Aiming at the problem of low recognition accuracy of electricity stealing behavior， an electricity stealing behavior recognition model based on joint neural network was proposed. Firstly， the acquired user electricity consumption data was processed， and the user electricity consumption data was two-dimensionally processed by using the Gramian angular field method. Then， for the electricity consumption data of different dimensions， a user electricity stealing behavior recognition model based on the joint neural network was proposed， and the features of the one-dimensional electricity consumption data and the two-dimensional electricity consumption data were extracted by using the convolutional neural network（CNN） and the bidirectional long short-term memory（BiLSTM） neural network. The case analysis shows that the proposed joint neural network model has an accuracy rate of more than 90% for the recognition of electricity stealing behavior， which proves that the established evaluation model provides a practical solution to the electricity stealing problem.

Permanent magnet synchronous motors play an important role in industrial automation， automotive， aerospace， household appliances， and other fields due to their high efficiency， high power density， good torque performance， and ease of control. However， in the field of electrical transmission， it is not only necessary to accurately control its operating status， but also to have complex industrial communication and human-machine interfaces. Traditional electrical drives mostly use proportional integral components for speed control， which may lead to integrator saturation issues， resulting in slower integration response after disturbance disappears. Complex industrial communication and human-machine interfaces require a large amount of chip resources and chip IO， resulting in high control costs. By improving the traditional proportional integrator and introducing output feedforward control， the response delay caused by integrator saturation was solved. By improving the external communication protocol and external human-machine interface hardware circuit， it solved the problem of resource and IO occupation that traditional transmission system design can only solve by increasing the number of chips. The improved system can complete all functions on a single chip 280039C DSP. Due to the absence of inter chip communication， the communication speed and response speed is greatly improved， and the on-site use is good.

In order to solve the problem of voltage and frequency fluctuations at the grid-connected point of the permanent magnet wind turbine during high and low voltage ride-through，a virtual synchronous generator （VSG） was proposed to control the grid-side converter to improve the inertia and damping of the permanent magnet wind power generation system，so that the voltage and frequency recovery of the grid-connected point could be supported. Through the active power-frequency adjustment function of the VSG，the active power output capacity of the grid-side converter was adjusted according to the system frequency variation，and the frequency fluctuation was suppressed under the action of the moment of inertia and damping coefficient. The virtual electromotive force was adjusted by adjusting the reactive power command of the VSG，so that the dynamic reactive power support could be provided for the grid during the voltage ride-through period，and the voltage of the grid-connected point could be adjusted. The continuous fluctuation of system power and frequency was effectively suppressed during the continuous high and low voltage ride-through period，and the grid-connected stability of the system was improved. Finally，the effectiveness of the proposed control strategy was verified by system simulation.

The semi-bridgeless dual Boost power factor correction（PFC） circuit can obtain higher conversion efficiency without increasing common mode noise，but the return current of the input AC terminal produces additional losses on the introduced diodes，which reduces the conversion efficiency. The replacement of diodes with MOS transistors was investigated. The MOS transistor was with very low impedance when turned on to short-circuit the MOS transister（or its body diode） and the inductor branch in parallel. Most of the return current was flowed through the introduced MOS transistor，the conduction loss was reduced and the conversion efficiency was improved. At the same time，a low impedance path was provided for the common mode noise through the MOS transistor and its body diode，and the commonmode noise level was not changed. The working process and design criteria of the circuit were given，and the correctness of the analysis was verified by circuit simulation. Finally，a 1.5 kW principle prototype was developed，in which it was verified that the circuit can further improve the conversion efficiency.

In order to improve the inertia support ability of grid-connected converter of new energy based on DC capacitor energy virtual inertia control to the power system，an adaptive virtual inertia control method based on linear quadratic regulator （LQR） was proposed. On the basis of analyzing the dynamic characteristics of the virtual synchronous generator （VSG），the mathematical relationship between the inertia parameters and the system frequency response characteristics was obtained，and an adaptive inertia control model based on LQR was established. The optimal feedback matrix was found by LQR cost function to make the state variables approach zero quickly and meet the control requirements with the least input energy. The simulation results show that the method can make the new energy grid-connected converter optimize the allocation of limited DC capacitance energy according to the system frequency fluctuation，quickly damp the frequency change，improve the system frequency response characteristics，enhance the inertia support ability of the new energy grid-connected converter to the power system，and ensure the stability of the power system.

The circulating current of metal sheath in high voltage cable is the superposition of induced current component and leakage current component. At present， the calculation of induced current component is relatively mature， but the calculation of leakage current component is insufficient， which affects the final circulating current calculation result and the related fault diagnosis effect. For this reason， a calculation method for leakage current component of sheath circulating current in high voltage cable was proposed. Based on the π-type equivalent circuit of "core—sheath"， an equivalent circuit for solving leakage current component of sheath circulating current was established， which take into account the shunting effect of leakage current and circulating current generated by leakage current of cable core. The final leakage current component was obtained based on the iterative method. A cable model with a linear arrangement was built using ATP-EMTP software. The results show that the relative deviation between calculation result and simulation value of this method is within 2% under different length combinations and load currents， compared with existing method， the accuracy is greatly improved. This method can better explain the reason of excessive circulating current at light load， and has a certain reference for the generation of sample data for fault diagnosis of cross-connected systems.

Three port converter （TPC） has the characteristics of high integration， high power density and single-stage conversion between ports， but at the same time， the energy flow is affected by the interaction of multiple variables. Model predictive control （MPC） was introduced to optimize the dynamic response of output voltage in order to improve the problem of slow response and large fluctuation of output value when traditional PI controller is used in the output voltage control loop of three port converter. The relationship between the prediction model and the control variable was obtained by modeling the average value of the state space of the three port converter， and the objective function was designed. The optimal value was obtained by traversal optimization， so as to improve the dynamic characteristics of the output voltage. The effect of output dynamic response of output voltage loop under traditional PI control and model predictive control was compared through experiments. The results show that model predictive control has obvious effect on improving output voltage stability.

The accurate recognition of power quality disturbance（PQD） is one of the main problems to be solved after PQD occurrence， which is of great importance for responsibility dividing and power market reform process accelerating. Massive quantities of power quality monitoring data prepare the ground for the recognition of PQD. Since the electrical characteristic is different for different PQD， the waveform difference between different power quality disturbances can be employed for the recognition of PQD. Combing the deep learning， the method for the recognition of complex PQD via bidirectional independently recurrent neural network（Bi-IndRNN）was proposed. In this way， the intrinsic characteristic of PQD was extracted， the internal correspondence between the input sequence and the output sequence was established， the dependence of the analysis result on the physical characteristic quantity was overcome， and the recognition accuracy of PQD was improved. The results illustrate that the diversity of complex PQD can be effectively responded， where the intrinsic characteristic hidden in complex PQD signal can be extracted directly， resulting in high accuracy.

Inverter products have been developing towards higher power density and more compact structure. To enhance the power density of inverter systems within a limited structural space， several prototypes need to be made and a large number of experiments conducted during the research and development process. To shorten the development cycle and reduce the number of prototypes and experiments， a thermal simulation method was proposed for the power unit of the inverter. Through this method， the system steady-state temperature rise of each scheme and the temperature rise of core components can be simulated during the research and development stage. It is possible to compare the advantages and disadvantages of different schemes without making prototypes or conducting experiments. Experimental results verify that the thermal simulation method has relatively accurate prediction of the cooling capacity of the thermal system. This method can effectively improve engineers' ability to control system thermal characteristics during the power unit design stage， quickly compare the advantages and disadvantages of different cooling solutions. It can also reduce the number of prototypes and experimental times， increase the first-time design yield， and reduce research and development costs.

Due to the extensive addition of new energy systems， the number and types of power quality disturbances in the system are also increased accordingly. However， the traditional power quality disturbance （PQD） classification method has the problem of low accuracy and efficiency， and it is difficult to adapt to the existing power quality management of power systems with high new energy penetration. Therefore， a PQD classification method based on graph convolutional neural networks （GCNNs） and Gramian angular field （GAF） was proposed. First， the original PQD signal was normalized and polar coordinate transformation was processed， then GAF was used to graphically transform different kinds of PQD one-dimensional signals to generate two-dimensional images containing different PQD features， and finally， GCNNs were used to train and classify the different kinds of PQD images to achieve the classification of different PQDs. In the experiment part， the IEEE-39 node system was used to simulate and simulate different types of PQD curves， and the method proposed was used for verification. The experiment results show that the proposed method can automatically extract and optimize the features， and meet the high efficiency and accuracy of PQD identification and classification.