The wireless charging technology for electric boats charging applications has gradually attracted attention. Due to the long wireless charging transmission distance of electric boats, the coupling coefficient and anti-misalignment capability of a magnetic coupler will decrease sharply with an increase in the transmission distance, thus reducing the transmission efficiency and output power. A novel solenoid magnetic coupler is proposed, and the finite element simulation models of magnetic couplers with different structures are built to compare their performance. In addition, the proposed solenoid magnetic coupler is optimized, and the corresponding magnetic shielding device is configured. Compared with those of the traditional solenoid magnetic coupler, the use of magnetic core and Litz wires of the novel solenoid magnetic coupler are reduced by 32% and 38%, respectively, thus effectively improving its coupling coefficient and anti-misalignment capability. Finally, a set of wireless charging prototype was built, and the transmission efficiency can reach 88%. The experimental results were consistent with the simulation results.

For a wireless power transfer system with one single planar rectangular pad, a metal object detection (MOD) method based on vertically decoupled coil sets is proposed. The detection coils are combined in a reasonable manner to form a detection coil set, which is decoupled from the transmitting and receiving coils simultaneously. When there is no metal object, the induced voltage of the detection coil set will be close to zero and minimally affected by the offset of the receiving coil. When a metal object falls on the top of the transmitting coil, eddy currents will be generated and induced voltage will appear in the detection coil set. The detection accuracy of the proposed method is not significantly affected by the offset of the receiving coil, and there is no need to reset the detection circuit or adjust the threshold for different offset positions. The parameters of the detection coil are optimized by finite element simulations. Based on a 300 W wireless charging system, MOD experiments were conducted, and results show that the proposed method can accurately identify metal objects even in offset situations. This method solves the problem of a planar detection scheme being affected by the offset of the receiving coil, with high detection sensitivity and great engineering application value.

Aimed at the problems that the capacity of a lithium-ion battery is difficult to obtain in the direct prediction of its remaining useful life (RUL) and the characteristic parameters are redundant or insufficient in the indirect prediction, an indirect prediction method for the RUL of lithium-ion battery based on the data preprocessing technology and improved support vector regression (SVR) is proposed. First, multiple indirect characteristic parameters are extracted at the battery charging and discharging stage. Second, principal component analysis is used to remove the redundancy of various parameters, and a fusion health indicator (HI) with sufficient information is reconstructed. Third, a fusion HI prediction model and a capacity prediction model based on the whale optimization algorithm and SVR are built, and the complete ensemble empirical mode decomposition with adaptive noise is used to decompose the fusion HI into several modal components. Each component is input into the fusion HI prediction model for HI prediction, and the prediction results are superimposed into the capacity prediction model to realize indirect RUL prediction. Finally, the NASA battery degradation data set is used for verification, and results show that the mean absolute error and root mean square error of RUL prediction results for the proposed method are controlled within 1.78% and 2.5%, respectively, which effectively improves the prediction accuracy of RUL. The research provides a new idea for the RUL prediction of lithium-ion batteries.

In view of the poor cooling effect on a lithium-ion battery module (LIBM) when using one single cooling method, a phase change material (PCM)-air cooling coupled battery cooling system was proposed. First, the influences of different inlet air velocities (IAVs), battery gaps (BGs) and inlet air temperatures (IATs) on the highest temperature and maximum temperature difference of LIBM were discussed. Second, with the IAV, BG and IAT as design variables and the highest temperature and maximum temperature difference as objective functions, the design variables and objective functions were numerically fitted by a response surface model. Third, the non-dominated sorting genetic algorithm-II is used to optimize and predict the response surface model. Finally, the optimization and prediction results were verified by simulations based on computational fluid dynamics. Results show that the optimized cooling system can reduce the highest temperature in LIBM by 4.68 ℃ (13.8%) and the maximum temperature difference by 0.57 ℃(30.3%).

In the traditional prediction of lithium batteries for electric vehicles, the state-of-health (SOH) prediction is usually regarded as a whole, and the result of single SOH prediction is obtained accordingly. However, in the actual operation of a car, the single prediction of SOH has a large error, and its prediction effect is not satisfying. To improve the accuracy of battery SOH prediction for electric vehicles, a novel prediction method based on variational modal decomposition (VMD) and sparrow search algorithm (SSA) optimization of kernel-based extreme learning machine (KELM) integrated prediction model, i.e., VMD-SSA-KELM, is proposed. First, the battery SOH sequence is decomposed by VMD to reduce the influence of SOH fluctuations. Meanwhile, the Person correlation method is used to reduce the influence of noise and improve the accuracy of prediction. The KELM is introduced, which improves the accuracy of prediction while retaining the advantages of extreme learning machine. The proposed model is validated based on the operation data of four electric vehicles, and experimental results show that compared with the VMD-DBO-KELM, VMD-POA-KELM, VMD-KELM and VMD-ELM models, the proposed model has a prediction trend which is the same as that of the original data, while the results of other models fluctuate a lot. The root mean square error of results predicted by the novel model is less than 0.2%, the prediction accuracy becomes higher, the prediction efficiency is faster and the time used is shorter, indicating that the proposed method has higher accuracy and better robustness.

The accurate open-circuit voltage-state of charge (OCV-SOC) curve is a basis for ensuring the modeling accuracy of lithium-ion battery. The OCV-SOC curves of LiFeO₄ battery obtained by a low-current OCV test cannot describe the OCV characteristics at a non-testing point, while those obtained by an incremental OCV test are interfered by the polarization effect. Therefore, based on the analysis of the characteristics of LiFeO₄ battery, a high-precision OCV-SOC curve acquisition method for LiFeO₄ battery is proposed by combining the low-current OCV test and incremental OCV test. This method takes the incremental discharge curve which is fitted in piecewise form as its optimization object, designs constraints based on the low-current OCV test data and first-order RC equivalent circuit model, and uses the differential evolution method to acquire the OCV-SOC optimization curve. Experimental results show that the OCV-SOC optimization curve can accurately simulate the OCV characteristics of LiFeO₄ battery. Compared with the OCV-SOC curve obtained by the low-current OCV test, the battery modeling and SOC estimation based on the OCV-SOC optimization curve has a higher accuracy, with the model accuracy increasing by 41.8% and SOC estimation accuracy increasing by 58.3%.

To improve the fuel economy of fuel cell buses, an energy management strategy based on model predictive control was proposed to solve the problem that the rule-based energy management strategy depends on engineering experience and parameter calibration. This strategy was implemented in a layered control manner to reduce the amount of computation. At the top level, a dynamic programming energy management strategy was used to determine the best reference trajectory for the lithium battery's state-of-charge (SOC). At the bottom layer, radial basis function neural network was used to predict the future vehicle speed, with the goal of predicting the efficient output of fuel cell power and SOC to follow the best reference trajectory in the time domain. In addition, the optimal decision making on power distribution was solved by the quadratic programming algorithm. The determination of the speed prediction model and prediction time-domain, the influence of SOC reference trajectories on the proposed strategy, and the changes in SOC and fuel economy under different strategies were studied in detail. Results show that when the initial SOC is 0.7, the fuel economy under the proposed model predictive control-based strategy can reach more than 92.99% of the fuel economy under the dynamic programming strategy, which is at least 11.78% higher than that under the rule-based strategy. Furthermore, the proposed strategy exhibits better fuel economy as the initial SOC increases.

The service performance and remaining life of a supercapacitor which serves as a novel energy storage device are part of the factors affecting the operation reliability of a system. A 2.7 V/10 F supercapacitor was taken as the research object, an aging experiment and test platform for energy storage devices was built, and the rules for the performance degradation of the supercapacitor and influencing factors were analyzed by collecting the variations in its characteristics. To solve the contradictory relationship between small samples and prediction accuracy of the supercapacitor lifetime prediction training set, a lifetime prediction method combining the non-dominated sorting genetic algorithm II (NSGA-II) with an elite retention strategy and support vector regression (SVR) was proposed, and the average values of charge-discharge cycles and charge-discharge duration were determined as the characterization of supercapacitor performance degradation. Moreover, a dual-objective model with the minimum sample size and highest prediction accuracy in the training set was established. Compared with the BP (back propagation), SVR and LSTM (long short-term memory) methods, the proposed NSGA-II-SVR method improved the evaluation indicators of root mean square error and mean absolute percentage error by 5 to 10 times in supercapacitor lifetime prediction under the premise of a small sample size of the training set. Finally, a generalization experiment was carried out using the open-source data set of lithium-ion batteries provided by NASA, proving that the NSGA-II-SVR method has strong generalization capability.

The accurate chip junction temperature prediction of power semiconductor modules is significant for their reliable operation. However, the research about freewheeling diode (FWD) chip junction temperature prediction is not sufficient at present, and the models of temperature sensitive electric parameters and the corresponding evaluation mechanism are lacking. A peak reverse recovery current model is deduced based on the turn-off process of FWD, and the relationship of peak reverse recovery current with bus voltage, load current and junction temperature is established. Then, a parameter identification approach for the peak reverse recovery current model is proposed based on the particle swarm optimization algorithm with random mutation, and the result of double-pulse test is used to extract the model parameters and verify the model. To reduce the model error caused by the linear distribution and linear changes of carriers, the exponential terms of load current and bus voltage are further added, thus optimizing the prediction of peak current. Finally, a junction temperature prediction approach for FWD chips in power semiconductor modules is put forward based on the optimized model, and this approach is more accurate compared with the look-up-table method, indicating that it can satisfy the requirements of FWD junction temperature prediction.

To ensure the safe and reliable operation of a three-phase bridge leg, it is necessary to set a reasonable dead-time. Due to its own characteristics, the voltage variation in the dead-time of a gallium nitride high-electron mobility transistor (GaN HEMT) is different from those of traditional Si devices. In this paper, the problem of large reverse conduction voltage drops in GaN HEMT power devices is analyzed, and an improved online dead-time compensation method based on GaN HEMT power devices is proposed. This method can not only avoid the influence of circuit noise on the judgment of current direction, but also reduce the phase voltage error and current harmonics, thus improving the circuit stability. Finally, the effectiveness of the improved online dead-time compensation method was verified by simulation experiments and the construction of an experimental platform, indicating that it has obvious advantages over the traditional dead-time compensation method under the condition of large reverse conduction voltage drops.