LCL filter has superior high-frequency harmonic suppression capability, and therefore it has been widely used in grid-connected inverters. In order to suppress the resonance peak problem of LCL filters, active damping is usually used. Although the resonance peak is suppressed by this control strategy, the digital control delay can cause the inherent resonance point to shift. To solve these problems, a control strategy of active damping superposition based on the virtual impedance model is proposed, which not only suppresses the resonance peak but also avoids the shift of the inherent resonance point, and expands the effective damping zone to (0, f_s/3). Finally, the simulation system of LCL single-phase grid-connected inverter and the hardware in-loop prototype are built to verify the correctness of the control strategy. The experimental results show that the control strategy has good anti-interference ability, dynamic performance and steady performance.

Modular multilevel matrix converter （M3C） is a low-frequency power transmission AC-AC converter used for offshore wind power generation. In order to improve the reliability and stability of M3C operation, it is necessary to have an efficient and accurate diagnosis method for the open circuit fault of IGBT in its submodules. Therefore, a deep learning fault diagnosis method based on the combination of convolutional neural network （CNN） and gated loop unit （GRU） is proposed. On the basis of analyzing the operating conditions of M3C submodule, wavelet packet analysis is performed on the original fault data, the high-frequency components of which are converted into two-dimensional fault images through temporal image conversion as the training and validation dataset for deep learning, and the features of the high-dimensional data are extracted by CNN, and then the data is optimized and trained by GRU, so as to realize the diagnosis identification of the M3C fault categories. Compared to traditional methods, this method has more accurate and fast fault diagnosis capabilities.

At present, the detection of the thickness of the scale layer on the equalizing electrode in the converter station is carried out by shutting down the transmission system and manually disassembling, which has significant time, labor costs, and blind screening of the equalizing electrodes with the scale layer. There is an urgent need for a non disassembly online detection method for the scale layer on the equalizing electrode. Ultrasonic guided waves have the advantages of high detection efficiency and low energy attenuation. Therefore, based on ultrasonic guided waves, a non disassembly online detection method for the thickness of the scale layer on the equalizing electrode is proposed, and its effectiveness is verified through the simulation and experimental analysis. Firstly, by drawing the dispersion curve of polyvinylidene fluoride （PVDF） filled water pipeline, the approximate excitation frequency range of ultrasonic guided waves is determined to be 40~100 kHz, and the longitudinal L（0,2） mode is also determined. Then, a finite element model of the axial longitudinal section of the PVDF water filled pipeline is further established, and the optimal excitation frequency is determined to be 60 kHz through calculation. Based on this model, the variation patterns of scale layer thickness and received signal amplitude and time are analyzed. At the same time, an experimental platform is built to verify the ultrasonic guided wave detection method of the equalization electrode. The research results indicate that using longitudinal L（0,2） ultrasonic guided waves to detect the scale layer on the equalizing electrode has advantages such as fewer modes and weaker frequency dispersion, and has good discrimination ability for the equalizing electrode with different thickness of scale layers. This method provides a certain application basis for non-destructive testing of the scale layer on the equalizing electrodes.

The resistance-capacitance matching (RCM) relationship of 10 kV DC voltage transformer(DCVT) based on the principle of resistance capacitance voltage division (RCVD) is not easy to meet under actual working conditions, making its bandwidth unable to meet the requirements. According to the the equivalent circuit and parameters of 10 kV DCVT based on the principle of RCVD under actual operating conditions, the frequency characteristic transfer function is established, and the influencing factors of DC voltage divider ratio and frequency characteristic are analyzed. The quantitative effect of the initial accuracy of the resistor on the DC voltage divider ratio and the degree of influence of the initial accuracy of the capacitor and stray capacitance on frequency characteristic are obtained. A method of parallel connection of the resistance capacitance matching unit (RCMU) in the low-voltage arm of RCVD is proposed, which combines the selection of resistance-capacitance parameters of the main body of RCVD，and the DC voltage divider ratio can be calibrated through the adjustable resistor in RCMU, and the frequency characteristics of the DCVT can be compensated through the adjustable capacitor in RCMU. Finally, a prototype experiment is conducted, and the results show that the matching resistance calibrates the DC voltage division ratio from 10 000:5.015 to the design value 10 000:5, and the matching capacitance reduces the amplitude error of 10 kV DC voltage transformer from 14% to nearly 1% within the bandwidth of 50 Hz to 3 kHz.

As one of the most effective means to suppress low-frequency oscillations, power system stabilizers (PSS) have been widely used in power systems. The most widely used power system stabilizer (PSS) in China today is PSS2B. However, due to its own structural problems, PSS2B has a general suppression effect on low-frequency interregional oscillation modes. The new multi-band stabilizer PSS4B is expected to solve this problem. However, the multi-parameter and multi-degree freedom of PSS4B bring difficulties to its parameter tuning. An iterative design approach is proposed for PSS4B stabilizers based on the contoured controller Bode (CCBode) plot. A PSS4B stabilizer is viewed as two filters connected in series: a band-pass filter and a phase compensator. A space searching approach is used to tune the phase compensator which ensures that the phase frequency response of the PSS4B stabilizer is bounded within an acceptable limit. A CCBode plot then helps tune the magnitude frequency response of the band-pass filter so as to achieve a better stability performance over a wide range of frequency band. Case studies of Kundur′s four-machine-two-area system is reported to demonstrate the effectiveness of the proposed method.

With the expansion of power grid scale and the higher requirements of security, the difficulty in solving the security constrained unit commitment (SCUC) is increasing. Aiming at the characteristics of constraints of active power over transmission lines and 0-1 on/off integer variables in the SCUC, two prediction methods are constructed based on improved K-nearest neighbor algorithm, which are used to identify the active transmission power constraints and determine the values of partial integer variables respectively. At the same time, considering the influence of load parameters on the values of integer variables, the action interval of the integer variables is limited to improve the prediction accuracy. Before solving the problem, two forecasting methods can quickly predict the active transmission power constraints and the values of partial integer variables. Using this information, a simplified SCUC model can be built, and then the optimization solver can be used to solve the model directly, shortening the solution time of the SCUC. Finally, the correctness and effectiveness of the proposed method are verified by the standard test system and an actual provincial grid data.

The zero-sequence characteristic difference between faulty line and healthy line is weakened by resonant grounding system, so a line selection method based on transient zero-sequence admittance is proposed in this paper. Firstly, the transient zero-sequence admittance formulas of faulty line and healthy line under single-phase grounding fault are deduced theoretically, and the characteristic differences between them are compared. On this basis, a single-phase grounding fault line selection method based on the maximum mean value index of transient zero-sequence admittance is constructed. Finally, based on PSCAD/EMTDC software, the performances of the proposed method are verified and tested comprehensively. The results show that proposed method has the advantage of strong resistance tolerance to the transition resistance. Only the transient zero-sequence admittance in half power frequency cycle after the fault needs to be calculated, so the calculation load is small, the speed is fast and the implementation is easy. At the same time, the line selection accuracy is still high under the interference factors such as noise interference, current transformer reverse connection, the changes in network topology, line parameters and sampling frequency.

High propotional distributed photovoltaic integration changes the operation mode of the distribution networks, and leads to a series of problems such as excessive active power losses, reduced service life of regulating equipment, and exceeding node voltage limits in the distribution networks. Based on this background, firstly the voltage and reactive power optimization problem is modelled as a Markov decision process, which is solved by using a model-free deep reinforcement learning method that captures the intermittence of PV and load fluctuation from historical operating data. A graph convolutional network-proximal policy optimization (GCN-PPO) algorithm is proposed which improves the perception of reinforcement learning agent on graph data of distribution networks by embedding the graph convolutional network. Finally, an arithmetic analysis is carried out with a modified IEEE 33-node test system to verify the effectiveness of the proposed method and its advantages over other methods. The results show that the trained reinforcement learning agent based on graph convolutional networks exhibits better performance when the topology of the distribution network changes and the measurement data are lost.

Medium-term load forecasting is influenced by multiple external variables such as temperature, holidays, and weekends. Although long short-term memory (LSTM) networks have shown strong predictive ability in short-term load forecasting, they cannot establish a good correlation between multiple external variables and predicted load in medium -term load forecasting. To address the above issues, parallel LSTM structures and time series N-node tree LSTM (t-N Tree LSTM) structures are proposed. By introducing branch structures and tree structures to construct finer feature granularity, modeling of medium -term load forecasting is achieved. Finally, experiments are conducted on the 2017 global energy forecasting competition dataset GEFCom2017, and the results show that finer feature granularity is beneficial for obtaining higher accuracy prediction results in the medium-term load forecasting process, verifying the effectiveness of the parallel LSTM model and t-N Tree LSTMs model.

The district cooling system is a special and large-scale category of central air conditioning load with the potential to participate in the power system frequency regulation. Based on the model prediction control method, a control strategy for the district cooling system is proposed to participate in secondary frequency regulation. First, the dynamic model of district cooling system and the dynamic model of thermal power system frequency regulation are established. Then, with the goal of minimizing the deviation between building temperature and human comfort temperature, thermal power unit frequency regulation commands and the frequency deviation, the model predictive controller is designed to optimally allocate the frequency regulation commands of the district cooling system and thermal power units. Finally, simulations were performed on a 10-unit 39-node system containing district cooling system, and the results show that the participation of district cooling system in frequency regulation can improve the system frequency regulation effect and the participation in frequency regulation does not have a large impact on the user comfort level.

In order to solve the constraints of many factors such as ambient temperature, wind speed and solar irradiance on photovoltaic power generation prediction, a long short-term memory （LSTM） neural network model based on similar day analysis and improved whale algorithm optimization to realize short-term prediction of photovoltaic power is proposed. Firstly, the Pearson correlation coefficient is used for feature selection to remove meteorological characteristics that are not correlated with the output power of photovoltaics. Secondly, according to the actual situation that the power generation of photovoltaic power plants is close under similar meteorological conditions, gray relation analysis （GRA） is used to select dates similar to the meteorological characteristics of the forecast day as the training set. Then, an improved whale algorithm（IWOA） is proposed to optimize the hyperparameters of LSTM deep neural network to minimize the root mean square error of the prediction model. Finally, the historical data of photovoltaic power generation of Yulara Desert No.3 photovoltaic power station in Australia is used as experimental data, and the GRA-IWOA-LSTM neural network model is used to make predictions. The simulation results show that the prediction results of the GRA-IWOA-LSTM model are more accurate than the prediction effects of other models under different weather types.

The combined heat and power system effectively integrates electricity and heat energy, providing significant advantages in energy utilization. To analyze the economic and energy-saving effects of combined heat and power dispatch, an optimization dispatch model is developed considering time-of-use electricity prices and load fluctuations, which includes wind farm, energy hub, heat pump, and heat storage tank. The optimization objective is to minimize daily operating costs while considering practical constraints such as steady-state combined electrical and thermal power flow, dynamic characteristics of energy storage device, and pipe temperature drop equations. To enhance optimization efficiency, equation linearization and piecewise McCormick convex envelope technique are utilized, transforming the original non-convex optimization problem into a mixed integer quadratic programming （MIQP） problem. The proposed model's validity is verified through case studies, comparing the impact of wind-to-heat conversion and heat transfer loss on daily operating costs and wind curtailment in four scenarios. The optimal connection node for integrating the wind farm into the heating network was also identified. The results demonstrate the system's capability to convert unstable wind energy into stable heat energy, thereby enhancing wind energy integration capacity and improving overall economic performance.

In view of the unreasonable layout of distributed photovoltaics will bring major impact to the distribution network, a double-layer optimized locating and sizing model of distributed photovoltaic considering the load and PV system output timing is proposed. The upper-layer optimization aims to screen a set of combined data of PV access nodes and installed capacity. The lower-layer optimization takes the network loss, voltage offset and minimum investment cost as the objective function, and feeds back the optimal planning results to the upper optimization layer while solving the high-dimensional and nonlinear power factor optimization problem, so as to determine the optimal access node and installed capacity of distributed photovoltaics. In addition, the adaptive artificial bee colony slime mold algorithm with improved crossover operator is introduced to solve the model, which has excellent global search ability, local development ability and renewal mechanism of the individual, which can obtain more ideal high-quality solutions for such models. The simulation results show that the improved slime mould algorithm not only considers the economy, but also significantly improves the active power loss and power quality of the distribution network compared to other algorithms..

A large number of wind power, photovoltaic and other new energy equipment are connected to the power grid, and the system gradually presents the characteristics of low inertia, which will bring a series of frequency safety problems.Traditional frequency characteristic analysis methods are difficult to accurately grasp the changes in system frequency characteristics after the participation of new energy in inertia support and primary frequency modulation. In order to explore the frequency characteristics and situation changes of the system considering the contribution of wind power-photovoltaic-energy storage to frequency modulation, a new frequency response model is constructed based on the traditional frequency response model and the contribution of wind power-photovoltaic-energy storage to frequency modulation is defined. The frequency characteristic transfer function is used to set the parameters of the system frequency modulation unit. Then the frequency stability index of the power system is quantitatively analyzed, and the influence of the inertia time constant and the control coefficient of each frequency modulation unit on the frequency stability is analyzed from the angle of frequency domain. Finally, the theoretical analysis is verified by MATLAB/Simulink platform. The method of analyzing the frequency characteristics of the new power system from the perspective of time domain and frequency domain can accurately predict the frequency situation of the system considering the contribution degree of wind power-photovoltaic-energy storage frequency modulation, and provide guidance for the safe operation of the new power system frequency considering the inertia support of wind power-photovoltaic-energy storage and primary frequency modulation.

With the high sensitivity and fast detection speed of cavity ring-down spectroscopy (CRDS), CRDS has been applied to the field of online detection of H₂S of characteristic decomposition components of insulating gases, which is of great significance for the fault state diagnosis of insulated electrical equipment. Ring-down time as the key parameter for concentration calculation, how to fit the ring-down curve to achieve the accurate solution of the ring-down time has become a major difficulty in this technical field. The high-precision fitting algorithm of the decay curve is the key factor determining the accuracy of CRDS. STM32F407 models of single-chip microcomputers are selected to establish the decay curve model, and the Levenberg-Marquardt (LM) algorithm is used to fit and combine the attenuation signals collected by H₂S gas at 1578.1nm to build a CRDS experimental platform for verification. Simulations and experiments are conducted based on the LM algorithm, and compared with the least squares method. The simulation results show that the coefficient of determination R² of the microcontroller fitting reaches 0.996, which is highly close to the offline software platform. The experimental results at different gas concentrations show that the fitting accuracies show an increasing trend. The accurate measurement of trace H₂S is achieved by concentration calibration experiment, the correlation coefficient between the actual value of the gas and the measured value is 0.9925, and the absolute errors of the result do not exceed 0.5×10^-6. Compared with the concentration measurement results based on the least squares method, the root mean square values of the concentration data based on the LM algorithm are smaller and the detection accuracies are higher.

Under the background of "Dual Carbon", the large-scale grid connections of new energy distributed power sources are characterized by intermittency and wide range load, which results in a wide dynamic range characteristics of current signals at the new energy grid connection point. Traditional current transformers are prone to exceeding error limits when the primary current is much lower than the rated current due to insufficient excitation of the iron core and low initial magnetic permeability. When the primary current exceeds the rated current, accuracy degradation may occur due to saturation of the iron core, making it difficult to measure power energy accuratly. In order to meet the requirements of high accuracy and wide range for current transformers, a passive zero-flux composite iron core current transformer based on electromotive force compensation is proposed. Through the introduction of unequal secondary windings and compensation windings on the main and auxiliary iron cores and selection of different iron core structure and materials according to the magnetic field distribution characteristics of the main and auxiliary iron cores, the main iron core works in a state of approximately zero magnetic flux. The control block diagram and transfer function of the current transformer with this structure are derived. Combined with simulation analysis and experimental verification, it is shown that the current transformer with this structure meets the accuracy requirements of 0.1 precision current transformer within the range of 0.1% to 200% of the rated primary current. An effective solution is provided for measuring current signals with a wide dynamic range in power systems under the background of new energy.

Aiming at the massive X-DR images generated by crimping quality detection of transmission line strain clamps, an intelligent recognition method for crimping defects is proposed based on YOLO-MS model. A X-DR image dataset including 6 typical crimping defects is constructed using the field crimping quality detection images of strain clamps, and the image preprocessing is carried out by Gaussian filtering, histogram equalization, and gamma correction. The multi-scale (MS) object detection network YOLO-MS is constructed using CSPDarknet, CBAM-PANet, and Head-4. The model is trained and tested using concentrated training and testing samples from a dataset. The results show that the YOLO-MS model can effectively detect 6 types of strain clamp crimping defects, with a mean average precision of 92.57%, and a detection speed of 26 frames per second. It can be used to assist transmission line operation and maintenance personnel to carry out automatic recognition and defect detection of strain clamp crimping images.