To address issues such as nonlinearity and time-variation in the power control of variable-speed pumped storage units,an adaptive power control method based on BP neural networks is proposed.By establishing a simpli-fied mathematical model of the variable-speed pumped storage system,a 3-5-1 feedforward neural network controller with online learning capability is designed.This controller uses the error backpropagation algorithm to adjust network weights in real time.Matlab/Simulink simulation results show that,compared to the traditional PI controller,the BP neural network controller exhibits near-zero overshoot,small steady-state error,and smooth,oscillation-free response during step response.
On the premise of ensuring route safety, an intelligent route planning algorithm based on multi-strategy particle swarm optimization is proposed with the lowest energy consumption as the goal. The energy consumption model is established based on airline meteorological information, and a multi-objective fitness function is constructed with energy consumption, airline smoothness and collision penalty as optimization objectives. An adaptive mean velocity updating strategy is proposed, and nonlinear inertia weight is introduced, and Tent chaotic mapping is adopted to dynamically adjust the development ability and exploration ability of the algorithm to avoid falling into local optimal solution. The strategy of updating the position of the worst particle is put forward, and the particles with the worst fitness are updated by coordinate cross, which improves the overall particle quality of the population and speeds up the efficiency of population optimization. The simulation results show that the improved multi-strategy particle swarm optimization algorithm is superior to PSO algorithm in optimization ability and convergence speed.
To improve the compensation performance of Active Power Filter (APF), this paper proposes a three-level APF controller based on dual DSP, which improves resource allocation. The main DSP is responsible for load current sampling, compensation current sampling, DC side voltage sampling, harmonic and reactive compensation current extraction, current tracking control calculation, PWM wave modulation, and other tasks, while the auxiliary DSP is responsible for sharing some operations and signal sampling, Including harmonic calculation, grid voltage sampling, touch screen display, power outage data storage, etc., communication between the two DSP is achieved through dual port random access memory RAM; More reasonable allocation of software tasks has improved the parallel computing ability of dual DSP. And it has been verified in a three-level APF prototype that it can achieve a switching frequency of 20 kHz, reduce the grid current distortion rate from 15 to 2%, and achieve a harmonic filtering rate of 86.7%, demonstrating good harmonic filtering ability.
The fiber optic temperature measurement host utilizes the principle of Raman scattering of light waves to monitor the temperature of the environment in real-time. Due to the extremely harsh working environment in mines, optical fibers are prone to compression deformation caused by foreign objects (falling coal blocks, mechanical equipment) after long-distance laying, leading to abnormal transmission of light waves and issuing invalid alarm signals. Invalid alarm signals can greatly interfere with the reliability and stability of coal mine safety monitoring systems, causing great difficulties to production operations. On the basis of in-depth analysis of the principle of Raman scattering of optical waves in temperature measurement hosts, this article proposes a segmented filtering model for fiber optic abnormal force signal temperature measurement band based on nonlinear polynomials. Through experimental analysis, it has been proven that the model can prepare to identify abnormal data generated by non environmental temperature effects in the optical path, greatly improving the reliability and effectiveness of the fiber optic temperature measurement host.
When visual and auditory are not available on some occasions,tactile rendering systems can provide an ef-fective way for people to obtain information.A wearable tactile information expression system based on electrotactile rendering is designed and developed.Because this is a kind of passive stimulation,dynamic display method which displayed one electrode at a time according to some specific orders is necessary to improve the effect of perception of the proposed system and this method is applied on an 8x8 spherical electrode array to obtain information through people's wrist.Three experiments including voltage threshold experiment,spatial resolution experiment and simple pat-tern recognition experiment are implemented and the average success rate of recognition is above 90％.These experi-mental results verify the effectiveness of the proposed system and lay a foundation for identifying complicated pat-terns in the future.
In this work, we explore the predefined-time practical consensus problem for integrator-type multi-agent systems by event-driven control. Based on a time base generator (TBG), a new type of event-driven control inputs and event-driven conditions are proposed for each agent, which are more general than the existing results. Under such control scheme, the agents are able to reach practical consensus at a completely predefined time. We provide the predefined-time practical consensus analysis of the closed-loop system by introducing a novel Lyapunov-based sufficient condition for predefined-time stability proposed in this paper. Furthermore, we give the proof of Zeno-freeness for each time instant. Lastly, a numerical example is presented to demonstrate the reliability of the derived results.
Aiming at the anti-interference performance requirements of high-power IGBT drive circuit in active power filter, this paper designs IGBT drive and protection circuit based on M57962AL chip. The voltage change of VCE is detected by two diode series resistors to the chip. Combined with the output characteristic curve of IGBT, the threshold of IGBT short circuit protection current can be set by adjusting the resistance value. The operation experiment of active power filter shows that the design can effectively maintain the stability of the fault output signal, so it has stronger anti-interference performance and improves the reliability of IGBT short-circuit protection.
The energy industry is undergoing unprecedented changes as it pursues global trends towards decarbonization, decentralization, and digitalization. Rapid development and deployment of big data and artificial intelligence technology over the past few decades have transformed the power generation industry in turning into a smarter industry that can monitor and adjust the status of key assets in real time. Digital twins, which are currently in the spotlight, are a technology that reproduce real physical assets using physical and data-driven models to simulate or predict the state of a system. There is a wide range of assets suitable for digital twins in power generation. They aim to identify changes in the system to detect anomalies or make maintenance decisions. The need for digital twins for energy transformation continues to grow. This paper provides a review of digital twin technology specific to the power generation industry. Among the power generation systems, digital twin studies for the combined cycle gas turbine, wind turbine, solar, and nuclear power plant were classified according to the lifecycle, complexity, and type of digital twin model, and the specific features and limitations of each application were analyzed. The goal is to provide readers with a curated summary of use cases which they may find useful in applying to their own work. The paper also explores the challenges and potential future research directions for increasing efficiency, availability, reliability, and solving environmental problems.
Because flexible robots have flexible components such as reducers, there are problems of accuracy deviation and end vibration in the process of external interference and trajectory tracking. This leads to the proposal of a Sliding Mode Control Approach Based on RBF Neural Network (SMC-RBF) parameter optimization. This method is mainly applied to reduce the end vibration and running position error of flexible robot. Firstly, the Newton-Euler method is used to establish the dynamic model of robot considering joint flexibility. At the same time, the experiment optimizes the Sliding Mode Control (SMC) method through RBF neural network. The experiments verify the control methods of the two-joint flexible robot and the six-joint flexible robot respectively. In the control of two-joint robot, the maximum tracking curve error of SMC is only about 0.25 rad under the interference of pulse signal; And the recovery time is only about 1 s. In the control of 6-joint robot, the maximum error of RBF-sliding mode control method on XYZ axis is 0.7 mm, 0.25 mm and 1.25 mm respectively; The error on three axes is smaller than that of traditional PD control method. The results demonstrate that the tracking error of the improved mode control is small, the chattering phenomenon of the robot system is weakened as well.
This article proposes an improved A* algorithm aimed at improving the logistics path quality of automated guided vehicles (AGVs) in digital production workshops, solving the problems of excessive path turns and long transportation time. The traditional A* algorithm is improved internally and externally. In the internal improvement process, we propose an improved node search method within the A* algorithm to avoid generating invalid paths; offer a heuristic function which uses diagonal distance instead of traditional heuristic functions to reduce the number of turns in the path; and add turning weights in the A* algorithm formula, further reducing the number of turns in the path and reducing the number of node searches. In the process of external improvement, the output path of the internally improved A* algorithm is further optimized externally by the improved forward search optimization algorithm and the Bessel curve method, which reduces path length and turns and creates a path with fewer turns and a shorter distance. The experimental results demonstrate that the internally modified A* algorithm suggested in this research performs better when compared to six conventional path planning methods. Based on the internally improved A* algorithm path, the full improved A* algorithm reduces the turning angle by approximately 69% and shortens the path by approximately 10%; based on the simulation results, the improved A* algorithm in this paper can reduce the running time of AGV and improve the logistics efficiency in the workshop. Specifically, the walking time of AGV on the improved A* algorithm path is reduced by 12s compared to the traditional A* algorithm.
为实现水利水电工程的智能化设计,构建了智能设计系统.系统基于建筑信息模型(Building Information Mod-eling,BIM)技术对工程结构进行三维建模,并对所建模型进行有限元分析以验证模型是否符合约束条件,系统内置可随时调用和修改模板资源库,以便对设计工程结构进行优化.将系统投入应用,应用结果显示,系统建立的BIM 模型达到了LOD400精度等级,具有较高的建模精度,所建模型的静力分析结果均符合设计要求.将智能设计系统与其他系统进行对比,对比结果显示,设计系统的响应时间和建模时间比其他系统更短,使用体验较好.可见,基于BIM技术的水利水电工程智能设计系统具有较为优秀的设计性能,设计速度较快,实现了工程设计的智能化、数字化,有利于提高工程设计的效率.
为了获得无像素点密集、边缘细节丢失以及扭曲的高质量矿用摄像仪视频图像,设计了基于残差编解码网络的矿用摄像仪视频图像畸变校正方法.确定包括径向畸变、离心畸变和薄棱镜畸变的矿用摄像仪视频图像畸变类型,以U-Net构架为基本条件,构建由上下采样、编解码以及全卷积层三大模块组成的残差编解码网络,将拍摄获取矿用摄像仪视频图像作为该网络的输入,对网络进行特征提取、非线性映射与反卷积重组操作完成残差编解码网络的训练,最终输出矿用摄像仪畸变细微环节重组视频图像,实现矿用摄像仪视频图像畸变校正;并采用随机梯度算法SGD不断优化残差编解码网络获取最佳输出结果.实验表明:该方法可实现矿用摄像仪视频图像畸变校正,获得最佳质量矿用摄像仪视频图像;且该方法校正后矿用摄像仪视频图像具备较高峰值信噪比.
In order to improve the training efficiency of tennis players,an automatic ball picking robot is proposed to reduce the burden of ball picking.In the ball picking robot,machine vision technology is first introduced to detect and identify tennis balls,and then PID control algorithm and fuzzy theory are introduced to build the intelligent control system of the ball picking robot.The algo-rithm test shows that the loss rate of the machine vision technology adopted by the research is only 0.01％.In addition,the simulation and application test of the ball picking robot show that the robot designed by the research can complete the ball picking work on the tennis court,and the maximum endurance time is 16.7 hours.The above results show that using machine vision technology,PID and fuzzy theory to design the ball picking robot control system is feasible and effective,which is of great significance to the future devel-opment of tennis and robot control technology.
In order to solve the problem of on-line traceability of gasoline vapor recovery detectors in gas stations,the research and development scheme of portable intelligent advanced equipment is put forward,and research is focused on module selection,hardware design,software development and performance verification.The calibration device of the gasoline vapor recovery detector developed by this project can complete the air tightness test,flow calibration and pressure calibration of the gasoline vapor recovery detector,the utility model makes up the shortcomings of the scattered equipment and the complicated operation when the common in-struments are calibrated.The instrument has built-in automatic pressurization device,which can accurately pressurize to the required limit without external air source,and with built-in electronic pressure sensor,it can realize fast air tightness detection,as well as built-in air pumping power source,it can provide stable airflow power for gasoline vapor recovery detectors and realize high efficiency flow calibration with built-in high precision roots flowmeter.The instrument has built-in environmental parameter sensor,which can realize real-time monitoring of temperature,humidity and atmospheric pressure.The device has been tested by the National Academy of Metrology,the metrological performance and uncertainty meet the requirements of national standard GB20952-2020"Emission standard of air pollutant for gasoline filling stations",national metrological calibration code JJF1948-2021"Calibration Specification for Gasoline Vapor Recovery Detectors"etc,on-line calibration of gasoline vapor recovery detectors can be carried out to ensure that the measurement value of oil and gas recovery detector can be traceable,accurate and reliable.
Aiming at the problem of slow and long migration time of virtual machines,a dynamic adaptive migration method of VMware virtualization platform based on cloud computing is proposed.The VMware virtualization platform is built through cloud com-puting.According to the historical dirty page rate and data jitter characteristics,the operating business characteristics and network bandwidth usage are obtained,and the bandwidth coefficient is adjusted adaptively to reasonably allocate network bandwidth;Accord-ing to the size of memory,dirty page speed and bandwidth transmission speed,the constraints of physical resource migration are for-mulated.Entropy coding,dictionary coding and run-length coding are used to complete the compression of dirty page memory and complete dynamic adaptive migration.The experimental results show that the proposed method can make reasonable use of bandwidth resources,reduce downtime,improve the efficiency of virtual machine migration,and ensure the quality of service of running business under the condition of limited physical resources.
In order to solve the problem of load imbalance caused by the incremental migration mechanism of virtual machines,the research of load balancing algorithm is proposed.First,the total amount of resources contained in virtual machines is calculated,the overall computing capacity of server nodes and cluster systems is calculated by using load entropy,and the load status is set uni-formly by proportion analysis,the number of incremental migration is obtained,and the load balancing task allocation model is con-structed,Normalize the resource utilization rate of incremental migration jobs,use standard deviation to complete the load calculation,and obtain the average value of the migration job set.While reducing the queue waiting time of virtual machines,you can obtain the waiting time of a single task.Set the load partition of resources under the incremental migration status,which can effectively achieve the load balancing design and calculation.The experimental results show that the proposed method can complete load balancing well,and the waiting time of the algorithm is long,and the balancing method is effective.
The weld obtained by laser welding is narrow after welding,which is easy to cause welding defects.In order to im-prove the accuracy of initial guidance and tracking of narrow gap butt weld in laser welding of non-standard products,the initial point identification method based on edge template matching was studied.The image of weld feature is processed by the method of dynamic ROI region extraction.Finally,a 3D laser welding seam tracking system based on active vision is designed.The experiment shows that the average error between the theoretical coordinates calculated by the research system and the actual coordinates is 0.027 0,which indicates that the system has a high calibration accuracy for weld points.The variation range of weld deviation of the research method is basically maintained within±0.25 mm,which is significantly lower than other advanced weld tracking models.And meet the requirements of weld tracking accuracy,and the system runs stably.This research system improves the positioning accuracy of the robot in the welding process,and can meet its requirements for real-time performance.
In order to further enhance the signal processing capability of sensors,an error compensation method based on FPGA and genetic algorithm is proposed,taking the signal processing of time grating sensors as the research object.Among them,FPGA is used as the main control chip to construct a control system for sensor signal processing,laying the foundation for subsequent signal measurement;At the same time,in order to improve the accuracy of signal measurement,genetic algorithm is used to identify the pa-rameters of the time grating sensor,and finally,error compensation is carried out based on the identification results.The results show that compared with traditional signal processing methods,the adaptive signal processing method designed in this study can achieve better signal processing results,resulting in more concentrated and stable signals;After error compensation,the system error is smal-ler and the accuracy is higher.In summary,it can be seen that the sensor signal processing method designed in this study can effec-tively improve the signal processing level of sensors,and is suitable for signal processing of various sensors,with certain practical ap-plication value.
To improve the recognition and tracking performance of traffic vehicles,an improved SSD-MobileNetV3 network and SORT algorithm recognition and tracking method is proposed.Among them,in the automatic vehicle detection and recognition part,an improved SSD-MobileNetV3 network is adopted;Then,an improved SORT algorithm is used to achieve automatic vehicle track-ing.The results show that compared with other object detection and recognition methods,the average accuracy and precision of the improved SSD-MobileNetV3 network in vehicle detection and recognition are 90.4％and 94.1％,respectively;Compared with other target tracking methods,the improved SORT vehicle tracking algorithm is efficient and stable,and the multi target tracking accuracy reaches 94.7％.In summary,the vehicle automatic recognition and tracking algorithm proposed in this study is feasible and can be used in the supervision of traffic vehicles.
At present,the world's cultural exchanges are becoming more and more frequent,and machine translation technology is indispensable.Aiming at the problems existing in traditional machine translation,this paper improves the machine translation model and proposes a combined translation model.This model introduces attention mechanism and graph convolution neural network to opti-mize the translation of graph information.The model also considers the factors such as word information location and sentence transla-tion length,and uses IWSLT data set and WMT14en-de data set to verify the model performance.The BLEU scores of the proposed model in the two data sets are 44.9 and 47.3;Translation accuracy reaches 0.903 and 0.941;When the sentence length is between 10 and 20 words,the BLEU score of the model is 44.67.The experimental results show that the proposed model has better translation performance and wider translation scope,which provides a new research idea for image information translation.
