The series impedance measurement device (SIMD) measures the impedance of a new energy grid-connected inverter by injecting voltage disturbances. However, the rated fundamental current of the inverter under test will flow through the SIMD and thus limit the switching frequency. To address this issue, an SIMD based on resonance shunt is proposed. The proposed topology and control strategy ensure the impedance measurement accuracy in the range of 5 Hz–1 kHz, and achieve a goal that the fundamental current hardly flows through the SIMD and thus the SIMD only bears a small disturbance current. At the same time, an impedance measurement method based on single-phase voltage disturbance injection is adopted to realize the impedance measurement of a three-phase inverter, which improves the device’s energy density and measurement efficiency. Finally, the validity of the proposed topology and control strategy was verified by hardware-in-the-loop experiments.

With the rapid construction of urban rail transit, the problems such as the low utilization rate of regenerative braking energy have become increasingly prominent. The energy storage system is one of the effective ways to solve the utilization problem of regenerative energy in urban rail, and its reasonable control can accelerate its response speed and improve the stability of urban transit systems. First, aimed at the hybrid energy storage systems of urban rail transit, the characteristics of vehicle-mounted and ground-mounted energy storage systems including their safety, flexibility, transmission loss and energy saving efficiency are analyzed. Second, with the combination of the characteristics of energy storage media, four main energy storage approaches (i.e., battery, flywheel, supercapacitor and hybrid energy storage) for the vehicle-mounted and ground-mounted energy storage systems are summarized separately. On this basis, five control strategies for hybrid energy storage systems are elaborated and analyzed from multiple perspectives, such as their control modes and application statuses. Finally, the key technologies of control strategies for urban rail hybrid energy storage systems are summed up, and the development direction in the future is also projected.

The accurate prediction of the futural capacity of lithium batteries is crucial for their health management systems. Aimed at the impact of capacity regeneration on the capacity sequences of lithium-ion batteries, a long short-term memory (LSTM) neural network-based remaining useful life prediction model is proposed, which is based on capacity sequence decomposition and reconstruction. First, the CEEMDAN technology is used to decompose the original sequence into various intrinsic mode components, so as to eliminate the phenomenon of capacity rebound. Second, the Hurst technique is utilized to reassemble each component into multiple systematically strong sub-sequences, which can reduce the time consumption during the testing phase. Third, the whale optimization algorithm is applied to multiple sub-sequences to optimize the parameters of LSTM, and the outcomes from various sets of predictions are further aggregated to derive the ultimate prediction results. Finally, the effectiveness of the proposed model is verified on two sets of lithium-ion battery failure datasets with different degradation characteristics. Experimental results show that the proposed method can maintain both the mean absolute error and root mean square error in certain ranges.

The state-of-charge (SOC) prediction for lithium-ion batteries is an important part of a battery management system, and accurately predicting their SOC is of significance for ensuring their safe operation. On this basis, empirical mode decomposition (EMD), long short-term memory (LSTM) recurrent neural network and Attention mechanism are used to build an SOC prediction model for a lithium-ion battery. The voltage, current, surface temperature and the voltage decomposed by EMD in its charging and discharging processes are taken as the characteristics of the model. The LSTM layer of the model can capture the battery time information from the historical cycle data of the lithium-ion battery, and the long-term dependence relationship is also summarized. The Attention layer of the model can capture the nonlinear and dynamic relationship between feature input and SOC output, so as to optimize the weight relationship between each cycle stage and each feature and further improve the SOC prediction accuracy. Experiments were carried out based on the public battery data set of University of Maryland, and results show that compared with that of a model which only uses LSTM, the average prediction errors of the proposed model under three different working conditions were reduced by 49.36%, 45.61% and 46.93%, respectively.

Capacity and impedance are usually used as direct characteristics to determine the health status of lithium-ion batteries, but they cannot be easily measured in a direct manner. Aimed at the problem of remaining useful life (RUL) prediction for lithium-ion batteries, an indirect prediction method based on expectile regression (ER) is proposed. First, several indirect health indicators which can characterize the performance decline of batteries are extracted from the charging process of batteries, and the correlations with capacity are analyzed using the Pearson and Spearman correlation analysis method. Second, a novel ER-extreme machine learning (ER-ELM) model is developed by combining with ELM. The BFGS quasi-Newton method is introduced to optimize the parameters of the ER-ELM model, and a lithium-ion battery RUL prediction model is established. Finally, the model is verified using the CALCE lithium-ion battery data set. Results show that compared with the standard ELM model, the ER-ELM model has stable output of prediction results, the maximum root mean square error of prediction results is 0.028 9, and all the values of goodness of fit are above 98%, indicating a better performance in terms of tracking capacity decline and regeneration.

To solve the problems in most dual-output wireless power transfer (WPT) systems such as unnecessary cross-coupling, low space utilization and requirement for additional compensation components, the circuit decoupling method was used to establish a constant-current and constant-voltage dual-output WPT system, which can realize load-independent constant-current and constant-voltage output and zero voltage switching with near zero phase angle characteristics. On this basis, a theoretical model of T-shaped decoupling circuit and dual-output WPT system was established, and mathematical theoretical derivation was also conducted to verify the correctness of the proposed system. In addition, the system was simulated and verified, and the simulation results were consistent with the theoretical derivation. Finally, a system magnetic coupler was designed, and an experimental prototype was built, with operating frequency of 85 kHz, constant-current output of 5 A and constant-voltage output of 60 V. Experimental results verified the correctness of the theoretical analysis, and the maximum efficiency of the system can reach 93.4%.

The hybrid power supply technology can effectively integrate the advantages of different power supply units, which is one of the key technologies to realize high power density pulse load power supply systems. Aimed at the problems in the existing aerospace servo pulse load power supply system such as a large capacity and a low power density, a battery-active capacitor pulse load hybrid power supply system and the corresponding optimal capacity configuration technology are proposed. First, the system architecture and operation state are analyzed, and the optimization target and key constraints are established. Second, the capacity of the battery and active capacitor are optimized using an analytical method, and the distribution manner of pulse load power between the battery and active capacitor is established, so as to realize the volume minimization of the power supply system. Finally, an experimental prototype of the pulse load power supply system with a constant power of 2.7 kW and a peak power of 35.0 kW was constructed in the laboratory, and experimental results verified the theoretical analysis.

To solve the problem of common-mode voltage spikes caused by dead-time in the traditional model predictive pulse pattern control algorithms, an improved model predictive pulse pattern control algorithm was proposed. First, the pulse pattern design approach of the traditional model predictive pulse pattern control algorithm was introduced. Then, the influence of dead-time on this algorithm was analyzed, and the pulse pattern was optimized considering the dead-time effect to eliminate the common-mode voltage spikes generated during the dead-time. Finally, the traditional model predictive control based on common-mode voltage suppression, the traditional model predictive pulse pattern control algorithm and the proposed algorithm were experimentally compared. Experimental results show that compared with the traditional model predictive pulse pattern control algorithm, the proposed algorithm effectively suppressed the common-mode voltage amplitude to one-sixth of DC-side voltage without significantly increasing the total harmonic distortion of current ortorque ripple, thereby eliminating the common-mode voltage spikes during the dead-time.

To address the computational burden associated with traditional model predictive control strategies when a permanent magnet synchronous motor (PMSM) is driven by three-level converters, an optimization strategy is proposed. First, a discretized model of the PMSM is developed. Second, the Nelder-Mead algorithm is applied to the selection of optimal voltage vectors for the three-level converter, iteratively seeking the voltage vector that minimizes the current control errors. Finally, the output voltage vector is determined in conjunction with the DC-side capacitor voltage. The proposed optimization strategy involves a maximum of seven current predictions within a control cycle. Experimental validation demonstrates that compared with the traditional model predictive control strategy, the proposed optimization strategy reduced the computation time by approximately 45%, without compromising the steady-state control performance. Moreover, this strategy exhibited excellent DC-side capacitor voltage balancing capability.

Under the indirect vector control, the output torque from an induction motor is easily affected by changes in the rotor time constant. To improve the torque control accuracy of the motor, an online correction algorithm for the rotor time constant is necessary. However, the existing parameter correction algorithms mostly rely on accurate mathematical models of the motor, which are susceptible to model parameter errors and other non-ideal factors, making it difficult to ensure the accuracy of output torque. To address these issues, based on the traditional reactive power-based rotor time constant correction algorithm, the calculation errors of reactive power in actual systems is considered, and a method for calibrating and compensating for reactive power calculation errors is proposed. This method can reduce the output torque error of induction motors from 4% to less than 1%, and its effectiveness was verified by experimental results.