We proposed an improved echo state neural network model for the analysis and prediction of long-term behavior of complex systems. The model introduced the delayed feedback of hidden layer state to reflect the influence of the past time information on the current state of the system,  avoiding the shortcomings of weak memory ability and difficulty of obtaining optimal parameters in traditional echo state network methods.

We considered the persistence problem of homeomorphism on noncompact metric spaces. By using the definitions of persistence, equicontinuity, strongly topological stability, and persistent shadowing property of homeomorphisms, we prove that homeomorphisms that are equicontinuity and topologically stable are persistent, homeomorphisms have persistent shadowing properties if and only if they are persistent and have pseudoorbital shadowing properties, and an expansive homeomorphism with persistent shadowing property is strongly topologically stable.

We considered a class of non-autonomous second-order evolution equations with delay. Firstly, we obtained the existence and uniqueness of solution by using Faedo-Galerkin approximation method in C_Ht. Secondly, by means of operator decomposition, the D_CHt-pullback asymptotic compactness of the process {U(t,τ)}_t≥τ was verified, which proved the existence of time-dependent pullback attractor for evolution equations with delay.

Firstly, we considered the identification problem of the dissipation coefficients for a class of degenerate elliptic equations. By treating the unknown dissipation coefficients as control functions, treating the solutions of the equation as state variables, and defining the objective  functional as the sum of the error between the state and the measurement values  and the artificial regularization term, we transformed the coefficient identification problem into an optimal control problem. Secondly,  the coefficient identification problem was studied  by using the research method of the optimal control problem. We gave the expression of the optimal control and proved the uniqueness of optimal control under appropriate conditions.

Based on the theory of variable exponent Fourier-Besov function spaces, we used Littlewood-Paley decomposition tools, Fourier localization methods and Banach contraction mapping principle. By establishing estimations for both linear and nonlinear terms, we proved the global well-posedness and the Gevrey class regularity of the solutions to the fractional Boussinesq-Coriolis equations in critical variable exponent Fourier-Besov spaces.

Firstly, by using differential inequality techniques, we gave  a prior estimate of the L⁴ norm  and solution of temperature for bidirectional flow media under the Newtonian cooling boundary conditions. Secondly, by using a prior estimate of the solution and setting an appropriate energy function, we proved that the solutions decayed algebraically with spatial variable in a semi-infinite pipe.

We considered the dynamic bifurcation  problem of a class of cross-reaction-diffusion models with Holling-Ⅱ functional response function under non-homogeneous Dirichlet boundary conditions. Firstly, the critical crossing conditions for the corresponding linearization problem eigenvalues were obtained by using the spectral analysis theory. Secondly,  the environmental carrying coefficient was selected as the bifurcation parameter, the analytical expression of the dynamic transition type and bifurcation solution of the system was obtained by using the center manifold reduction and the dynamic bifurcation theory. Finally, by using the finite difference method, the pattern change patterns of the system were given under  different parameter conditions.

By using operator semigroup theory and upper and lower solution monotone iterative methods, we discuss the existence of mild solutions to  initial value problems for a class of Conformable fractional evolution equations  with Volterra-type integral operators in Banach spaces, where T_α represents the  Conformable fractional derivative operator with order 0<α<1, A is a coherently closed linear operator. Under the condition that the nonlinear term satisfies the appropriate inequality, the existence of the mild solution to the equation is obtained.

Firstly, we introduce the notion of Gorenstein strongly FP-injective modules by means of acyclic complexes of injective modules and the theory of  Hom functors. Secondly, we study homological properties of Gorenstein strongly FP-injective modules  by using the Horseshoe Lemma and the method of constructing pull-back diagrams, and prove that the class GSFI of Gorenstein strongly FP-injective modules is injectively resolving, with respect to closed under arbitrary direct products and direct summands, and if the Gorenstein strongly FP-injective dimension is finite for every R-module, then (^⊥GSFI, GSFI) forms a complete hereditary cotorsion pair.

We considered  the construction problem of non-direct product triangular norms on product lattices, and gave a class of non-direct product triangular norms on product lattices,  thus solving  an open problem of whether there were other forms of non-direct product triangular norms on product lattices.

We considered the model averaging problem of the proportional hazard model in  the right-censored data with fragmentary covariates. We first used the maximum likelihood estimation method  to estimate the parameters in the model, and then used the model averaging method based on the information criterion  to select the weights. The simulation results show that the model averaging method has higher prediction accuracy than the model selection method, and  the superiority and feasibility of the proposed method are verified by the analysis of breast cancer examples.

For the high-dimensional quantile regression model with massive data, firstly, a subsampling algorithm based on the decorrelation score function was constructed to estimate the low-dimensional parameters of interest. Secondly, we derived the limit distribution of the proposed estimates and calculated the subsampling probability under the L-optimal criterion according to the asymptotic covariance matrix, giving an efficient two-step algorithm. The simulation and empirical analysis results show that the  optimal subsampling method is significantly superior to  the uniform subsampling method.

We considered the inverse problem of Cauchy problem of two dimensional parabolic equations, which was seriously ill-posed. Firstly, a regular approximate solution of the problem was obtained by using Landweber iterative regularization method, and  Fourier transform was used to obtain  the exact solution of the problem. Secondly, the Holder type error estimation between the exact solution and the regular solution was given under the selection rules of the posterior regularization parameters, and stronger prior conditions were used to give  the error estimation at the end point x=1. Finally, numerical examples were given to demonstrate the effectiveness of the proposed method. The results show that the proposed method has a faster  convergence rate than existing methods.

We proposed a semi-parametric accelerated hazard regression model with measurement errors based on the current status data. Firstly, the unknown baseline cumulative hazard function was approximated by using I-spline, and parameter estimates of the model were obtained based on Sieve maximum likelihood estimation method. Secondly, a simulation extrapolation method was used to correct  estimation error caused by  measurement errors in covariates. Thirdly, the numerical simulations were carried out to verify the effectiveness of the proposed method as well as the impact of ignoring measurement error in covariates. Finally,  the proposed method was applied to study cardiovascular and cerebrovascular disease mortality, we obtained estimation of hazard function for  cardiovascular and cerebrovascular disease mortality. The experimental results show that the proposed method is effective.

By improving the Swin Transformer coding and decoding network,  combined with the skip-linking and depth supervision mechanisms, we proposd a new image region segmentation method  of  neonatal brain based on self-attention mechanism of shifted windows to  address the issues of low signal-to-noise ratio and poor tissue contrast in segmentation of nuclear magnetic resonance imaging (MRI) images of the neonatal brain. The method could achieve accurate segmentation of multifunctional regions of the neonatal brain images after preprocessing the MRI images, and further improve the segmentation accuracy by using the maximum connected domain algorithm. The experimental results on the dHCP dataset show that the method is superior to existing methods, providing potential possibilities for early detection and intervention of neonatal brain injury.

Aiming at the problem that there were non-uniform scattering media  (such as atmospheric turbulence, smoke, haze, etc.) incomplex images， which led to different propagation and scattering characteristics of light in different regions, making it difficult to accurately restore the visibility of the image, we proposed a defogging method of  complex image based on  high and low frequency feature enhancement and transmittance correction. Firstly, we designed low frequency feature enhancement methods 
based on singular value decomposition and Gamma inflection point correction. Secondly, based on Shearlet transformation decomposition and nonlinear transformation, we obtained a high frequency feature enhancement method. Thirdly, we used soft cutout to refine the estimated transmittance and constructed a transmittance correction strategy. Finally, by integrating the above three methods, based on atmospheric light values and refined transmittance, image dehazing was completed. After enhancing high and low frequency features respectively, we superimposed  the two  to obtain an enhanced dehazing image. The visual perception and objective evaluation indicators  of dehazing images have been  verified that the proposed method has good dehazing effect and  can effectively restore the detailed information of complex images,  improving the overall visual quality of the images.

Aiming at the shortcomings of existing methods in the image segmentation performance of small and medium-sized organs in the abdomen, we proposed  a network model based on local and global parallel coding  for multi-organ image segmentation in the abdomen. Firstly, a local coding branch was designed to extract multi-scale feature information. Secondly, the global feature coding branch adopted the  block Transformer, which not only captured the global long distance dependency information but also reduced the computation amount through the combination of intra-block Transformer and inter-block Transformer. Thirdly, a feature fusion module was designed to fuse the context information from two coding branches. Finally, the decoding module was designed to realize the interaction between global information and local context information, so as to better compensate for the information 
loss in the decoding stage. Experiments were conducted on the Synapse multi-organ CT dataset, compared with the current nine advanced methods, the average Dice similarity  coefficient  (DSC) and Hausdorff distance (HD) indicators achieve the best performance, with 83.10% and 17.80 mm, respectively.

Aiming at  the problem of sparse entity relationship type data with unclear feature information when using global pointer networks for entity relationship extraction, as well as the problem of class imbalance and incorrect labeling in the data, we proposed a entity-relation joint extraction model based on  contrastive learning and gradient penalty methods while utilizing an enhanced RoBERTa pre-trained model. Experimental results on the Alibaba Tianchi Chinese medical information processing benchmark CBLUE2.0 dataset show  that this model outperforms the global pointer network, and can more  effectively extract  entity relationship from complex data.

Aiming at the problem that the traditional flower pollination algorithm (FPA) was significantly affected by initial parameters and prone to local optima or convergence failures, we proposed  a quantum-inspired flower pollination algorithm (QFPA). By incorporating quantum systems into the FPA,  the  pollination search process was made more efficient, thereby improving global search capabilities. Additionally, trajectory analysis was employed to better enable the population to escape from local optima and further reduce errors. In order to verify  the effectiveness of the method, firstly, the  QFPA was evaluated using selected benchmark functions. Secondly,  the best evaluated  QFPA was used  to optimize the hyperparameters of the long short-term memory network (LSTM) model. Finally, the experiments were conducted on an air quality dataset after removing noise  using the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) algorithm, and compared with several other commonly used optimization algorithms. The experimental results show that QFPA  enhances the global search capability and convergence properties of optimization algorithms. The QFPA-LSTM model improves the accuracy and efficiency of long-term time series predictions, with a root mean square error of 10.93 μg/m³, thus providing a reliable solution for air quality prediction in practical applications.

Aiming at  the non-selective expansion  and training deficiencies of the DoubleMix algorithm during data augmentation, we proposed a supervised contrastive learning text classification model based on double-layer data augmentation, which effectively improved the accuracy of text classification when training data was scarce. Firstly, keyword-based data augmentation was applied to the original data at the input layer, while selectively enhancing the data without considering sentence structure. Secondly, we  interpolated  the original and augmented data in the BERT hidden layers, and  then send them to the TextCNN for further feature extraction. Finally, the model was trained by using Wasserstein distance and double contrastive loss to enhance text classification accuracy. The comparative experimental results on SST-2, CR, TREC, and PC datasets show that the classification accuracy of the proposed method is 93.41%, 93.55%, 97.61%, and 95.27% respectively, which is superior to classical algorithms.

Aiming at the  problems that ignored by the existing embedded methods: the influence of the latent representation of instance
 correlation on pseudo-label learning, and the calculation error was caused by the fixed graph matrix, which increased with the deepening of iterations. We proposed a multi-label feature selection method with latent representation and dynamic graph constraints. Firstly, the proposed method used the latent representation of instance correlation to construct the pseudo-label matrix, and combined it with linear mapping and minimizing the Friedman norm distance between the pseudo-label and the ground-truth label to  ensure a high similarity between pseudo-labels and the ground-truth labels. Secondly, the dynamic graph was constructed by using the low-dimensional manifold structure of pseudo-labels to alleviate the problem of increasing calculation error with iteration depth caused by a fixed graph matrix.  The comparative experimental results with seven advance methods on 12 datasets show that the overall classification performance of the proposed method is superior to  the existing advanced methods, and it  can better deal with multi-label feature selection problems.

Based on the intelligent reflecting surface-assisted multi input single output (MISO) secure wireless communication system, we proposed a joint active-passive beamforming algorithm with the objective of maximizing the system secrecy rate. We considered the joint optimization design problem of  beamforming vectors for  base station transmission and passive intelligent reflecting surface (IRS) phase shift matrix under non-ideal channel state information. In order to solve the non-convex fractional planning problem, two auxiliary variables were introduced through Charnes-Cooper transformation to transform the single fractional problem into a difference form, at the same time,  an alternating iterative optimization combined with  semidefinite relaxation (SDR) method was adopted to obtain an easy-to-solve convex problem. The simulation experiment results show that  compared with the traditional algorithms, the proposed algorithm effectively improves the system security, the confidentiality performance by 10%—30%, and the secrecy rate does not decrease significantly under certain channel state information error, which has strong robustness.

Aiming at the problem that  existing methods in  human-computer interaction systems could not  accurately capture the user’s operational intentions and had poor adaptability to dynamic environments, resulting in poor  accuracy of human-computer interaction. In order to improve the accuracy of human-computer interaction in the operation process of collaborative robots, we proposed a multi-point  touch based approximate optimal human-computer interaction control method. Firstly, based on human-computer interaction for multi-point touch action matching, we established an image conduction function for interactive gesture action sequences, extracted interactive gesture features, analyzed image similarity feature components, and obtained the fuzziness set of action judgments based on pixel values to achieve matching of multi-point touch actions and accurately capture user operation intentions. Secondly, considering the motion conditions and friction factors of the robot, we established an approximate optimal constraint equation for friction to ensure the balance and stability of the robot’s interaction and movement. Finally, we obtained the expected response of the interactive arm, described the human-computer interaction state under multi-point touch conditions through Lagrange equation, established the interaction action dynamics equation, introduced interaction control variables, and used adaptive fuzzy control system to output approximate optimal control results to improve dynamic environment adaptability. We also adjusted control strategies according to actual situations to better meet the needs of human-computer interaction. Experimental results  show that the proposed method can effectively achieve human-computer interaction control, with recognition rates of over 94%, and a small delay difference of 0.03 ×10^-3 s during control,  with fast iteration convergence speed and better control effect.

Aiming at  the problems that wireless fidelity (WiFi) and bluetooth low energy (BLE) fingerprint localization methods required a large number of labeled training samples and that the accuracy and stability of single-mode localization were difficult to  meet the requirements of large-scale localization scenarios, we proposed a semi-supervised manifold constraint localization method that fused WiFi and  BLE signals. The experimental results show that compared with a single feature, the normalized variance of each dimension of the proposed  method is stable below 0.08, and the accuracy of localization is improved by about 25 percentage points.  When the semi-supervised learning method is used to construct manifold constraints separately, the number of labeled samples required in the localization process can be reduced by about 90%. Therefore,  this method can greatly reduce the  number of required label samples, and effectively improve the stability and accuracy of localization.

Aiming at the problem of short network life cycle due to low energy efficiency in wireless sensor networks, we proposed a novel improved Harris hawk optimization algorithm based clustering protocols for wireless sensor networks (IHHOC). IHHOC adopted the improved Harris hawk optimization algorithm to obtain the optimal cluster head set. Firstly, the population was initialized by the Sobol sequence and the fitness function was defined by considering the three parameters of residual energy, the distance to the base station, and the density of nodes, and the optimal solution was finally obtained by iterating through the exploration, transition, and exploitation one after another. Secondly, Gaussian stochastic wandering strategy was used to avoid IHHOC falling into local optimum. After clustering, the optimal forwarding nodes were found in the neighboring clusters of the cluster head based on the residual energy, distance from the cluster head and base station to further reduce the network energy consumption. The simulation experiment results show that IHHOC can effectively improve the network energy efficiency, increase the network throughput, and extend the network life cycle.

Based on the sliding mode method of nonlinear chaotic systems, the sliding mode control and synchronization of Sprott-D uncertain fractional-order systems were studied according to the fractional-order stability theory and synchronous control method, and the results were verified by using MATLAB simulation program. The results show that the Sprott-D uncertain fractional-order chaotic systems corresponding to the master-slave systems can achieve adaptive sliding mode synchronization under certain assumptions.

We proposed a stimulated Brillouin scattering (SBS) optoelectronic oscillator (OEO) based on a self-polarization-stabilization dual-loop structure.By utilizing the narrow bandwidth gain spectrum of SBS to achieve the conversion from phase modulation to intensity modulation (PM-IM), and selecting the oscillation mode of the OEO. A polarization self-polarization-stabilization dual-loop structure was constructed. After the input light in each loop was reflected by a 45° Faraday rotator mirror. It would  return to its path through a 45° Faraday rotator and different lengths of single-mode fiber, ensuring that the polarization state of the output signal and the input signal always differed by 180°, thereby eliminating external mechanical vibrations and temperature disturbances. Since the loop structure was reflective and bidirectional, the required fiber length was reduced by half. The experimental results show that the OEO can achieve frequency tuning by changing the pump wavelength, and  can generate 
microwave signals of 1—16 GHz, with a side mode suppression ratio (SMSR) of 67.14 dB at 10 GHz, and a phase noise of -116.3 dBc/Hz@10 kHz.

Aiming at the problems of low sensitivity, poor signal-to-noise ratio, complex installation, and the need for additional power supply of the traditional power grid dance detector in the face of electromagnetic interference and continuous power supply, we proposed a new double-ended Mach-Zehnder (M-Z) fibre optic seismic detection system. The system adopted double-ended fibre optic M-Z optical path structure, which could achieve long-distance, large-scale, multi-dimensional on-line monitoring and structural safety warning to locate the position of the dance in real time.  The geophone mechanical unit of this system adopted the push-pull structure of silicone elastomer and brass mass block, and after parametric modelling and finite element analysis by ANSYS software, its intrinsic frequency was determined to be 106.77 Hz, with the relative error of only 1.65% compared to the experimental data. The experimental results show that the relative sensitivity of the detector in the range of 0—140 Hz is 41.25—66.78 dB(rad/Pa), which confirms its ability to detect the high sensitivity of the dance signal.

At the theoretical level of M06-2X/SMD/6-311+G(d,p), we studied  the reaction mechanism between Edaravone (Eda) keto isomers and superoxide hydrogen radical ·HO₂  in aqueous liquid phase at 1 atmospheric pressure and 310.15 K temperature. The results show  that there are three processes in the reaction of Eda keto isomers with ·HO₂:  H extraction,  addition and single electron transfer.  The H extraction reaction is mainly achieved through ·HO₂ extraction of heterocyclic H and methyl H,  and the free energy barrier of the reaction is  77.1—78.7 kJ/mol. The addition reaction can be realized by the process of ·HO₂ addition to unsaturated C,  and the free energy barrier of addition is 48.2—95.0 kJ/mol. The most advantageous exothermic reaction is the addition of C atoms connected to methyl groups on heterocycles,  with a free energy barrier of 48.2 kJ/mol. The free energy barrier of single electron transfer is 141.1 kJ/mol,  which is impossible.Therefore,  the  Eda  keto isomer in aqueous liquid phase can eliminate ·HO₂ by H extraction and addition reactions.

Online and offline software were used to predict,  screen and assist in the calculation of the B-cell and T-cell epitopes of the allergen Met e1 from Metapenaeus ensis. Firstly, the amino acid sequence of Met e1 protein was retrieved from the uniprot protein database. Secondly, the physicochemical properties,  signal peptides,  and transmembrane regions of Met e1 were analyzed  by using online tools ExPASY ProtParam,  SignalP-5.0 Server,  and TMHMM Server v.2.0. The secondary structure of Met e1 was jointly predicted and  calculated by using the PSIPRED online tool,  SOPMA online tool,  and DNAstar software,  while the tertiary structure of  Met e1 was predicted and calculated by using Swiss model online software. Thirdly, the linear B-cell epitopes were comprehensively predicted and calculated by using DNAStar offline software and IEDB online software,  CD4+T and CD8+T-cell epitopes were predicted and calculated by using the IEDB online software.  Finally, the obtained results were  screened for B-cell and T-cell epitopes. The results show  that the B-cell epitopes of Met e1 protein are located at amino acids 15—28， 45—49， 92—95， 125—130， 150—153 and 255—258,  and T-cell epitopes are located at amino acids 78—84 and 223—232.

Taking the quasi-protected area of Jiangbei water source  area in Jiamusi City， Heilongjiang Province as the research  area, we selected  seven typical land use types  to study the distribution characteristics of nitrogen and phosphorus in groundwater and aeration zone under  different land use types. The results show that the NH⁺₄ content in the irrigated land is significantly lower (p<0.05, same below) than that in other land use types within the depth range of 0—180 cm in the aeration zone. The distribution characteristics of dissolved nitrogen and phosphorus in residential area aeration zone are the most different from those in other land use types. The contents of NO^-₃ (0—150 cm),  NO^-₂ (0—120 cm),  and DP (0—90 cm) in the residential area aeration zone are significantly higher than those in other land use types,  and their contents decrease with the increase of soil depth. For all land use types of groundwater, NO^-₃ is the main component of TN,  and the contents of NO^-₃ and TN in the groundwater of residential areas and irrigated land are significantly higher than those in other land use types.  Agricultural production,  manure and wastewater are the main sources of NO^-₃ in groundwater at residential areas. The results of Pearson correlation analysis show that the content of NO^-₂ in groundwater is significantly   positively correlated with the contents of NH⁺₄,NO^-₂,NO^-₃,  and conductivity  in the soil at 0—30 cm depth. There is a significant   positive correlation between NH⁺₄ content and soil pH  value at 150—180 cm depth. TN  is significantly   positively correlated with NH⁺₄ and NO^-₂ contents in 0—30 cm soil. The research results  provide a theoretical basis for an in-depth understanding of the relationship between the distribution characteristics of  nitrogen and phosphorus in the aeration zone soil and the quality of shallow groundwater under different land use types.