In order to meet the requirements of electronic experimental teaching for oscilloscope function and performance，the wireless oscillograph is designed，its parameters can be set flexibly，and its application is not restricted by space． The system uses FPGA( Field Programmable Gata Array) as the main controller for signal processing，and the hardware is mainly composed of signal conditioning module，A/D ( Analog-to-Digital) conversion module，wireless transmission module and power module． The host interface is designed based on Labview visual environment． The parameters and signal transmission between data acquisition system and software client is realized by WIFI technology． And the observation effect is not limited by the space position of the measured point． The system can collect and observe 4 signals simultaneously． After a series of tests，we confirm that the waveform shape of the signal with highest frequency 700 kHz can be distinguished，and the frequency and amplitude of the signal with highest frequency of 2 MHz can be accurately tested． The designed instrument and the laboratory traditional oscilloscope complement each other，and can meet the equipment requirements for undergraduate experimental teaching and extracurricular practice．

In order to solve the problem that the battery car capacitance is damaged because charging voltage of the battery charging systems used in the market is not matched with the model of the battery car capacitance，a new battery charger is proposed． The battery charging system uses NCP1654 as the core control chip，and the Boost Chopper is used as the main circuit． The new silicon carbide semiconductor device is used as the main switching device to convert alternating current into direct current． When the AC voltage on the power supply changes from 180 V to 260 V，the device can operate normally． The intelligent battery car charging system outputs two DC voltages，42 V and 27 V respectively，and the maximum output current are both 2 A． The load regulation is 0． 1，and the peak-to-peak value of output noise ripple voltage is less than 1． 5 V． The protection circuit in the charging device prevents spike voltage and surge current during start － up，and prevents over charge problem． Considering the user experience of the intelligent battery charging system，the touch screen is selected as the operation interface． The STM32 demoboard is used to detect the output voltage and output current，control the output voltage amplitude of the charging system． The STM32 can also detect the output voltage and current and control the amplitude of the output voltage． After testing，the battery charging system reached the design requirements．

In order to improve the optical absorption of active layer in polymer solar cells，a PSC ( Polymer Solar
Cell) with multiple optical spacers is proposed． By adjusting the refractive index distribution of multiple optical
spacers to change the distribution of incident light in the active layer，enhance the absorption of the active layer
and improve the performance of the device． Further，we use the transfer matrix method to calculate the device and
explore the influence of a PSC with multiple optical spacers consisting of undoped ZnO film and CZO ( Cs doped
Zinc Oxide) films with different doping concentration of 0． 002 5 mol /L，0． 005 mol /L and 0． 01 mol /L to the
distribution of light field and short circuit current． The simulation results show that the multiple optical spacers
provide a cascade of refractive index between the ITO ( Indium Tin Oxide) substrate and the active layer when the
CZO films with a decreasing doping concentration is deposited on ITO successively，resulting in an increase of light
absorption in active layer and Jsc significantly．

In order to solve the control problem of refrigeration system with time-delay characteristics，an improved model-free adaptive predictive control algorithm is proposed． The improved model-free adaptive predictive algorithm is applied to the refrigeration system control scheme． And in order to get more stable control effects，better performance of the algorithm，and make design of MFAPC ( Model-Free Adaptive Predictive Control) more flexible． Firstly，a parameter estimation control scheme that introduced two parameters L1 and L2 is proposed，It improves the pseudo-partial derivative parameter estimation process of the MFAPC，therefore the algorithm can improve the stability effect while achieving the goal of reducing reference． Then，The control output error sum is added to the control input criterion function． Finally，typical linear and nonlinear time-delay system，tracking time-varying signal system，lithium bromide refrigeration system are introduced． The tracking control problems are simulated and compared． The simulation results show that the improved model-free predictive control algorithm can achieve more stable output result，with faster response speed and better control performance，solving the refrigeration system control problem of time-delay characteristics．

In order to improve the system vibration and instability caused by the delay and perturbation phenomena in actual Liurie control system，based on the theory of robust stability theories，Lurie systems with time-varying delay uncertainty are discussed． Combined with lemma and cross terms defined method，the new stability criterions of time-delay dependent and time-delay independent are deduced in two sector regions:［0，V］and ［V1，V2］by constructing a new Lyapunov functional． The feasibility and superiority are illustrated by the numerical example．

In automatic question and answering system，traditional way of the answer extraction depends on participle of answer fragment and the accuracy of semantic comprehension from the context，which consumes manpower and time during the extraction process． To solve the above problems，the approach of step answer extraction is adopted where the final answer extraction is conducted through the extracted answers from the sentences． The model of answer sentence extraction is built in combination of Bi-LSTM ( Bi-directional Long Short-Term Memory) and Max Pooling during the extraction process． The experimental results show that the MＲＲ ( Mean Average Precision) index of this model is close to 0． 75 in the extraction of the answer sentence．

In order to overcome the shortcomings of particle swarms in solving complex problems of multimodal functions，such as the convergence speed is slow and it is easy to fall into local optimal values． A multimodal delayed particle swarm optimization algorithm based on Gaussian learning is designed． Firstly，the improved Gaussian learning is introduced to improve the convergence speed of the algorithm． Then for the four evolution states，a delay factor is introduced into the algorithm to avoid the local optimal problem． Simulation experiments are performed by six unimodal multimodal test functions． It is verified that GLPSO ( Gaussian Learning PSO) has better convergence speed when dealing with six functions． And it is verified that the GLMDPSO ( Gaussian Learning Multimodal Delayed PSO ) algorithm has better global search ability when dealing with complex problems of multimodal functions． The improved algorithm solves the multimodal function optimization problem，the algorithm could effectively jump out of the stagnation state，improve the convergence speed and have better optimization ability．

In order to solve the problem that the ICP ( Iterative Closest Point) algorithm has strict requirements on the initial point cloud position and is easy to be trapped in local optimum，a new registration method is proposed． Firstly，based on the idea of complementary advantages，the artificial glowworm algorithm and particle swarm algorithm are combined to propose an AAGPSO ( Adaptive Artificial Glowworm-Particle Swarm Optimization) ，accelerating the convergence speed of the algorithm and improving the accuracy of the solution．Secondly，for a different initial location of the point cloud，the improved AAGPSO algorithm is introduced into the ICP registration algorithm making ICP algorithm optimized，and solving the local optimum problem of the ICP
algorithm． The improved algorithm accelerates the overall registration efficiency． Finally，experimental data are used to compare the original ICP registration method with the improved registration method，and the error analysis is carried out． The AAGPSO algorithm improves the registration accuracy，accelerates the algorithm convergence speed．

Considering that the difficulty of free-hand sketch recognition，the low recognition accuracy and handcrafted features，we propose a new CNN ( Convolutional Neural Networks) structure DCSN ( Deeper-CNNSketch-Net) which is specifically designed to accommodate the unique characteristics of free-hand sketch． Firstly we use a larger convolution kernel in the first layer to obtain the structural information of the sketch． Then we use smaller stride in the first layer to keep the feature information． At last，we increase the network layers to deepen the network depth． In order to improve the recognition accuracy，we propose two new data enhancement methodsmall graphics reduction strategy and tail removal strategy to increase the diversity of data sets，then we use the extended data sets to train the DCSN network． The experimental results show that our model can achieve 70． 5% recognition accuracy on the largest free-hand image dataset，which has a good performance than the existing freehand sketch recognition methods．

In order to analyze the key factors affecting the survival of P2P ( Peer to Peer) platform for the phenomenon of“life is short”of the P2P online lending platform，we collected the relevant variable data． First，we apply the bagging method and random forest to classify the P2P platform dataset and extract the important variables that affect the classification results． Secondly，the Cox proportional hazard model and the accelerated death model in the survival analysis are used to conduct empirical research，and continue to explore the P2P online loan platform“survival”or the key factors of“death”． It indicates that the most important influencing factor is whether the user funds are deposited in the bank． This result provides reference for investors and regulators，and provides a basis for accelerating the construction of the management and restraint system for P2P online loan platform funds．

The traditional K-means algorithm is very sensitive to the selection of cluster initial points and the calculation of distance metrics，it is possible that the K-means algorithm can converge to local optimal solutions．Aiming at this problem，an improved K-means algorithm，namely K-means clustering algorithm optimal matching algorithm． The improved algorithm firstly selects the initial point of the traditional K-means clustering algorithm and analyzes the clustering result． Then，it conducts experimental tests from the selection of the initial clustering center and the determination of the distance algorithm respectively，and introduces the contour coefficient evaluation clustering． The experimental results prove that the optimal matching algorithm of K-means clustering algorithm has better stability and higher clustering accuracy．

In order to solve the problems of effectiveness and safety of electromagnetic stimulation in the treatment of epileptic neurological diseases，the equal effect model of pyramidal neurons in hippocampal ca3 region，which is prone to epileptic seizures，was established． The firing activity of the neurons was studied through Matlab simulation under the conditions of both direct-current and alternating-current electric field． The results are as follows: first，neuron firing frequency is modulated by direct-current electric field，and greatly sensitive to the direction of the electric field． The dynamic characteristics of neurons were affected significantly by the change of the electric field intensity when the direction of the field was negative; the effects were not significant when the direction of field was positive． Second，the dynamic characteristics of neurons showed great dependence on the external field frequency by alternating-current electric field stimulation． The neurons firing may appear discharging status such as period bursting，synchronized spiking，period spiking，and chaotic firing by different stimulation frequency． It has great theoretical significance to treat the neurological disorders such as Epilepsy by means of electromagnetic field therapy，by which the discharging status of neurons can be affected better by alternating-current electric field than that by direct-current electric field． It will be helpful for doctors to make secure and efficient therapy schedules to treat neurological disorders with the method of electromagnetic fieldtherapy．

Because most university libraries still use traditional lending system ssuch as readers' card and all in one-card，an intelligent face recognition lending system based on cloud computing is proposed． Compared with the face recognition system that finishes by computing in local devices，which has the shortcomings of slow operation and limited storage resources，this system adopts a cloud platform with high performance computing power and massive storage capacity． Uploads the collected face information to the cloud platform by collecting face data，and the cloud platform uses binarization，light compensation，gray change，contrast enhancement and other methods to save and back up the face features information． When scanning the face is tested，the face
information is uploaded to the cloud and compared with the backup information for identification． The test results show that the intelligent face recognition lending system based on cloud computing has better accuracy，real-time and stability，which reduces the burden and cost of local storage，and it can better meet a need of user's intelligent lending．

Due to the poor performance of the traditional course database retrieval system and the influence of noise，the retrieval accuracy is low，which can not meet the user's demand for Stac ( Statistical Analysis) course database retrieval． To this end，the design of the automatic retrieval system of Stac course database based on scene theory is proposed． Under the scene theory，the design of the database automatic retrieval system adding the word segmentation module，uses the combined ambiguity statistical method to distinguish synonymous or polysemous words in the Stac course database; and uses the web spider to find the web link address，reading the content，and performing all the goals address retrieval． When the collection volume reaches a certain scale，several independent search engines are called and cooperated with each other to establish an index library，collect data according to the Stac curriculum resource data specification standard，and use the index engine to input all the collection results into the system． By identifying the characteristics of the scene，the CD-ＲOM database is created，the retrieval process is designed，the behavior of each machine is closely monitored，and noise interference is avoided． After processing by the central DB ( Data Base) Server，the address lists are merged to form a new resource list for the user to retrieve． Experimental results show that the retrieval accuracy of the system can reach up to 98%，providing systematic support for multi-image retrieval．