132 results about "Echo state network" patented technology

The invention discloses an intelligent substation communication link fault accurate positioning method and system, and the method comprises the steps: obtaining a configuration file and equipment state data, carrying out the processing of the configuration file and the equipment state data, and generating a standardized link feature vector and a marking data set; constructing a hybrid deep learning model, and optimizing parameter configuration of the hybrid deep learning model by adopting an optimization algorithm to obtain a parameter-optimized hybrid deep learning model; training by using a real fault sample in combination with a virtual fault sample generated by a generative adversarial network, optimizing a time sequence prediction capability through an echo state network, and generating a fault positioning model; in combination with the link state data, outputting a fault link positioning result and confidence evaluation through multi-stage confidence evaluation and topological correlation analysis; and carrying out virtual-real corresponding verification in combination with the configuration file, carrying out parameter optimization on the fault positioning model, and outputting a fault positioning system. The problems that the fault positioning precision is low, the response speed is low, and complex fault scenes cannot be processed are solved.

The invention provides an adaptive data encryption method based on risk driving, and belongs to the technical field of data security based on deep learning. The method comprises the following steps: firstly, constructing a multi-dimensional input feature by collecting context information such as a communication behavior sequence, a topological structure and an equipment state; secondly, designing a communication risk intelligent assessment module based on an echo state network and a multi-head attention mechanism to realize joint identification of communication behavior types and risk levels; then, constructing an encryption strategy decision module composed of a rule tree and a strategy neural network cooperatively, and obtaining an optimal disturbance level, a disguise level, an encryption algorithm, a channel type and other strategy combinations in a limited strategy space according to an identification result; and finally, executing structure disturbance, behavior camouflage and data encryption packaging according to a strategy result, and performing secure transmission through a hidden channel. The method breaks through the limitations of lack of context modeling, uncontrollable strategy acquisition, extensive encryption granularity and the like in the traditional communication encryption technology, and has remarkable technical advantages.

The invention provides a distribution network traveling wave fault point intelligent positioning method and system based on deep time sequence feature learning, and belongs to the technical field of intelligent power distribution detection based on deep learning. Firstly, a high-speed traveling wave sensor is arranged on a distribution line, multi-dimensional three-phase voltage and current signals are collected, and a data set of fault types, phases, grades and positions is constructed; then, fault state features are extracted through topology perception normalization, symmetric component mapping and two-channel time sequence modeling, and accurate recognition of fault types, related phases and grades is achieved through an attention mechanism; furthermore, a fault intelligent positioning model based on a topological graph is constructed, a tower sensing weight and a line information bearing weight are introduced, and space-time embedding is extracted through a graph attention network and an echo state network, so that line fault classification and accurate positioning are realized. Finally, through combination of off-line model training and on-line system deployment, real-time identification and positioning of power distribution network faults are realized, and accuracy, robustness and response speed of fault diagnosis are effectively improved.

The invention relates to a boiler flue gas temperature dynamic compensation method based on CFD-reinforcement learning, and belongs to the technical field of boiler operation control. The method comprises the steps that boiler operation parameters and flue gas temperature data are collected in real time, a multi-scale coupling three-dimensional CFD model is constructed, a macroscopic gas-solid two-phase flow-combustion reaction-radiation heat transfer coupling algorithm and a microcosmic pulverized coal particle combustion dynamics algorithm are fused, and hearth and flue temperature distribution is simulated and output; building a CFD-reinforcement learning fusion agent based on a CFD model, taking similar working condition agent parameters as initial values, combining historical and virtual data staged training, and reinforcing the weight of a key temperature measurement point through a focusing regulation and control strategy; an echo state network-predictive control collaborative algorithm is adopted, real-time data are input to obtain an initial adjustment amount, and after security constraint optimization, a compensation execution mechanism is controlled to act. The flue gas temperature is accurately and dynamically compensated, the adaptability to complex working conditions is improved, and equipment safety and operation efficiency are both considered.

The invention discloses a power distribution network power prediction method, system and device based on multi-source heterogeneous data fusion, and a storage medium, and relates to the technical field of power distribution network data processing and prediction, and the method comprises the steps: constructing a data autocorrelation matrix based on the comprehensive operation and maintenance data of a power distribution network; performing eigenvalue decomposition on the data autocorrelation matrix, and constructing a principal component matrix; performing dimension reduction processing on zero-mean data obtained in the process of constructing the data autocorrelation matrix by using the principal component matrix to obtain operation and maintenance feature fusion associated data of the power distribution network; an echo state network model is established and trained, the operation and maintenance feature fusion associated data of the power distribution network is used as the input of the echo state network model, and the output power of the power distribution network is predicted; according to the method, the data dimension is reduced, the calculation complexity is reduced, and the output power of the power distribution network can be predicted more accurately; the operation efficiency and the management level of the power distribution network can be improved, and powerful support is provided for stable supply of a power system.

The invention discloses a chemical process multi-parameter intelligent monitoring method based on deep learning, and the method comprises the following steps: collecting and preprocessing the data of a chemical production process, and obtaining a standardized input sequence; according to the chemical process flow diagram and the equipment topological relation, a standardized process structure matrix is obtained; inputting the standardized input sequence and the standardized process structure matrix into a constraint decoupling echo state network model, and outputting a hidden state sequence; performing multi-scale gating fusion on the hidden state sequence, and outputting a parameter prediction result; calculating a residual error based on a parameter prediction result and an actual observation value, and outputting an anomaly detection result and channel-level attribution information; triggering a self-adaptive correction process to form a monitoring model after self-adaptive correction; and outputting a parameter prediction result, an anomaly detection result and channel-level attribution information based on the monitoring model after adaptive correction. According to the invention, the constraint decoupling echo state network is adopted, and multi-parameter intelligent monitoring of the chemical process is realized.

The invention discloses a method for estimating the energy state of a high-speed lithium battery, which belongs to the field of electric automobiles and comprises the following steps: performing multi-time scale modeling and hierarchical feature extraction on current, voltage and power data of the lithium battery through a deep echo state network, independently adjusting hyper-parameters of each layer by using a parameter differentiation strategy, and calculating the energy state of the lithium battery according to the hyper-parameters; therefore, different dynamic characteristic inputs can be adapted. Furthermore, sudden change output signals in the prediction process are removed through a modal decomposition noise reduction module, and the prediction precision and stability are improved. Experimental verification shows that the method shows high-speed prediction performance, high precision and high usability under various real automobile working conditions, not only has the capabilities of rapid training and efficient prediction, but also can effectively avoid the problem of gradient disappearance in a traditional recurrent neural network, and meanwhile, keeps higher precision and robustness.

The invention discloses a walking dysfunction intelligent evaluation system and method based on a flexible electronic technology, and the method comprises the following steps: S1, setting a multi-modal flexible sensor array, and synchronously collecting multi-modal data; s2, synchronously aligning the multi-channel signals by using an edge calculation unit to obtain a standardized gait feature tensor; s3, based on the dynamic Bayesian network, forming a high-dimensional hidden state probability distribution sequence of a complete gait cycle; s4, outputting a gait anomaly typing result and a walking dysfunction risk level based on the echo state network; s5, in combination with a knowledge rule engine and historical gait data, performing multiple verification and dynamic Bayesian network model parameter adaptive updating on a gait anomaly typing result; and S6, generating a rehabilitation evaluation report according to the gait anomaly typing result and the walking dysfunction risk level. According to the method, dynamic Bayesian network time sequence modeling and an echo state network intelligent discrimination algorithm are combined, and active detection and grading evaluation of walking dysfunction are achieved.

The invention discloses a dynamic quantitative feeding control system and method based on real-time working condition feedback of a reaction kettle, and the method comprises the steps: collecting the temperature, pressure, flow and other data of the reaction kettle, carrying out the denoising and standardization, and generating a working condition input data set; calculating heat release intensity, an energy residual error and a material residual error based on the working condition input data set, and constructing a physical feature vector; inputting the physical characteristics into an improved echo state network, and outputting temperature and pressure prediction; according to a prediction result and equipment limitation, temperature, pressure and rate boundary conditions are constructed to generate a security constraint set; self-evolution waveform optimization is adopted based on safety constraints, a candidate waveform set is generated, and an optimal feeding curve is selected; and inputting the optimal waveform to control feeding, establishing a trust propagation graph, calculating score gating and triggering safety interlocking. According to the invention, accurate prediction and safety control of reaction kettle feeding are realized by improving the echo state network and optimizing the self-evolution waveform.

The invention discloses a real-time visual processing method and system based on ESN-CV coprocessing, and relates to the technical field of visual processing, and the method comprises the steps: firstly, synchronously collecting a camera image and road surface humidity data of a physical sensor, and extracting an image reflection intensity distribution matrix as a visual feature; then humidity time sequence data and visual features are input into an echo state network, a dynamic weight coefficient matrix is generated through spatio-temporal feature fusion, the dynamic weight coefficient matrix is injected into a predefined convolutional layer of a target detection network, kernel weight parameters are adjusted in an element-by-element superposition mode, and the feature extraction capacity of a high-sensitivity area is enhanced. And after a detection result is output, the system triggers closed-loop feedback through a confidence coefficient deviation value: when the deviation exceeds a limit, the actual offset is calculated by combining a high-precision map, an error correction vector is generated, the state of the ESN reserve pool is updated by utilizing a Hadamard product, and the weight generation logic of the next frame is optimized in real time.

The invention provides a charging fault real-time diagnosis system based on edge calculation, and relates to the technical field of charging fault diagnosis. The method comprises the following steps: constructing a three-dimensional space model of a target charging area, and dividing the target area into M monitoring sub-areas; collecting charging fault data of each monitoring sub-region; preprocessing the charging fault data; constructing a fault risk assessment model by using an echo state network, and optimizing model hyper-parameters by using a pollen propagation algorithm; deploying a fixed edge node in a high-risk area in the fault risk level of each monitoring sub-area for continuous monitoring, planning an inspection path of a mobile edge node, and executing monitoring according to a dynamic period; receiving monitoring data of fixed and mobile edge nodes in real time, and constructing a fault prediction model based on a support vector machine to obtain a fault probability value; and the fault probability value is compared with a preset multi-level early warning threshold value, the fault level is judged, and the early warning information is output, so that the safety and the reliability of the charging process are ensured.

The invention provides an intelligent detection method and system for multiple pollutants in the urban solid waste incineration process, and belongs to the technical field of artificial intelligence. Inputting the input variables into the multi-module echo state network model to obtain predicted concentrations corresponding to the to-be-detected pollutants output by the multi-module echo state network model; wherein the multi-module echo state network model comprises two sub-modules, each sub-module is constructed for one pollutant, and each sub-module is provided with an independent input layer, a reserve pool and an output layer. According to the method, the multi-pollutant collaborative detection model based on the multi-module echo state network is constructed, and the concentration of various pollutants in the urban solid waste incineration process is accurately predicted through state sharing among the sub-modules.

The invention relates to the field of circuit breaker grid-connected and off-grid conversion, in particular to a system for realizing grid-connected and off-grid conversion based on a residual current protection circuit breaker, which is characterized in that a data sensing module acquires data through a Modbus-RTU protocol, processes the data through a linear active disturbance rejection control algorithm and outputs a pure data sequence; the power-off state analysis module predicts the power-off probability in combination with an echo state network algorithm, judges the island state through an improved island detection algorithm, fuses multi-source information by utilizing a D-S evidence theory, and judges whether the mains supply is powered off or not to generate a control instruction; the cooperative verification module carries out priority ranking on PCS based on an analytic hierarchy process and a fuzzy evaluation method and carries out cooperative control on off-grid mode switching by confirming the state of the commercial power, and the commercial power recovery module calculates the maximum recoverable power according to the rated power of the PCS and the SOC of a battery through load priority scheduling and a greedy algorithm, and gradually recovers load power supply; the system significantly improves the anti-interference capability of grid-connected and off-grid conversion, and is suitable for regions with unstable power grids. Customers feed back that the system is charged depending on an original photovoltaic system under the condition that the mains supply is in long-term power failure, so that the power utilization condition is greatly improved, and the power utilization difficulty is relieved.

The invention particularly relates to an excitation signal waveform intelligent optimization method for a high-power narrow-linewidth fiber laser, and the method comprises the steps: obtaining an excitation signal, carrying out the processing of a phase modulation waveform of the excitation signal, and extracting an input feature set needed by frequency spectrum prediction; constructing a frequency spectrum prediction model based on the decoupling echo state network, inputting the input features into the frequency spectrum prediction model based on the decoupling echo state network, realizing nonlinear mapping between the phase modulation waveform of the excitation signal and the frequency spectrum response thereof, and obtaining a prediction frequency spectrum; constructing a composite cost function including a mean square error, a spectrum sideband leakage penalty term and a spectrum smoothing term based on the difference between the predicted spectrum and the target spectrum; based on a composite cost function, an Adam optimization algorithm is adopted to iteratively correct a phase waveform in a frequency domain, intelligent waveform optimization of phase modulation is realized, and a threshold value of a stimulated Brillouin scattering effect is reduced. The intelligent optimization of the phase modulation waveform of the excitation signal can be realized.

The invention provides a method and a system for optimizing power closed-loop response time of a wind and light storage system. The method comprises the following steps: collecting multi-source historical operation data of the wind and light storage system within continuous preset duration; carrying out global optimization on key parameters of the echo state network by adopting a sethead whale migration algorithm; based on the collected data, outputting an advanced prediction result of the output of the wind turbine unit, the output of the photovoltaic unit and the change trend of the energy storage SOC in a future time period through the optimized echo state network; taking the advanced prediction result as input, and based on a weighted objective function taking power tracking error minimization and power closed-loop response time minimization as targets, adopting an adaptive snake vulture optimization algorithm to solve the objective function to obtain optimal control parameters of the wind-solar storage system; the optimal control parameters are analyzed into real-time control instructions through the FPGA controller so as to drive the energy storage unit, the wind generating set and the photovoltaic generating set to execute power regulation and control, and collaborative optimization of power closed-loop response time shortening and tracking precision improvement is achieved.

The invention discloses a heat exchange station thermal load modeling method integrated with an echo state network, and belongs to the technical field of industrial process intelligent detection. According to the method, dynamic time sequence characteristics and a nonlinear mapping relation of industrial process data are extracted by constructing a multi-layer echo state network (ESN) deep architecture; performing adaptive optimization on the key hyper-parameters of each layer of ESN by adopting a particle swarm optimization algorithm so as to obtain optimal reservoir dynamic characteristics and output weight; through a layer-by-layer progressive deep learning strategy, cascading the storage pool state of the front-layer ESN with an original input variable to serve as input of a subsequent layer, and realizing layer-by-layer abstraction and representation learning of features; and finally, adopting an average integration strategy to fuse the prediction output of the multi-layer ESN, and improving the prediction precision of the model. The deep ESN ensemble learning framework provided by the invention not only can fully mine complex dynamic laws in industrial time series data, but also can effectively inhibit an overfitting phenomenon of a single model, and achieves the purpose of improving the robustness and reliability of a soft measurement model.

The invention relates to the technical field of motor control, in particular to a speed stabilization control method and system for a hybrid power range extender, and the method comprises the steps: obtaining the operation disturbance of an HEV, training an echo state network to learn a time sequence dynamic state, enabling the disturbance to be engine crankshaft torque fluctuation, sampling into a historical sequence according to a control period, and enabling the network to comprise an input layer, a reservoir layer and an output layer; the reservoir is provided with a nonlinear recursive unit of leakage integral, and multi-step prediction is output; the relative angular displacement and relative angular velocity of an engine crankshaft and a planet carrier, the angular velocity of a second motor and the current engine torque are obtained at the control moment; driving the network multi-step look-ahead by using the current disturbance and the historical sequence, covering a control and prediction time domain, and forming a time margin not less than calculation and execution lagging; discrete state space prediction control is constructed based on the disturbance sequence, input and output constraints are applied, and a current state is used as an initial value for extrapolation; an optimal motor torque sequence meeting constraints is obtained through real-time iterative quadratic programming in the current period, and feed-forward compensation and smooth compromise are achieved; and caching the next period segment and issuing the next period segment to an execution mechanism on time. According to the invention, the problem that the HEV power system is easy to cause insufficient compensation or overcompensation during compensation can be solved.

The invention discloses a quick-response dynamic wireless power supply system and method suitable for mobile equipment. The method comprises the following steps: S1, acquiring running state data of the mobile equipment in real time; s2, preprocessing the operation state data; s3, predicting a power supply position track and a power demand of the mobile device based on the improved self-attention echo state network; s4, constructing a dynamic power supply control instruction set; s5, an array type wireless energy emission control module is driven, and energy directional transmission is carried out; s6, collecting parameters of a receiving end in real time at the receiving end of the mobile equipment; s7, correcting the control instruction set through the reinforcement learning adjustment model; and S8, updating network parameters and control strategies through incremental learning. According to the method, the improved self-attention echo state network and reinforcement learning are combined, intelligent prediction and closed-loop control of dynamic wireless power supply of the mobile equipment are achieved, and the method has the advantages of being fast in response, high in stability and high in self-optimization capacity.

The invention belongs to the technical field of DDoS attack prediction, and provides a DDoS attack prediction method and system based on a chaotic mapping echo state network. The method comprises the following steps: S10, determining an optimization objective function of a harmony search algorithm as a variance value predicted by an echo state network, and initializing related parameters of the harmony search algorithm; s20, generating an initial weight matrix by using the Logistic-tent chaotic mapping; s30, updating the harmony memory bank, selecting a weight matrix with the optimal fitness from the updated harmony memory bank, and constructing and training an ESN prediction model; and S40, processing the network flow sequence data detected in real time by using the trained ESN prediction model so as to perform DDoS attack prediction. According to the method, the echo state network DDoS attack prediction model based on chaotic mapping is adopted, the time efficiency is improved, and the DDoS attack situation can be effectively predicted in real time.

By accurately predicting the industrial aging process (IAP), such as the slow deactivation of catalysts in chemical plants, maintenance events can be scheduled further in advance, ensuring the cost - effectiveness and reliable operation of the plant. So far, these degradation progressions have typically been described by mechanical models or simple empirical prediction models. To accurately predict the IAP, data - driven models are proposed, comparing some traditional stateless models (linear and kernel ridge regression, and feed - forward neural networks) with more complex state - recursive neural networks (echo state network and long - short - term memory network). In addition, variants of the stateful models are discussed. In particular, stateful models that use mechanical pre - knowledge about the degradation dynamics (hybrid models). Stateful models and their variants may be more suitable for generating near - perfect predictions when trained on a sufficiently large dataset, while hybrid models may be more suitable for better generalization in the case of smaller datasets under changing conditions.

The invention relates to an engine flame tube life prediction method, equipment and a program product. The method comprises the following steps: acquiring a crack propagation parameter of a target flame tube to be subjected to life prediction; inputting the crack propagation parameters into a pre-trained life prediction agent model, and outputting a residual life prediction value of the target flame tube through the life prediction agent model; wherein the life prediction agent model is obtained by training the echo state network by taking a flame tube crack propagation data set obtained by simulation as a training sample, and the flame tube crack propagation data set comprises a plurality of groups of crack propagation parameters and corresponding residual life prediction values. According to the method, the complex nonlinear relationship between the crack form and the life can be efficiently mapped, and the residual life prediction value can be quickly and accurately output; moreover, the method does not need to depend on complex real-time simulation calculation, can be directly connected with actual monitoring data, and effectively reduces the operation and maintenance cost.

The invention relates to an adaptive multi-scale decomposition echo state network chaotic time sequence prediction method suitable for meteorological prediction, and belongs to the field of chaotic time sequence prediction. Comprising the following steps: performing multi-scale decomposition on an input sunspot number sequence by using an HP filter; each sub-sequence obtained through decomposition is distributed to an ESN sub-network; inputting a weight in the ESN sub-network by adopting a Xavier weight initialization method, introducing a composite activation function in reservoir nonlinear updating, and performing state splicing by adopting a state splicing strategy; a sparse random matrix is constructed, a unit matrix scale factor is introduced, and the spectral radius is strictly controlled; each ESN sub-network independently trains an output weight, ridge regression is adopted for solving, integration and reconstruction are carried out after corresponding component prediction is completed, and if a target sequence is a sunspot activity index, the model finally outputs a sunspot number prediction value in a future time step; according to the method, high-precision, stability and robustness prediction of the chaotic time sequence can be realized.

The invention discloses an ammonia nitrogen concentration prediction method, a computing device and a computer readable storage medium wherein the ammonia nitrogen concentration prediction method comprises: determining a plurality of characteristic factors affecting the ammonia nitrogen concentration, and collecting historical data corresponding to each characteristic factor wherein the ammonia nitrogen concentration is the ammonia nitrogen concentration in the sewage treatment process; determining a correlation coefficient between each characteristic factor and the ammonia nitrogen concentration, and determining training data for a variable structure echo state network model from the historical data based on the correlation coefficient; constructing the variable structure echo state network model; and optimizing the variable structure echo state network model by using the training data, and predicting the ammonia nitrogen concentration in the sewage treatment process by using the optimized variable structure echo state network model. Therefore, the ammonia nitrogen concentration can be accurately determined in the sewage treatment process.

The invention belongs to the technical field of artificial intelligence and network predictive maintenance, relates to a fault prediction method based on a multi-source migration echo state network, and aims to solve the problem of low prediction precision caused by insufficient early samples of a fault, and the method comprises the following steps: firstly, extracting migration knowledge from a plurality of historical fault source domains by using the echo state network; secondly, measuring the similarity between the source domains and the target domain and the difference between the source domains by adopting a dynamic time warping distance, and selecting a plurality of similar source domains with complementary information; then, a plurality of prediction sub-models are established in the target domain in combination with migration knowledge of the selected source domain, and a final prediction model is obtained through integrated learning; and finally, performing future value prediction on the key variables acquired in real time by using the model. According to the method, multi-source historical information can be fully utilized under the small sample condition, the fault prediction precision and generalization ability are remarkably improved, and the method is suitable for fault prediction scenes with multi-source time series data in industrial processes, mechanical equipment, power systems and the like.

The invention discloses an expressway traffic flow robust prediction method based on a Huber loss function, which analyzes traffic flow data collected by an expressway detector through a machine learning algorithm to realize accurate prediction of future traffic flow. The prediction method comprises the following steps: step 1, acquiring time sequence data from each sensor of a highway, and calculating and capturing the output x (t) of a time sequence dependency echo state network (ESN) reserve pool; 2, designing an echo state network model with a Huber loss function and a two-norm regularization item, converting the model into an equivalent form of linear constraint, and solving the model by using an alternating direction multiplier method (ADMM); and step 3, adjusting a threshold value c of a Huber loss function according to the prediction performance, obtaining an echo state network model with the strongest robustness and generalization performance, and obtaining a result similar to a result without an outlier under a high outlier proportion.

The invention discloses a power distribution network stability control method based on multiple agents. The method comprises the following steps of deploying the agents, collecting data and constructing graph data; establishing an improved graph echo state network and forming a structured input sequence; inputting an input sequence to carry out time sequence evolution, and forming a dynamic echo state vector; exchanging dynamic echo state vectors with a neighborhood agent, and splicing to form comprehensive time sequence state information; each agent analyzes the comprehensive time sequence state information to generate a local stability control instruction; when disturbance occurs, disturbance updating comprehensive time sequence state information is constructed, and a disturbance updating local stability control instruction is generated; and periodically and continuously executing dynamic echo state vector updating, comprehensive time sequence state information construction and local stability control instruction generation in a normal operation period. According to the invention, the improved graph echo network is adopted to realize stable control of the power distribution network, and the method has the advantages of strong real-time performance and fast disturbance response.

The application discloses a power distribution network stability control method based on multiple agents, comprising the following steps: deploying agents, collecting data and constructing graph data; establishing an improved graph echo state network and forming a structured input sequence; inputting the input sequence for time evolution and forming a dynamic echo state vector; exchanging the dynamic echo state vector with neighboring agents to splice and form comprehensive time sequence state information; each agent analyzes the comprehensive time sequence state information to generate a local stability control instruction; when a disturbance occurs, the comprehensive time sequence state information is updated and a disturbance updated local stability control instruction is generated; and the dynamic echo state vector is updated, the comprehensive time sequence state information is constructed and the local stability control instruction is generated periodically and continuously in a normal operation cycle. The application realizes power distribution network stability control by using an improved graph echo network, and has the advantages of strong real-time performance and fast disturbance response.

Procedure for determining the reliability of the modeling of the dynamics of a traffic intersection, comprising the steps: S1 - Recording sensor data regarding the movement of road users at the traffic junction; S2 - Using a trained deep learning network model with the sensor data, wherein the deep learning network model includes at least one echo-state network unit (ESN), and wherein a result of the deep learning network model is a model of the dynamics of the traffic node; S3 - Determining a learned regularization coefficient of the deep learning network model, where the regularization coefficient indicates the reliability of the modeling of the dynamics of the traffic junction.

The present invention belongs to the field of physical layer security, and discloses a time-delay feature extraction method, system, device and medium based on weight similarity. The method includes: generating a chaotic time series as a data set; constructing an echo state network; selecting multiple time delays, and constructing multiple data set pairs according to different time delays; translating a certain amount for each data set pair to generate a corresponding data set pair; training the network for two data set pairs with the same time delay respectively, and extracting weights; performing similarity measurement on the two extracted weight matrices; and inferring the time-delay information according to the statistical analysis result of the similarity measurement. The present invention can analyze the time-delay key of a two-dimensional or even higher-dimensional chaotic system, not limited to one-dimensional time delay, and does not require prior knowledge of the chaotic system structure and mathematical model. Only a small amount of data is needed to analyze the key, saving computing power and having high accuracy.

A transformer health state evaluation method based on a leaky-integrator echo state network includes the following steps: collecting monitoring information in each substation; performing data filtering, data cleaning and data normalization on the collected monitoring information to obtain an input matrix; inputting the input matrix into a leaky-integrator echo state network to generate trainable artificial data, and dividing the artificial data into a training set and a test set in proportion; constructing a deep residual neural network based on a squeeze-and-excitation network, and inputting the training set and the test set for network training; and performing health state evaluation and network weight update based on actual test data. Considering that a deep learning-based neural network needs a large amount of data, the present disclosure uses the leaky-integrator echo state network to generate the artificial training data.