Multi-performance collaborative optimization system for liquid battery electrolyte

Abstract

The invention relates to the technical field of liquid battery electrolyte, in particular to a liquid battery electrolyte multi-performance collaborative optimization system which comprises a parameter acquisition module, a constraint condition presetting module, a data driving model building module, a multi-objective optimization engine and a result output and verification module. The system constructs a data driving model based on an attention mechanism neural network by acquiring historical data of an electrolyte formula and process parameters, and maps a relationship between the parameters and comprehensive performance scores; performing multi-objective optimization by adopting a hybrid strategy combining simulated annealing and Bayesian optimization, and outputting a Pareto optimal solution set; and a data closed-loop updating model is verified through experiments, so that continuous optimization is realized. According to the method, the problems of multi-parameter coupling, multi-target conflict and low optimization efficiency of the electrolyte are solved, and the research and development efficiency and the comprehensive performance of the electrolyte are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of liquid battery electrolyte technology, and in particular to a multi-performance synergistic optimization system for liquid battery electrolytes. Background Technology

[0002] As a core component of electrochemical energy storage systems, the performance of the electrolyte in liquid batteries directly determines key indicators such as energy density, cycle life, safety, and operating temperature range. With the increasing demands on battery performance from applications such as electric vehicles and large-scale energy storage, developing high-performance electrolytes with synergistic optimization of multiple performance characteristics has become a key research direction in the industry.

[0003] In existing technologies, the optimization and development of electrolytes has long faced multiple challenges. Common approaches include empirical trial-and-error or single-objective optimization strategies, such as introducing novel lithium salts or specific solvents to improve ionic conductivity, or adding functional film-forming additives or adjusting solvent ratios to improve interfacial stability. While these methods have made some progress in improving specific properties, they have failed to systematically solve the fundamental problem of synergistic optimization of multiple electrolyte properties.

[0004] However, existing optimization techniques still have significant limitations:

[0005] 1. The final performance of an electrolyte is the result of a complex coupling effect between formulation parameters (such as the type and ratio of lithium salts, solvents, and additives) and process parameters (such as mixing sequence, temperature, stirring intensity, and time). Most existing methods separate formulation design from process optimization, or rely solely on empirical correlations, lacking a unified modeling system that can simultaneously accommodate and handle these two types of multidimensional parameters and reveal the complex nonlinear mapping relationship between them and overall performance. This leads to blind spots in the optimization process, making it difficult to fully exploit the performance potential of the electrolyte through synergistic parameter adjustment.

[0006] 2. Faced with multiple, even conflicting, performance targets such as ionic conductivity, high-temperature stability, and wide-temperature-range performance, traditional trial-and-error methods or simple grid search methods are extremely costly and time-consuming in searching the vast parameter space, easily getting trapped in local optima, and unable to efficiently obtain the Pareto optimal solution set that characterizes the best equilibrium relationship. Although some studies have attempted to introduce computational simulation to aid design, they often lack an intelligent optimization engine specifically designed for the high-dimensional, continuous parameter space of electrolytes, and cannot achieve fast, targeted global optimization.

[0007] 3. Existing optimization models are typically in an "open-loop" state, meaning that experimental verification results are only used to determine the merits of a specific formulation, without effectively feeding back into the optimization model to correct and improve its predictive accuracy. This "one-off" optimization model prevents the model from continuously evolving with data accumulation and adapting to new R&D needs, thus limiting the long-term optimization efficiency and accuracy of the system.

[0008] Therefore, this invention proposes a multi-performance synergistic optimization system for liquid battery electrolytes. Summary of the Invention

[0009] The purpose of this invention is to provide a multi-performance synergistic optimization system for liquid battery electrolytes, which achieves multi-performance synergistic optimization of electrolytes through data-driven models and multi-objective optimization.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a multi-performance synergistic optimization system for liquid battery electrolytes, comprising: a processor, a memory, and a computer program stored in the memory; the computer program is configured to, when executed by the processor, implement the following program modules:

[0011] The parameter acquisition module is used to acquire historical datasets of liquid battery electrolytes, including a set of formulation parameters and a set of process parameters.

[0012] The constraint preset module is used to receive input of multiple performance constraints corresponding to the target battery application scenario;

[0013] The data-driven model building module is used to construct a data-driven model that maps the relationship between electrolyte parameters and comprehensive performance scores based on the formula parameter set, process parameter set, and performance test results obtained under the performance constraints.

[0014] A multi-objective optimization engine is used to call the data-driven model to perform iterative simulation calculations with the goal of maximizing the comprehensive performance score, and output the Pareto optimal solution set.

[0015] The results output and verification module is used to output the Pareto optimal solution set, receive actual performance data obtained by the user through experimental verification based on a specific solution, and update the data-driven model using the actual performance data.

[0016] Preferably, the performance constraints target performance indicators including two or more of the following: ionic conductivity, high temperature stability, wide temperature range performance, wettability with electrode materials, and dendrite suppression ability.

[0017] Preferably, the formulation parameter set includes the concentration of lithium salt LiPF6, the volume ratio of solvent ethylene carbonate (EC) to ethyl methyl carbonate (EMC), and the concentration of film-forming additive LiPO2F2; the process parameter set includes mixing temperature and stirring time; wherein the solvent comprises ethylene carbonate (EC) and ethyl methyl carbonate (EMC), and the volume ratio of EC to EMC is 3:7 to 4:6.

[0018] Preferably, the data-driven model is a machine learning model; the machine learning model is a neural network model containing an attention mechanism, used to capture the nonlinear relationship between formula parameters, process parameters and battery long-cycle performance.

[0019] The neural network model assigns differentiated weights to different input parameters through an attention weight calculation function to capture the non-linear coupling relationship between parameters and performance, and outputs a comprehensive performance score.

[0020]

[0021] Where α i It is the attention weight, W a It is the attention weight matrix, h i It represents the hidden state of the input features. LeakyReLU is a non-linear activation function, exp is an exponential function. It is the normalized denominator.

[0022] Preferably, the additive includes the film-forming additive LiPO2F2, and its concentration is 0.5%-2% of the total mass of the electrolyte.

[0023] Preferably, the multi-objective optimization engine employs a hybrid optimization strategy. First, it performs a global exploration based on the simulated annealing algorithm, where the probability P of accepting a suboptimal solution is determined by the Metropolis criterion.

[0024]

[0025] Where Δf is the fitness difference, T k Given the current temperature; then, based on the Bayesian optimization algorithm, a local fine search is performed using the previous exploration results as prior knowledge, and finally, the Pareto optimal solution set is output, where each solution represents a candidate combination of electrolyte parameters.

[0026] Preferably, the result output and verification module receives actual verification data and retrains the data-driven model to form a closed-loop optimization; when the average absolute percentage error between the actual verification data and the corresponding model prediction data is greater than 10%, the model retraining process is automatically triggered, and the actual verification data is used to incrementally learn the model.

[0027] Preferably, a method for synergistic optimization of multiple performance characteristics of a liquid battery electrolyte is characterized by comprising the following steps:

[0028] S1: Obtain the historical formula parameter set and process parameter set of the electrolyte and the corresponding historical performance test results;

[0029] S2: Receive multiple performance constraints set by the user;

[0030] S3: Based on the historical data, train a data-driven model;

[0031] S4: With the goal of maximizing the overall performance score, the data-driven model is invoked to perform iterative simulation calculations and output the Pareto optimal solution set;

[0032] S5: Output the Pareto optimal solution set and receive actual verification data from the user for a specific solution to update the data-driven model.

[0033] Preferably, the step of updating the data-driven model includes: comparing the actual validation data with the corresponding simulated prediction data and calculating the mean absolute percentage error; when the error exceeds a preset threshold of 10%, triggering the model retraining process and updating the model parameters using an incremental learning algorithm.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] 1. This invention integrates the collection of electrolyte formulation parameter sets and process parameter sets by setting a parameter acquisition module and a data-driven model construction module. By constructing a data-driven model (a neural network model with an attention mechanism) that maps the relationship between electrolyte parameters and comprehensive performance scores, this invention can systematically reveal the nonlinear relationship under the complex coupling effect of formulation and process, optimize blind spots in traditional processes, and explore electrolyte performance through parameter synergistic adjustment.

[0036] 2. This invention introduces a multi-objective optimization engine and adopts a hybrid optimization strategy. First, it performs global exploration based on the simulated annealing algorithm, and then uses intelligent optimization engines such as the Bayesian optimization algorithm (whose acquisition function is the expected improved EI function) to adaptively adjust the high-dimensional continuous parameter space of the electrolyte. It efficiently performs iterative simulation calculations in the huge parameter space and automatically outputs the Pareto optimal solution set representing the best balance relationship. This overcomes the shortcomings of traditional "trial and error method" and grid search method, which are high cost, long cycle and easy to get trapped in local optima, and achieves fast and targeted global optimization.

[0037] 3. This invention achieves closed-loop optimization through a result output and verification module. The system innovatively feeds back the user's actual verification data for a specific solution to the model update process. When the deviation between the verification data and the model's predicted data exceeds a preset threshold, the system automatically triggers a retraining process and updates the data-driven model using incremental learning and other methods. This closed-loop mechanism enables the model to continuously evolve with data accumulation, constantly correcting and improving prediction accuracy. This effectively addresses the limitations of existing "one-off" open-loop optimization modes, thereby ensuring the long-term optimization efficiency and accuracy of the system. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the multi-performance synergistic optimization system for liquid battery electrolyte of the present invention;

[0039] Figure 2 This is a flowchart of the multi-performance synergistic optimization method for liquid battery electrolyte of the present invention;

[0040] Figure 3 This is an example diagram illustrating the construction of a data-driven model for implementing this invention;

[0041] Figure 4 This is a flowchart illustrating the workflow of the multi-objective optimization engine implemented in this invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Please see Figure 1 , Figure 2 , Figure 3 and Figure 4 As shown, this embodiment discloses a system-level method for synergistic optimization of multiple performance characteristics of a liquid battery electrolyte, including:

[0044] A multi-performance synergistic optimization system for liquid battery electrolytes, characterized in that it includes: a processor, a memory, and a computer program stored in the memory; the computer program is configured to, when executed by the processor, implement the following program modules, as detailed below. Figure 1 :

[0045] The parameter acquisition module is used to acquire historical datasets of liquid battery electrolytes. The historical datasets include formula parameter sets and process parameter sets.

[0046] The constraint pre-setting module is used to receive inputs of multiple performance constraints corresponding to the target battery application scenario, transforming complex physicochemical problems into structured data problems and clarifying the optimization dimensions and boundaries;

[0047] The data-driven model building module is used to construct a data-driven model that maps the relationship between electrolyte parameters and comprehensive performance scores based on the formula parameter set, process parameter set, and performance test results obtained under the performance constraints.

[0048] A multi-objective optimization engine is used to maximize the overall performance score by calling a data-driven model to perform iterative simulation calculations and output the Pareto optimal solution set.

[0049] The results output and verification module is used to output the Pareto optimal solution set, receive actual performance data obtained by the user through experimental verification based on a specific solution, and update the data-driven model using the actual performance data.

[0050] In this embodiment, the core principle of the multi-performance collaborative optimization system for liquid battery electrolytes lies in constructing a computational paradigm that integrates data perception, intelligent modeling, multi-objective decision-making, and continuous evolution, thereby systematically solving the inherent problems of multi-parameter coupling, multi-objective conflict, and low optimization efficiency in electrolyte research and development.

[0051] Furthermore, the performance constraints target two or more of the following performance indicators: ionic conductivity, high temperature stability, wide temperature range performance, wettability with electrode materials, and dendrite suppression ability.

[0052] Furthermore, the formulation parameter set includes the concentration of lithium salt LiPF6, the volume ratio of solvent ethylene carbonate (EC) to ethyl methyl carbonate (EMC), and the concentration of film-forming additive LiPO2F2; the process parameter set includes mixing temperature and stirring time.

[0053] Furthermore, the data-driven model is a machine learning model; the machine learning model is a neural network model incorporating an attention mechanism, used to capture the nonlinear relationship between formulation parameters, process parameters, and the long-cycle performance of the battery. See details... Figure 3 ;

[0054] The neural network model is defined by its attention weight calculation function:

[0055] Where α i It is the attention weight, W a It is the attention weight matrix, h i It represents the hidden state of the input features. LeakyReLU is a non-linear activation function, exp is an exponential function. It is the normalized denominator;

[0056] Differentiated weights are assigned to different input parameters to capture the nonlinear coupling relationship between parameters and performance, and output a comprehensive performance score. Through a dynamic weight allocation function, key parameters that have a significant impact on the final performance (such as specific additive concentration and process temperature) are automatically identified and strengthened, thereby accurately depicting the complex nonlinear mapping relationship between formulation and process at the mathematical level, overcoming the limitations of traditional experience-based reliance or simple linear regression.

[0057] Furthermore, the additives include film-forming additive LiPO2F2, and its concentration is 0.5%-2% of the total mass of the electrolyte.

[0058] Furthermore, the multi-objective optimization engine employs a hybrid optimization strategy. First, it performs a global exploration based on simulated annealing, where the probability P of accepting a suboptimal solution is determined by the Metropolis criterion:

[0059]

[0060] Where Δf is the fitness difference, T k Given the current temperature; then, based on the Bayesian optimization algorithm using the prior exploration results as a priori knowledge, a local fine-grained search is performed, finally outputting the Pareto optimal solution set, where each solution represents a candidate combination of electrolyte parameters. See details... Figure 4 ;

[0061] In this embodiment, the multi-objective optimization engine employs a hybrid optimization strategy primarily to address the contradiction between global exploration and local optimization in stages and with differentiation. Simulated annealing accepts inferior solutions with a certain probability in the early stages of high-temperature optimization, endowing the system with a strong ability to escape local optima and enabling extensive exploration of the parameter space. Bayesian optimization, on the other hand, builds upon this by utilizing existing point information to construct a surrogate model, guided by acquisition functions such as expected improvement, to perform efficient local mining. This sequential and combined optimization mechanism ensures that the system can quickly and directionally converge to the Pareto front, representing the optimal performance balance, within a vast high-dimensional continuous parameter space.

[0062] Step 1: Simulated annealing global exploration. Overall performance score: DJ = ω1 * conductivity + ω2 * high-temperature stability + ω3 * cycle retention;

[0063] Where DJ is the fitness function, and the weights are ω1 = 0.4, ω2 = 0.3, and ω3 = 0.3.

[0064] Starting from an initial temperature T0 = 1000, the temperature is reduced according to a cooling coefficient δ = 0.95. At each step, the probability of accepting a suboptimal solution is determined by the Metropolis criterion:

[0065] P = exp(-T)k / ΔDJ);

[0066] Where ΔDJ is the fitness difference, and T k The current temperature is used. This process iterates multiple times, aiming to escape local optima and find potential high-performance regions.

[0067] The second step involves Bayesian optimization of the local fine-grained search. Using the high-performance regions discovered through simulated annealing as prior knowledge, a Gaussian process surrogate model is constructed. The expected improvement function is then used as the acquisition function, and multiple iterations are performed to accurately locate the Pareto optimal solution set.

[0068] The system of this invention uses simulated verification sample data to build and train a data-driven model. It employs a hybrid optimization strategy for the multi-objective optimization engine of CAPTCHAs. Comparative tests were conducted in a simulated data center environment using 10 samples, each including details shown in Table 1.

[0069] Input parameters:

[0070] LiPF6 concentration (M): Lithium salt concentration, which affects ionic conductivity and stability;

[0071] EC volume fraction (%): The proportion of the solvent ethylene carbonate, which affects film formation and viscosity;

[0072] EMC volume fraction (%): The proportion of solvent ethyl methyl carbonate, complementary to EC to 100%;

[0073] LiPO2F2 concentration (%): Concentration of film-forming additive, which improves interfacial stability;

[0074] Mixing temperature (°C): A process parameter that affects the uniformity of dissolution;

[0075] Stirring time (h): A process parameter that affects the degree of mixing.

[0076] Output performance:

[0077] Ionic conductivity (mS / cm): measures the conductivity of an electrolyte.

[0078] High temperature stability (%): Capacity retention after storage at 60°C for 24 hours;

[0079] Cycle retention (%): Capacity retention after 500 cycles at a 0.5C rate;

[0080] Overall performance score DJ: Calculated as DJ = 0.4 × ionic conductivity + 0.3 × high temperature stability + 0.3 × cycle retention rate.

[0081] Table 1 Training Data Driven Model Data Table

[0082]

[0083]

[0084] Furthermore, the result output and verification module receives actual verification data and retrains the data-driven model to form a closed-loop optimization. When the average absolute percentage error between the actual verification data and the corresponding model prediction data is greater than 10%, the model retraining process is automatically triggered, and the actual verification data is used to perform incremental learning on the model.

[0085] In this embodiment, the dataset is used to verify the accuracy of the Pareto optimal solution set output by the multi-objective optimization engine. The system of this invention updates model parameters through incremental learning to achieve closed-loop optimization. To verify the updating of model parameters and the achievement of closed-loop optimization, a comparative test with three verification samples was conducted in a simulated data center environment. Experimental verification was carried out based on the parameter combination recommended by the optimization system. For details, please refer to Table 2.

[0086] Table 2 Updated Data-Driven Model Data Table

[0087]

[0088] Furthermore, a method for synergistic optimization of multiple performance aspects of a liquid battery electrolyte is characterized by comprising the following steps, detailed in the following document. Figure 2 :

[0089] S1: Obtain the historical formula parameter set and process parameter set of the electrolyte and the corresponding historical performance test results;

[0090] S2: Receive multiple performance constraints set by the user;

[0091] S3: Based on the historical data, train a data-driven model;

[0092] S4: With the goal of maximizing the overall performance score, the data-driven model is invoked to perform iterative simulation calculations and output the Pareto optimal solution set;

[0093] S5: Output the Pareto optimal solution set and receive actual verification data from the user for a specific solution to update the data-driven model.

[0094] Furthermore, the steps for updating the data-driven model include: comparing the actual validation data with the corresponding simulated prediction data and calculating the mean absolute percentage error; when the error exceeds a preset threshold of 10%, triggering the model retraining process and updating the model parameters using an incremental learning algorithm.

[0095] To verify the effectiveness of the multi-performance synergistic optimization system for liquid battery electrolytes, a comparative test was conducted in a simulated data center environment. The differences in key R&D indicators between the traditional trial-and-error method and the optimization system of this invention, using the same high-performance electrolyte (target: ionic conductivity ≥11.0 mS / cm, high-temperature stability ≥80%), are detailed in Table 3.

[0096] Table 3 Comparison between Traditional Time-Fault Method and Multi-Performance Collaborative Optimization System

[0097]

[0098] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0099] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A liquid battery electrolyte multi-performance synergistic optimization system, characterized in that, The system comprises a processor, a memory, and a computer program stored on the memory; the computer program is configured to implement the following program modules when executed by the processor: A parameter acquisition module configured to acquire a historical data set of a liquid battery electrolyte, the historical data set comprising a formulation parameter set and a process parameter set; A constraint condition preset module configured to receive input of a plurality of performance constraint conditions corresponding to a target battery application scenario; A data-driven model construction module configured to construct a data-driven model mapping electrolyte parameters to a comprehensive performance score based on the formulation parameter set, the process parameter set, and performance test results obtained under the performance constraint conditions; A multi-objective optimization engine configured to call the data-driven model for iterative simulation calculation with the goal of maximizing the comprehensive performance score, and output a Pareto optimal solution set; A result output and verification module configured to output the Pareto optimal solution set, receive actual performance data obtained by a user based on a specific solution for experimental verification, and update the data-driven model using the actual performance data.

2. The system for multi-property synergistic optimization of a liquid battery electrolyte of claim 1, wherein, The performance indicators targeted by the performance constraint conditions include two or more of ionic conductivity, high-temperature stability, wide-temperature-range performance, wettability with electrode materials, and dendrite suppression ability.

3. The system for multi-property synergistic optimization of a liquid battery electrolyte of claim 1, wherein, The formulation parameter set includes the concentration of lithium salt LiPF6, the volume ratio of solvents ethylene carbonate (EC) and methyl ethyl carbonate (EMC), and the concentration of film-forming additive LiPO2F2; the process parameter set includes mixing temperature and stirring time.

4. The system for multi-property synergistic optimization of a liquid battery electrolyte of claim 1, wherein, The data-driven model is a machine learning model; the machine learning model is a neural network model containing an attention mechanism, used to capture the nonlinear relationship between formulation parameters, process parameters, and battery long-cycle performance. The neural network model assigns different weights to different input parameters through an attention weight calculation function to capture the nonlinear coupling relationship between parameters and performance and outputs a comprehensive performance score. where a i is the attention weight, W a is the attention weight matrix, h i is the hidden state of the input feature, LeakyReLU is a nonlinear activation function, exp is an exponential function, is the normalization denominator.

5. The system for multi-property synergistic optimization of a liquid battery electrolyte of claim 3, wherein, The additive includes film-forming additive LiPO2F2, and its concentration is 0.5%-2% of the total mass of the electrolyte.

6. The system for multi-property synergistic optimization of a liquid battery electrolyte of claim 1, wherein, The multi-objective optimization engine adopts a hybrid optimization strategy, first performs global exploration based on a simulated annealing algorithm, and the probability P of accepting inferior solutions is determined by the Metropolis criterion: Wherein, Δf is fitness difference, T k is the current temperature; and based on the prior knowledge of the previous exploration results of the Bayesian optimization algorithm, a local fine search is performed, and finally a Pareto optimal solution set is output, wherein each solution represents a candidate electrolyte parameter combination.

7. The system for multi-property synergistic optimization of a liquid battery electrolyte of claim 1, wherein The result output and verification module re-trains the data-driven model by receiving actual verification data, forming a closed-loop optimization; when the average absolute percentage error between actual verification data and corresponding model prediction data is greater than 10%, the model re-training process is automatically triggered, and the model is incrementally learned using the actual verification data.

8. A method for multi-property synergistic optimization of liquid battery electrolyte, characterized in that, The method comprises the following steps: S1: Acquire historical formulation parameter set and process parameter set of the electrolyte and their corresponding historical performance test results; S2: Receive a plurality of performance constraint conditions set by a user; S3: Train a data-driven model based on the historical data; S4: Call the data-driven model for iterative simulation calculation with the goal of maximizing the comprehensive performance score, and output a Pareto optimal solution set; S5: output the Pareto optimal solution set, and receive actual verification data of a user on a specific solution, to update the data-driven model.

9. The method of claim 8, wherein the method is characterized by, The step of updating the data-driven model comprises: comparing the actual verification data with corresponding simulated prediction data, calculating the mean absolute percentage error; when the error exceeds a preset threshold of 10%, triggering a model retraining process, and updating the model parameters using an incremental learning algorithm.

Images