Design of an efficient battery model using evolutionary algorithms.

Deep neural networks (DNNs) are constructed by considering highly complicated configurations including network structure and several tuning parameters (number of hidden states and learning rate in each layer), which greatly affect the performance of speech processing applications. To reach optimal performance in such systems, deep understanding and expertise in DNNs is necessary, which limits the development of DNN systems to skilled experts. To overcome the problem, this paper proposes an efficient optimization strategy for DNN structure and parameters using evolutionary algorithms. The proposed approach parametrizes the DNN structure by a directed acyclic graph, and the DNN structure is represented by a simple binary vector. Genetic algorithm and covariance matrix adaptation evolution strategy efficiently optimize the performance jointly with respect to the above binary vector and the other tuning parameters. Experiments on phoneme recognition and spoken digit detection tasks show the effectiveness of the proposed approach by discovering the appropriate DNN structure automatically.

Surrogate-assisted evolutionary algorithms provide an effective means for complex and computationally expensive optimization problems. However, due to the scarcity of training samples, the prediction accuracy of frequently-used regression surrogate models can hardly be guaranteed as the difficulty of the problem increases, resulting in performance degradation of the whole algorithm. Since real-world problems rarely exist in isolation, it is expected to alleviate the above issue by properly exploiting the knowledge shared across different problems. In this context, this study proposes a novel evolutionary multitasking optimization algorithm assisted by a classifier rather than a regression model for expensive multitasking problems, where the accuracy of the classifier is boosted by knowledge transfer. To be specific, a support vector classifier (SVC) is first developed and integrated into a classic evolutionary algorithm, i.e., covariance matrix adaptation evolution strategy (CMA-ES). With a low computational cost, it helps CMA-ES to prescreen parent solutions from the current population. Following that, a knowledge transfer strategy is designed to enrich the training samples for each task-oriented classifier by sharing high-quality solutions among different tasks, where a PCA-based subspace alignment technique is employed. Extensive experiments indicate that the SVC-assisted CMA-ES gains significant superiority over general CMA-ES in terms of both robustness and scalability, and the knowledge transfer strategy further helps it earn a competitive edge over some state-of-the-art algorithms on expensive multitasking optimization problems.

ReaxFF is a computationally efficient force field to simulate complex reactive dynamics in extended molecular models with diverse chemistries, if reliable force-field parameters are available for the chemistry of interest. If not, they must be optimized by minimizing the error ReaxFF makes on a relevant training set. Because this optimization is far from trivial, many methods, in particular genetic algorithms (GAs), have been developed to search for the global optimum in parameter space. Recently, two alternative parameter calibration techniques were proposed, i.e. Monte-Carlo Force Field optimizer (MCFF) and Covariance Matrix Adaptation Evolutionary Strategy (CMA-ES). In this work, CMA-ES, MCFF and a GA method (OGOLEM) are systematically compared using three training sets from the literature. GA shows the smallest risk of getting trapped into a local minimum, whereas CMA-ES is capable of reaching the lowest errors for two third of the cases. For each method, we provide reasonable default settings and our analysis offers useful guidelines for their usage in future work. An important side effect impairing the parameter optimization is numerical noise. A detailed analysis reveals that it can be reduced, e.g. by using exclusively unambiguous geometry optimizations in the training set. Even without this noise, many distinct near-optimal parameter vectors can be found, which opens new avenues for improving the training set and detecting overfitting artifacts.

This article addresses various challenges associated with lithium-ion battery modeling. Lithium-ion batteries have a key role to play in mobile energy storage. One can potentially expand the envelope of lithium-ion battery performance, efficiency, safety, and longevity by using fundamental electrochemistry-based models for battery control. There are clear trade-offs between battery model fidelity and complexity, and a significant literature addressing these trade-offs. Electrochemistry-based battery models can be effective at capturing frequency-domain battery dynamics, especially at lower frequencies. When they are examined in this light, the commonalities between them and equivalent-circuit models become more visible. Constructing lithium-ion battery models certainly takes effort, and so does reducing these models for control design purposes. One important open challenge in lithium-ion battery modeling is the matching of sophisticated battery models to experimental data. Half-cell testing or insertion of a third reference electrode in a fuel cell can separate the contributions of the negative and positive electrodes, and researchers are pursuing other novel technologies for in-cell instrumentation and measurement.

The modeling of lithium-ion battery is an important element to the management of batteries in industrial applications. Various models have been studied and investigated, ranging from simple to complex. The second-order equivalent circuit model was studied and investigated since the dynamic behavior of the battery is fully characterized. The simulation model was built in Matlab Simulink using the Kirchhoff Laws principle in mathematical equations, while the battery's internal parameters were identified by using the BTS4000 (Battery tester) device. To estimate the full state of charge (SOC), the initial state of charge (SOC0) must be identified or measured. Hence, this paper seeks for the SOC estimation by using experimental terminal voltage data and SOC with Matlab lookup table. Then, the simulated terminal voltage, as well as the SOC of the battery are compared and validated against measured data. The maximum relative error of 0.015 V and 2% for terminal voltage and SOC respectively shows that the proposed model is accurate and relevant based on the error analysis.

Research on state-of-charge estimation of lithium-ion batteries based on an improved gas-liquid dynamics model

This paper deals with the problem of comparing and testing evolutionary algorithms, that is, the benchmarking problem, from an analysis point of view. A practical study of the application domain of four representative evolutionary algorithms is carried out using a relevant set of real-parameter function optimization benchmarks. The four selected algorithms are the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) and the Differential Evolution (DE), due to their successful results in recent studies, a Genetic Algorithm with real parameter operators, used here as a reference approach because it is probably the most familiar to researchers, and the Macroevolutionary algorithm (MA), which is not widely known but it shows a very remarkable behavior in some problems. The algorithms have been compared running several tests over the benchmark function set to analyze their capabilities from a practical point of view, in other words, in terms of their usability. The characterization of the algorithms is based on accuracy, stability and time consumption parameters thus establishing their operational scope and the type of optimization problems they are more suitable for.

In order to study the temperature distribution in a lithium-ion battery (LIB), a series of experiments including open-circuit voltage (OCV) test, discharge rate test, hybrid pulse power characteristic (HPPC) test, and discharge temperature rise test based on the Bernardi heat generation model were conducted. The thermal model (TM) of the LIB under different discharge rates was established. The temperatures of the cell at different discharge rates calculated by the TM, which leverages the user-defined function (UDF) to import a variable heat generation rate, were compared with the experimental data. Moreover, this TM was used to simulate the LIB temperature at different ambient temperatures and different heat-transfer coefficients and analyze the temperature distribution of cell. The results showed that the TM can accurately simulate the temperature rise of the LIB at different discharge rates, which provides a reference value for LIB thermal management.

Battery fast charging strategies have gained an increasing interest toward the convenience of battery applications but may unduly degrade or damage the batteries. To harness these competing objectives, including safety, lifespan, and charging time, in this article, we propose a novel health-aware multiobjective optimal charging strategy to simultaneously shorten the charging time and relieve the battery degradation. The multiobjective optimal charging problem is formulated based on a coupled electrochemical-thermal-aging battery model. Constraints are explicitly imposed on physically meaningful state variables to avoid hazardous operations. Charging duration and battery aging process are well traded-off. Strategies for minimum-time and health-aware fast charging are investigated using different input current bounds, subject to both side reaction and temperature constraints. The experimental results validate that the presented multiobjective health-aware optimal charging algorithm is capable of reducing the charging time from its benchmarks largely without sacrificing the state-of-health of the battery.

The incorporation of physics-based lithium-ion battery models in real-time optimization, control and estimation has the potential to substantially facilitate their efficient and safe operation.1,2 In addition to improved prediction of battery states, the access to predictions of variables internal to battery electrochemistry allows the formulation of more complete optimization and control problems than would be possible by simple circuit models.3 Various approaches have been employed to enable this integration by reducing the computational demands of electrochemical models, particularly for the macro-homogeneous ‘P2D’ models of Newman and co-workers.4 These may be broadly divided into two categories: simplifications of the full model subject to limiting assumptions,5–7 or reformulation techniques which exploit the underlying structure of model equations to achieve fast simulation.1,8,9 The single particle model (SPM) visualizes the electrodes in a lithium-ion battery as two spherical particles and considers the electrode reactions and lithium transport through the active particle. Importantly however, it neglects potential and concentration variations in the electrolyte phase, which limits its predictive ability to low-current scenarios. Reformulated P2D models offer the desired computational performance without sacrificing physical detail, but their utility is dependent on the precise estimation of a large number of parameters from experimental data.10 This talk focuses on the development of a lumped physics-based model for lithium-ion batteries. The governing equations of the P2D model are integrated over the volume of each region in a cathode-separator-anode representation. This results in a set of volume-averaged quantities and relations, with the temporal evolution of each averaged variable expressed as an overall balance containing internal source terms and interfacial fluxes, which need to be approximated. The averaged porous domains may thus be regarded as three ‘tanks-in-series’. The model equations are simulated, and the electrochemical variable predictions compared against both the full P2D model and SPM over a range of parameter combinations and operating conditions. Various approximations for the interfacial fluxes are also evaluated in terms of their impact on cell-level predictions. The model is also extended to incorporate thermal effects via volume-averaged energy balances. Model predictions for different galvanostatic discharge rates suggest that the lumped model is of intermediate sophistication between the SPM and the P2D models. This is most likely because it incorporates liquid phase transport in an average sense, in contrast to SPM, which ignores them entirely. The lumped model is conservative, and possesses the computational simplicity of SPM, while also incorporating elements from the P2D model, but with a smaller number of parameters, making it easier to parameterize and potentially facilitating use in real-time prediction and control. In addition, tank models have been extensively researched in control theory, enabling the extension of these concepts to battery systems through lumped models. The ‘tanks-in-series’ averaging procedure described herein may be generalized to any electrochemical battery model. Acknowledgements The authors are also thankful for financial support from the Battery500 Consortium and the Advanced Research Projects Agency (ARPA-E). The authors acknowledge financial support from the Department of Chemical Engineering and the Clean Energy Institute at the University of Washington. References A. M. Bizeray, S. Zhao, S. R. Duncan, and D. A. Howey, J. Power Sources, 296, 400 (2015).V. Ramadesigan, P. W. C. Northrop, S. De, S. Santhanagopalan, R. D. Braatz, and V. R. Subramanian, J. Electrochem. Soc., 159, R31 (2012).M. Pathak, D. Sonawane, S. Santhanagopalan, R. D. Braatz, and V. R. Subramanian, ECS Trans., 75, 51 (2017).M. Doyle, T. J. Fuller, and J. Newman, J. Electrochem. Soc., 140, 1526 (1993).S. Santhanagopalan and R. E. White, J. Power Sources, 161, 1346 (2006).B. S. Haran, B. N. Popov, and R. E. White, J. Power Sources, 75, 56 (1998).M. Guo, G. Sikha, and R. E. White, J. Electrochem. Soc., 158, A122 (2011).P. W. C. Northrop, V. Ramadesigan, S. De, and V. R. Subramanian, J. Electrochem. Soc., 158, A1461 (2011).P. W. C. Northrop, M. Pathak, D. Rife, S. De, S. Santhanagopalan, and V. R. Subramanian, J. Electrochem. Soc., 162, A940 (2015).V. Ramadesigan, K. Chen, N. A. Burns, V. Boovaragavan, R. D. Braatz, and V. R. Subramanian, J. Electrochem. Soc., 158, A1048 (2011).

A hierarchical active equalization topology based on Cuk converter and flyback converter was proposed to solve the problem of the inconsistency of series battery packs used in electric vehicle power cells. Using state of charge (SOC) as equalization standard, two equalization control strategies were developed, the second-order equivalent circuit model of lithium ion battery and open circuit voltage (OCV) characteristic curve were established to identify parameters and estimate SOC. The equalization of single battery in series battery was realized, and the inconsistency of series battery was reduced. Finally, a balanced experimental platform was built for verification. The layered active equalization framework gives full play to the advantages of the two kinds of active equalization, and improves the problem of the inconsistencies of the series battery. According to the SOC distribution of all the single cells in the series battery, the balancing control strategy is selected, which can effectively shorten the balancing time.

Lithium ion battery as a kind of new energy is a promising energy storage medium for electric and hybrid electric vehicles with their characteristics of lightness and high energy density. However, some accidents about battery fire or explosion remind us to focus on their reliability and safety issues. Thus, life prediction as the essential part should be considered during the design phrase of batteries to guarantee the safety and reliability level. In this paper, a new life prediction method based on gas production model of batteries has been proposed. Firstly, based on analysis result of actual gas production mechanism, types of chemical reactions existed in batteries can be determined. Then, through theoretical analysis of these chemical reactions combined with capacity change of battery, the whole gas behavior is divided into two stages. It includes the initial solid electrolyte interface formation stage and the stage of solid electrolyte interface being severely damaged and repaired. Gas production equation of each stage has been constructed. Based on these gas equations, quantitative relationship between the battery capacity and the characteristics of gas production is investigated. With a suitable end-of-life criterion, the application-oriented life prediction model for lithium ion batteries is established. Finally, the verification procedure is presented based on actual test data to validate our method and its advantages. This paper aims to achieve deeper understanding gas behavior of lithium ion batteries and further develop corresponding life prediction method to provide a theoretical basis for future design improvement.

This work investigates the modeling of lithium-ion batteries (LIBs) with a recurrent neural network (RNN), rather than with an equivalent circuit or similar type model as is typically used. The RNN is trained with dynamic battery data, such as vehicle drive cycle test results. Specialized characterization tests and model parameterization are not necessary, simplifying the process of battery modeling. A compact unified methodology consisting of an RNN with gated recurrent unit and deep feature selection structures is utilized. A total of two RNNs are evaluated, one with current as the input and another with power as the input. Both RNN forms accurately model LIB dynamic responses including battery nonlinear behavior at different temperatures. The models are compact in size, require fewer characterization tests compared to conventional equivalent circuit models, and can be further used as an LIB simulator in model-based design and hardware-in-loop applications to test battery management systems and other electronic components.

The covariance matrix adaptation evolutionary strategy (CMA-ES) is explored here as an improved alternative to well-established algorithms used in electromagnetic (EM) optimization. In the past, methods such as the genetic algorithm (GA), particle swarm optimization (PSO), and differential evolution (DE) have commonly been used for EM design. In this article, we examine and compare the performance of CMA-ES, PSO, and DE when applied to test functions and several challenging EM design problems. Of particular interest is demonstrating the ability of the relatively new CMA-ES to more quickly and more reliably find acceptable solutions compared with those of the more classical optimization strategies. In addition, it will be shown that due to its self-adaptive scheme, CMA-ES is a more user-friendly algorithm that requires less knowledge of the problem for preoptimization configuration.

This paper presents our study on the effect of perturbing variables and array size in optimizing a planar sparse array based on Danzer aperiodic tiling. A side lobe level below -12 dB has been achieved within a wide bandwidth from f 0 to 10 f 0 in a circular disk with radius of 5λ 0 , while below -14.2 dB in a circular disk with radius of 10λ 0 . It is also demonstrated that covariance matrix adaptation evolution strategy (CMA-ES) has outperformed other evolutionary algorithms, i.e. genetic algorithm (GA), particle swarm optimization (PSO).