A K-means clustering algorithm was proposed and a conditional Wasserstein generative adversarial network with gradient penalty（CWGAN-GP）to address the problem of imbalanced photovoltaic generation data caused by the low occurrence probability of extreme weather. A prediction approach combining bidirectional long short-term memory（BFLSTM）with convolutional neural network was introduced and incorporating channel attention mechanism to enhance the PV power prediction performance by integrating spatio-temporal features and dynamically adjusting the importance of feature channels. Firstly，correlation analysis and K-means algorithm were utilized to select and label various environmental factors. Then，extreme weather labels with fewer samples after clustering were selected，and CWGAN-GP was used for data augmentation.Finally，the augmented dataset was used to train the CNN-SE-BiLSTM prediction model for PV power prediction under extreme weather conditions.Simulation modeling was conducted using data from a certain PV power station，and the results demonstrate that augmenting the original extreme weather training set with CGAN-GP helps improve the prediction accuracy of the model. Moreover，CNN-SE-BiLSTM shows higher prediction accuracy among five weather categories compared to other traditional models，indicating that the proposed method is suitable for ultra-short-term photovoltaic power prediction.

Accurate and efficient multi-load forecasting is of great significance for the operation control and scheduling of integrated energy system（IES），in order to improve the load forecasting effect，a integrated energy system load prediction model based on least absolute shrinkage and selection operator（LASSO）and LSTM-GRU neural network was proposed. Firstly，in order to solve the problem of complex data caused by meteorological factors in the integrated energy system，a big data selection and analysis algorithm based on LASSO was studied to select and analyze the meteorological factors to obtain an effective data set. Secondly，the long short-term memory（LSTM）neural network was used to predict the system load，and the preliminary prediction value was obtained. Subsequently，the gated recurrent unit（GRU）was used to construct the error compensation model，and the compensation value of the prediction error was obtained through the training and learning of the prediction error. Finally，by reconstructing the output of the two，a more ideal prediction result was obtained. Through the simulation of the example，the proposed prediction model has higher prediction accuracy than the traditional LSTM neural network prediction model and the LSTM model optimized by particle swarm optimizer（PSO）.

The three-core underground cable in the distribution network works in a harsh environment. Due to the influence of aging，external force，high temperature and humidity，the insulation of long-term live operation is easily damaged and causes faults. Aiming at the single-phase grounding fault of cable in distribution network，a fault location method based on the refraction-reflection characteristics of traveling wave head was proposed. Firstly，the zero-mode voltage of each line was obtained，and the arrival time of the wave head was determined. The waveform in a power frequency period after the arrival of the wave head was selected，and the main frequency of the wave head was determined. The square wave with the same frequency as the main frequency of the wave head was selected as the initial traveling wave signal，and the attenuation signal waveform at the head and end of the initial traveling wave signal after refraction of different times at this frequency was calculated.If the correlation between the corresponding data of the waveform calculated at both ends met the requirements，it was assumed that the fault point is the actual fault point. Otherwise，continue to narrow the fault interval until the correlation between the waveform data is satisfied. Simulation results show that the method can quickly and accurately realize fault distance measurement，reduce the complexity of fault distance measurement，and has strong effectiveness and practicability.

The back-to-back active neutral-point-clamped（ANPC）converter，as the core component of the unified power flow controller of the distribution network，is widely used for the flexible closed-loop operation.However，the ANPC converter has more switching devices and commutation paths，which significantly increases the complexity of modeling.To solve the above problems，an average modeling method of the ANPC converter was proposed.The controlled source equivalent substitution method was used to replace the switching devices.The resistance-capacitance-inductance devices of other parts of the circuit were retained while averaging each branch，which greatly simplified the circuit model of the ANPC converter. On this basis，combined with the flexible control principle based on back-to-back ANPC converters，the average model of a back-to-back ANPC flexible control system was further built. By comparing the simulation results of the switch model and the average model，it was shown that the average model was consistent with the switch model in the internal branch and external characteristics，which verified the correctness and accuracy of the proposed average model and its feasibility in the application of flexible closed-loop system.

With the emergence of a large number of distributed power generations，various unbalanced and nonlinear loads are connected to the power grid，resulting in the distortion of the grid current，which brings adverse effects to both the power grid and electrical equipment. A multifunctional energy storage converter was proposed，which is capable of simultaneously realizing the compensation of reactive current，harmonic current and unbalanced current of the load. This converter used a phase-locked loop based on double second-order generalized integrators to generate the required phase angle and frequency，and at the same time obtained the amplitude of the positive sequence component of the voltage.The converter obtained the command current by obtaining the positive sequence component of the load current and performing coordinate transformation，and then controled through the difference between the command current and the output current of the inverter to track the command signal.Proportional resonant control was used to obtain three-phase current，and pulses were generated through three-dimensional space vector control to control the conduction of the switch tube. By comparing with traditional energy storage converters，digital and analog hybrid experimental results have proven the effectiveness and reliability of this scheme.

The accurate identification of hidden danger for power utilization in low-voltage substations plays an important role in improving the quality of power supply and reducing the risk of accidents.To improve the accuracy of identifying hidden danger in low-voltage substations，a low-voltage user hidden danger for power utilization identification model based on SSAE-SSA-GRU was proposed. Firstly，the user's original voltage data was normalized，and the feature parameters of the data were extracted through a stacked spares auto-encoder（SSAE）to solve the redundancy problem caused by the high dimensionality of the original voltage data. Then，the sparrow search algorithm（SSA）was introduced to optimize the hyperparameters of the gated recurrent unit（GRU）network，improving the accuracy of the model's fault diagnosis results.Finally，the performance of the established SSAE-SSA-GRU model was evaluated through numerical examples，verifying the effectiveness of the proposed method in identifying hidden danger for power utilization for low-voltage users. Compared with traditional methods for identifying abnormal electricity usage，the proposed method has good convergence and high accuracy.

The traditional predictive current control for three-phase LCL grid connected inverters has the problems of large steady-state ripple and unstable switching frequency，which is not conducive to the quality of grid connected current and the design of filters. Therefore，an improved three vector fixed frequency predictive current control method was proposed.By applying three vectors within one control cycle，the control degrees of freedom were maximized. Firstly，an active damping method based on capacitor voltage feedback was designed to suppress resonance. Then，a duty cycle calculation method based on inverter side current deadbeat control and a seven segment symmetric pulse generation scheme were proposed. The simulation results and test results verified the superiority of the proposed method in improving dynamic performance，steady-state performance and fixing switching frequency.

Space vector pulse width modulation（SVPWM）strategy is one of the key technologies for three-level inverters.However，conventional three-level SVPWM often experiences neutral point potential（NPP）oscillation and high switching loss when addressing the neutral point potential balancing issue.To tackle this challenge，firstly，the fundamental principles of the traditional nearest three vector（NTV）synthesis strategy and the virtual space vector pulse width modulation（VSVPWM）strategy were reviewed. And then，the drawbacks of the NTV approach based on the conventional region partitioning method was analyzed. A dynamic partitioning method was proposed to resolve the neutral point potential offset caused by conventional partitioning.Furthermore，to counteract the NPP oscillation of the NTV under high modulation，a strategy that employs a synthesis of four-effective voltage vectors was introduced. Compared to VSVPWM，the proposed strategy significantly reduced the inverter's switching loss.Ultimately，the effectiveness and superiority of the proposed algorithm were validated through experimental results.

In response to the existing problems of traditional diode based clamp type wind turbine inverters，a hybrid clamp type structure was introduced. Based on the analysis of its working principle，a method for switching the switch state and a vector optimization selection method that can avoid modulation waveform burr line voltage were proposed，achieving balanced control of clamp capacitor and DC bus capacitor voltage. A third-order low-pass filter method with initial angle calculation was proposed to optimize the virtual flux orientation control strategy for the control of hybrid clamp type rectifiers in the unit. A direct power tracking control method for rectifiers based on improved virtual flux orientation was also presented. Finally，simulation and experiments were conducted on the system，confirming the correctness and effectiveness of the proposed method.

The load of a submerged arc furnace has many advantages，such as large single-unit regulation capacity，good communication conditions，easy control，and low economic retrofitting，making it an important flexible regulation resource for the power grid.However，there is currently a lack of effective digital simulation methods for modeling submerged arc furnace systems and their regulation characteristics.A time-domain digital simulation model for the power supply system of an industrial submerged arc furnace was established，which included an electric furnace transformer，short network，furnace body，and reactive power compensation，based on the actual electrical operating conditions of the submerged arc furnace.The power regulation characteristics of the submerged arc furnace were determined by representing the electrode regulation link with the time-varying arc length calculated through the Newton downhill method.The equivalent time-varying resistance was obtained by solving the arc differential equation using the Runge-Kutta method，enabling simulation of the power regulation process of the submerged arc furnace.This was achieved by encapsulating it with a custom module and integrating it with the system model to realize time-domain digital simulation.It is applicable to both AC submerged arc furnace and various electrode control strategies，and can be widely used for simulating the load characteristics of submerged arc furnace，flexibility resource assessment，and as a simulation platform for participating in grid interaction research.