the battery life of a medical simulator based on dynamic weights and physical constraints in the present invention. Detailed Embodiments

[0062] In order to more clearly understand the above objects, features and advantages of the present invention, the present invention will be further described in detail below with reference to the drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other.

[0063] Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention may be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0064] Embodiment 1

[0065] As Figure 1As shown, the present invention provides a method for predicting the battery life of a medical simulator based on dynamic weights and physical constraints, including: According to the physical characteristics of the battery of the medical simulator, establish a physical prediction model of battery capacity with dynamic fusion weights, and an LSTM prediction model of battery capacity; According to the dynamic fusion weights, fuse the physical prediction capacity of the physical prediction model of battery capacity and the LSTM prediction capacity of the LSTM prediction model of battery capacity to obtain the final battery capacity prediction result; According to the final battery capacity prediction result, establish a physical constraint loss function for the physical prediction model of battery capacity and an LSTM loss function for the LSTM prediction model of battery capacity, and establish a joint loss function according to the physical constraint loss function and the LSTM loss function; Update the parameters of the physical prediction model of battery capacity according to the joint loss function, and update the parameters of the LSTM prediction model of battery capacity according to the joint loss function to obtain an updated physical prediction model of battery capacity and an updated LSTM prediction model of battery capacity; Use the updated physical prediction model of battery capacity and the updated LSTM prediction model of battery capacity to process the battery of the medical simulator to obtain the prediction result of the battery life of the medical simulator.

[0066] In daily use, the battery characteristics of the medical simulator include: a wide operating temperature range, capable of operating in an environment of -10°C to 40°C; the charge and discharge mode is high-frequency intermittent, requiring 3 to 5 charges and discharges within the test cycle of the medical simulator, which is significantly different from the daily use of real medical equipment; long-term continuous testing, requiring the medical simulator to run continuously for 72 hours; the life of the battery of the medical simulator will affect the test accuracy of the medical simulator.

[0067] Aiming at the battery characteristics of medical simulators, this paper proposes a medical simulator battery life prediction method based on dynamic weights and physical constraints. By integrating the battery capacity physical prediction model and the LSTM battery capacity prediction model, a hybrid model architecture is constructed to predict the battery life of the medical simulator. The battery capacity physical prediction model is based on the Arrhenius equation and introduces a temperature fluctuation compensation coefficient for correction. The battery capacity physical prediction model is established, which can accurately reflect the aging and attenuation law of the battery in the temperature range of -10℃~40℃, and adapt to the extreme working temperature requirements of the medical simulator; the battery capacity LSTM prediction model can capture the timing characteristics of high-frequency intermittent charging and discharging, solve the problem of insufficient timing feature capture of the traditional LSTM model, and adapt to the special charging and discharging mode of the medical simulator; introduce a dynamic fusion weight adjustment mechanism, and adaptively adjust the dynamic fusion weight according to the charging and discharging interval and average current, so that the model dynamically balances the contributions of the battery capacity physical prediction model and the battery capacity LSTM prediction model under different working conditions, enhances the prediction performance under high-frequency charging and discharging conditions, and improves the model adaptability; through the joint loss function, the parameters of the battery capacity physical prediction model and the battery capacity LSTM prediction model are synchronously updated to ensure that the prediction results fit the actual working conditions of the battery and improve the accuracy and reliability of battery life prediction.

[0068] In the present invention, the charge and discharge intervals of the medical simulator test cycle are and average current Get the dynamic fusion weight, the expression of dynamic fusion weight is as follows:

[0069] ;

[0070] in, is the dynamic fusion weight, is the Sigmoid function, is the weight matrix of dynamic fusion weights, is the bias term of dynamic fusion weight, is the average current of the medical simulator battery, is the charge and discharge interval of the medical simulator battery, A vector composed of the average current of the medical simulator battery and the charge-discharge interval. When the medical simulator is in the high-frequency mode, the charge-discharge interval of the medical simulator battery is generally less than 10 minutes, and the dynamic fusion weight at this time is about 0.1, which can strengthen the ability of the battery capacity LSTM prediction model to capture instantaneous current pulses. When the medical simulator is in the normal mode, the charge-discharge interval of the medical simulator battery is generally greater than 30 minutes, and the dynamic fusion weight at this time is about 0.5, which can balance the contributions of the battery capacity physical prediction model and the battery capacity LSTM prediction model. The dynamic fusion weight of the present invention can be dynamically adjusted according to different situations of the medical simulator battery, improve the adaptability of the battery capacity physical prediction model and the battery capacity LSTM prediction model to the high-frequency intermittent charge-discharge scenario, and improve the accuracy of battery life prediction, solving the problem that the traditional LSTM model in the prior art has insufficient capture of the high-frequency charge-discharge timing characteristics of the medical simulator and does not consider the battery decay rate.

[0071] In the present invention, a temperature fluctuation compensation coefficient is introduced to correct the Arrhenius equation to obtain a battery capacity physical prediction model, which can relate the battery capacity decay to the battery temperature, activation energy, and temperature fluctuation amplitude, indirectly reflect the aging rate of the medical simulator battery at different temperatures, and capture the impact of temperature changes on the battery capacity decay of the medical simulator, solving the problem that the traditional method cannot accurately describe the battery capacity decay characteristics of the medical simulator battery in a complex temperature environment and high-frequency charge-discharge working conditions, making the battery capacity prediction result more in line with the battery aging law in the actual use scenario. The expression of the battery capacity physical prediction model of the present invention is as follows:

[0072] ;

[0073] ;

[0074] ;

[0075] Among them, is the physical prediction capacity of the th charge-discharge cycle of the medical simulator battery, with the unit of mAh, is the initial capacity of the medical simulator battery, with the unit of mAh, is the natural constant, is the number of charge-discharge cycles of the medical simulator, that is, the number of test cycles of the medical simulator, is the temperature-dependent aging rate, with the unit of / s, is the frequency factor, is the th activation energy of the medical simulator battery in the th charge-discharge cycle, representing the difficulty of material aging of the medical simulator battery, The temperature change compensation coefficient for the first charge-discharge cycle. According to experience, the initial value of the temperature change compensation coefficient can be set to 0.67. is the temperature fluctuation amplitude for a single charge cycle, in K. is the gas constant. is the battery temperature, in K. is the maximum temperature for a single charge cycle, in K. is the minimum temperature for a single charge cycle, in K.

[0076] In the present invention, the capacity sequence , temperature and charge-discharge current are used as the inputs of the LSTM network, and the predicted capacity of the battery LSTM is used as the output of the LSTM network. The gating mechanism of the LSTM network is utilized to capture the temporal dependence of the medical simulator battery, learn the dynamic characteristics of the medical simulator battery during high-frequency intermittent charge and discharge processes, solve the problem of insufficient capture of high-frequency charge-discharge temporal characteristics in the prior art when dealing with specific working conditions of medical simulators, improve the prediction accuracy of the battery capacity change trend, and adapt to the actual usage conditions of the medical simulator battery in high-frequency intermittent charge and discharge scenarios. The expression of the battery capacity LSTM prediction model is as follows:

[0077] ;

[0078] ;

[0079] ;

[0080] ;

[0081] Wherein, is the LSTM predicted capacity of the medical simulator battery for the th charge-discharge cycle, are the trainable weights, is the bias, is the LSTM hidden state for the th charge-discharge cycle, is the LSTM hidden state for the th charge-discharge cycle, is the LSTM network model function, is the capacity sequence, is the temperature, is the charge-discharge current, is the reset gate, is the weight of the reset gate, is the bias of the reset gate, is the Sigmoid activation function. is the candidate gate, is the weight of the candidate gate, is the bias of the candidate gate, is the hyperbolic tangent function.

[0082] According to the dynamic fusion weight, the physical prediction capacity of the battery capacity physical prediction model and the LSTM prediction capacity of the battery capacity LSTM prediction model are fused to obtain the final battery capacity prediction result. The expression of the final battery capacity prediction result is as follows:

[0083] ;

[0084] Among them, is the final battery prediction capacity of the medical simulator battery for the th charge-discharge cycle, is the dynamic fusion weight, is the physical prediction capacity of the medical simulator battery for the th charge-discharge cycle, is the LSTM prediction capacity of the medical simulator battery for the th charge-discharge cycle; Through the dynamic fusion weight, the fusion of the battery capacity physical prediction model and the battery capacity LSTM prediction model is realized. By using the in-depth analysis of the battery aging mechanism by the battery capacity physical prediction model and the fitting ability of the battery capacity LSTM prediction model for the complex time-series data of the battery, the medical simulator can better adapt to the actual use scenarios under complex working conditions such as high-frequency intermittent charge-discharge conditions and extreme working temperatures, improving the accuracy and reliability of battery capacity prediction; Among them, the dynamic fusion weight represents the contribution of the battery capacity physical prediction model. When the dynamic fusion weight is, it means that the final battery prediction capacity completely depends on the battery capacity physical prediction model.

[0085] In the present invention, according to backpropagation, a physical constraint loss function of the battery capacity physical prediction model is established, and the objects of action are the temperature change compensation coefficient and the activation energy. The expression of the physical constraint loss function is as follows:

[0086] ;

[0087] ;

[0088] ;

[0089] ;

[0090] Among them, is the physical constraint loss function, is the first weight parameter and can be set according to experience. is the predicted battery capacity attenuation gradient. is the actual battery capacity attenuation gradient.

[0091] In the present invention, the LSTM loss function of the battery capacity LSTM prediction model is obtained through backpropagation, which is used to update the LSTM network parameters, enabling the battery capacity LSTM prediction model to better learn the temporal characteristics of battery capacity changes, improving the prediction accuracy and adaptability of the battery capacity LSTM prediction model under high-frequency intermittent charge and discharge conditions. The expression of the LSTM loss function is as follows:

[0092] ;

[0093] ;

[0094] Among them, is the LSTM loss function. is the gradient of each gating unit of the LSTM, which is calculated through backpropagation through time and involves the temporal gradient accumulation of the LSTM cell state and the hidden state . is the LSTM network parameter. is the second weight parameter and can be set according to experience. is the final battery capacity prediction result. is the actual battery capacity.

[0095] In the present invention, the combined loss function of the battery capacity physical prediction model and the battery capacity LSTM prediction model is established according to the physical constraint loss function and the LSTM loss function. The battery capacity physical prediction and the battery capacity LSTM prediction are combined, and the deviation between the predicted battery capacity and the actual battery capacity and the difference between the predicted capacity attenuation gradient and the actual capacity attenuation gradient are considered to solve the problem of inaccurate prediction of a single model in the prior art and the problem that it is difficult for traditional methods to balance battery data characteristics and battery physical change laws. When in complex high-frequency intermittent charge and discharge conditions, it ensures that the prediction results conform to the actual aging law of the battery, improving the accuracy and reliability of battery life prediction. The expression of the combined loss function of the present invention is as follows:

[0096] ;

[0097] Among them, is the combined loss function. is the third weight parameter and can be set according to experience. is the predicted battery capacity of the LSTM. is the actual battery capacity. is the predicted battery capacity attenuation gradient, is the actual battery capacity attenuation gradient.

[0098] After obtaining the joint loss function, according to the joint loss function, use the Adam optimizer to update the temperature change compensation coefficient and activation energy, so that the physical prediction model of battery capacity can better reflect the aging characteristics of the actual battery, and has higher prediction accuracy and adaptability under extreme working temperatures and high-frequency charge and discharge conditions, making the physically predicted battery capacity fit the actual battery aging rate under extreme temperatures. Specifically, it includes:

[0099] Update the temperature change compensation coefficient of the physical prediction model of battery capacity. The expression is as follows:

[0100] ;

[0101] where, is the temperature change compensation coefficient for the th charge and discharge cycle, is the temperature change compensation coefficient for the th charge and discharge cycle, is the learning rate of the temperature change compensation coefficient, is the joint loss function;

[0102] Update the activation energy of the physical prediction model of battery capacity. The expression is as follows:

[0103] ;

[0104] where, is the activation energy updated based on the physical constraint loss function for the th charge and discharge cycle, is the activation energy updated based on the physical constraint loss function for the th charge and discharge cycle, is the learning rate of the activation energy.

[0105] After obtaining the joint loss function, according to the joint loss function, use the adaptive learning rate method of the Adam optimizer to update the parameters of the LSTM prediction model of battery capacity. Adjust the LSTM network parameters through the first-order and second-order moment estimates of the joint loss function gradient, so that the LSTM model can consider both data fitting and physical law constraints at the same time, improve the prediction accuracy of the LSTM prediction model of battery capacity for the battery capacity change trend, and solve the problem that the traditional LSTM model in the prior art is difficult to capture the timing characteristics and adapt to high-noise data when dealing with the complex working conditions of high-frequency intermittent charge and discharge of medical simulator batteries, resulting in insufficient prediction accuracy. Update the parameters of the LSTM prediction model of battery capacity according to the joint loss function. The expression is as follows:

[0106] ;

[0107] ;

[0108] ;

[0109] Among them, is the updated LSTM network parameter, is the LSTM network parameter before update, is the learning rate of the LSTM prediction model for battery capacity, is the first-order moment of the gradient for the th charge and discharge cycle, is the first-order moment for the th charge and discharge cycle, is the second-order moment for the th charge and discharge cycle, is the second-order moment for the th charge and discharge cycle, is a constant, generally set to a very small constant according to experience to ensure numerical stability, is the first-order moment decay rate, is the th charge and discharge cycle of the gradient of the LSTM network parameter, is the th charge and discharge cycle of the element-wise square of the gradient of the LSTM network parameter.

[0110] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0111] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations therein.

Claims

1. A method for predicting the battery life of a medical simulator based on dynamic weights and physical constraints, characterized in that, Including: Based on the physical characteristics of the medical simulator battery, establish a dynamic fusion weight, a physical prediction model of battery capacity, and an LSTM prediction model of battery capacity; According to the dynamic fusion weight, fuse the physical prediction capacity of the physical prediction model of battery capacity and the LSTM prediction capacity of the LSTM prediction model of battery capacity to obtain the final battery capacity prediction result; According to the final battery capacity prediction result, establish a physical constraint loss function for the physical prediction model of battery capacity and an LSTM loss function for the LSTM prediction model of battery capacity, and establish a joint loss function based on the physical constraint loss function and the LSTM loss function; Update the parameters of the physical prediction model of battery capacity according to the joint loss function, and update the parameters of the LSTM prediction model of battery capacity according to the joint loss function to obtain an updated physical prediction model of battery capacity and an updated LSTM prediction model of battery capacity; Use the updated physical prediction model of battery capacity and the updated LSTM prediction model of battery capacity to process the medical simulator battery to obtain the medical simulator battery life prediction result.

2. The method for predicting the battery life of a medical simulator based on dynamic weights and physical constraints according to claim 1, wherein The expression of the dynamic fusion weight is as follows: ; Among them, is the dynamic fusion weight, is the Sigmoid function, is the weight matrix of the dynamic fusion weight, is the bias term of the dynamic fusion weight, is the average current of the medical simulator battery, is the charge-discharge interval of the medical simulator battery, is the vector composed of the average current and the charge-discharge interval of the medical simulator battery.

3. The method for predicting the battery life of a medical simulator based on dynamic weights and physical constraints according to claim 1, wherein The expression of the physical prediction model of battery capacity is as follows: ; ; ; Wherein, is the physically predicted capacity of the medical simulator battery for the th charge-discharge cycle, is the initial capacity of the medical simulator battery, is the natural constant, is the number of charge-discharge cycles of the medical simulator, is the temperature-dependent aging rate, is the frequency factor, is the th activation energy of the medical simulator battery for the th charge-discharge cycle, is the temperature change compensation coefficient for the th charge-discharge cycle, is the gas constant, is the battery temperature, is the maximum temperature for a single charge cycle period, is the minimum temperature for a single charge cycle period.

4. The method for predicting the battery life of a medical simulator based on dynamic weights and physical constraints according to claim 1, wherein The expression of the LSTM prediction model of battery capacity is as follows: ; ; ; ; Wherein, is the LSTM predicted capacity of the th charge-discharge cycle of the medical simulator battery, is the trainable weight, is the bias, is the th LSTM hidden state of the charge-discharge cycle, is the th LSTM hidden state of the charge-discharge cycle, is the LSTM network model function, is the capacity sequence, is the temperature, is the charge-discharge current, is the reset gate, is the weight of the reset gate, is the bias of the reset gate, is the Sigmoid activation function, is the candidate gate, is the weight of the candidate gate, is the bias of the candidate gate, is the hyperbolic tangent function.

5. The method for predicting the battery life of a medical simulator based on dynamic weights and physical constraints according to claim 1, wherein The expression of the final battery capacity prediction result is as follows: ; Among them, is the final battery predicted capacity of the nth charge-discharge cycle of the medical simulator battery, is the dynamic fusion weight, is the physical predicted capacity of the nth charge-discharge cycle of the medical simulator battery, is the LSTM predicted capacity of the nth charge-discharge cycle of the medical simulator battery.

6. The method for predicting the battery life of a medical simulator based on dynamic weights and physical constraints according to claim 3, wherein The expression of the physical constraint loss function is as follows: ; ; ; ; Among them, is the physical constraint loss function, is the first weight parameter, is the predicted battery capacity attenuation gradient, is the actual battery capacity attenuation gradient.

7. The method for predicting the battery life of a medical simulator based on dynamic weights and physical constraints according to claim 4, wherein The expression of the LSTM loss function is as follows: ; ; Among them, is the LSTM loss function, is the gradient of each gating unit of the LSTM, is the LSTM network parameter, is the second weight parameter, is the final predicted result of the battery capacity, is the actual battery capacity.

8. The method for predicting the battery life of a medical simulator based on dynamic weights and physical constraints according to claim 1, wherein The expression of the joint loss function is as follows: ; Among them, is the combined loss function, is the third weight parameter, is the predicted battery capacity of the LSTM, is the actual battery capacity, is the predicted battery capacity attenuation gradient, is the actual battery capacity attenuation gradient.

9. The method for predicting the battery life of a medical simulator based on dynamic weights and physical constraints according to claim 1, wherein The updating of the parameters of the physical prediction model of battery capacity according to the joint loss function specifically includes: Update the temperature change compensation coefficient of the physical prediction model of battery capacity, and the expression is as follows: ; Among them, is the temperature change compensation coefficient for the th charge-discharge cycle, is the temperature change compensation coefficient for the th charge-discharge cycle, is the learning rate of the temperature change compensation coefficient, is the combined loss function; Update the activation energy of the physical prediction model of battery capacity, and the expression is as follows: ; Among them, is the activation energy updated by the physically constrained loss function for the th charge-discharge cycle, is the activation energy updated by the physically constrained loss function for the th charge-discharge cycle, is the learning rate of the activation energy.

10. The method for predicting the battery life of a medical simulator based on dynamic weights and physical constraints according to claim 1, wherein The updating of the parameters of the LSTM prediction model of battery capacity according to the joint loss function is expressed as follows: ; ; ; Among them, is the updated LSTM network parameter, is the LSTM network parameter before update, is the learning rate of the battery capacity LSTM prediction model, is the first-order moment of the gradient of the th charge-discharge cycle, is the first-order moment of the th charge-discharge cycle, is the second-order moment of the th charge-discharge cycle, is the second-order moment of the th charge-discharge cycle, is a constant, is the first-order moment decay rate, is the second-order moment decay rate, is the gradient of the LSTM network parameter of the th charge-discharge cycle, is the element-wise square of the gradient of the LSTM network parameter of the

Images