39 results about "Hill climbing" patented technology

A power load prediction method, system and device based on GHOA-CNN-iTransform-LSTM, and the method comprises the steps: firstly obtaining the multi-source data of a power load and analyzing the causal relationship, then constructing an analysis model based on scene injection, constructing an improved hiking optimization algorithm GHOA based on a hierarchical strategy evolution framework, optimizing the hyper-parameters of the analysis model, and finally obtaining the multi-source data of the power load. And finally, after the analysis model is trained, power load prediction of the smart power grid is realized. By acquiring the multi-source data, analyzing the causal relationship and adaptively transforming the kernel of the analysis model, the analysis model can integrate the time and the causal logic for dynamic allocation, accurate perception of external events is realized, a stable deduction result is obtained from complex data, and the accuracy of the deduction result is improved. According to the method, the self-adaptive optimization of the optimal parameters is realized through the hierarchical strategy evolution framework, the search strategy is ensured to dynamically adapt to the high dimension and complexity of the parameter space, and the precision and adaptability of power load prediction are ensured.

The invention provides a method for adjusting a variable-speed pumped storage unit under a working condition of a water turbine, which comprises the following steps of: arranging a multi-source sensor array in a flow channel to collect parameters such as water head pressure, rotating speed fluctuation and active power, generating operation state data through space-time alignment and self-adaptive weighted fusion, and triggering re-calibration when a threshold value is exceeded; an efficiency optimization index is calculated, fluctuation is smoothed through moving average filtering, water head fluctuation is predicted, and early warning is conducted when the water head deviates from the optimal working condition; according to the efficiency index and the water head amplitude variation factor, a maintenance / early warning / emergency adjustment grade is divided, and a guide vane instruction is generated during high deviation; determining unit importance, evaluating resource requirements by using a multi-objective optimization algorithm, and dynamically monitoring deviation; and optimizing a rotating speed regulation scheme by adopting a self-adaptive hill climbing algorithm, evaluating an iterative optimization regulation path, and triggering parameter re-calibration when the deviation exceeds the limit. The operation efficiency, the adjusting precision and the stability of the variable-speed pumped storage unit under the working condition of the water turbine can be improved, and the energy consumption and the instability risk are reduced.

The application discloses a kind of model construction method and device of wind power generation system, the method wind power generation system, at least including wind turbine and permanent magnet synchronous generator, the method includes: the mechanical characteristics of wind turbine is modeled according to momentum theory data to obtain wind turbine model;Maximum power output point of wind turbine is tracked by analyzing wind turbine model and using hill climbing search method;Ideal conditions met by permanent magnet synchronous generator are obtained;Under ideal conditions, three-phase voltage equation under natural coordinate system, flux linkage equation under natural coordinate system, electromagnetic torque equation of motor, mechanical motion equation of permanent magnet synchronous generator are established to obtain permanent magnet synchronous motor model.The application can reduce the loss generated by mechanical rotation in power generation process, so that the power generation efficiency of system is higher.

The invention relates to the technical field of factory production scheduling, and provides a factory production scheduling plan automatic generation method and device and electronic equipment, task information of a to-be-scheduled task and production elements of a production factory are obtained, and the production factory is used for producing the to-be-scheduled task; and based on the task information of the to-be-scheduled task and the production elements of the production factory, determining an optimal scheduling plan of the production factory for the to-be-scheduled task through mixed integer programming and a hill climbing algorithm. By combining the accuracy of the mixed integer programming and the high efficiency of the hill climbing algorithm, a high-quality production scheduling scheme can be found, and calculation can be completed within acceptable time, so that automatic and efficient production scheduling is realized, and the production efficiency and the management level of a factory are improved.

A multi-agent path planning method based on an adaptive meme algorithm addresses the immaturity and slow convergence speed of current multi-agent path planning methods. This method employs a structure combining global optimization and local search, utilizing chromosomes as generated paths in a two-dimensional grid map and optimizing their fitness values ​​through strategies such as "selection" and "crossing." A hill-climbing algorithm is used as a sub-algorithm for local optimization to prevent generated solutions from getting trapped in local optima, thereby improving the accuracy of multi-agent motion control and enhancing the algorithm's convergence speed and optimization efficiency. This solves the problems of slow convergence speed and low path optimization accuracy that are difficult to address in current multi-agent path planning methods. Experimental results show that this method can accurately optimize multi-objective paths for multiple agents, and the average length of each path converges to the globally optimal path length with population iterations, achieving the expected results.

The application discloses a kind of climbing auxiliary service demand price curve drawing method, device and storage medium, wherein method includes: obtaining the system net load value of preset scheduling interval, system net load value includes system net load forecast value and system net load measured value;System net load value is classified as the typical day data set of different time periods;Calculate the system net load error of each scheduling interval, obtain the system net load error set corresponding to different time periods;According to system net load error set, the frequency of system net load prediction error in different error intervals is counted, and the error frequency of frequency relative to the data quantity in the system net load error set of current period is calculated;According to error frequency, probability density distribution histogram is drawn, and the climbing auxiliary service demand price curve of each period is drawn.The present application can comprehensively reflect electric energy cost and climbing auxiliary service benefit, and provide effective data support for building climbing auxiliary service market.

The invention provides an optical lens focusing processing method and device, equipment and a medium, and relates to the technical field of automatic focusing, and the method comprises the steps: obtaining the environment brightness information of a scene where imaging equipment is located at present, so as to determine a focusing strategy corresponding to the environment brightness information; collecting a target definition sequence according to the search resource configuration information in the focusing search process of the imaging equipment; under the condition that the definition trend feature corresponding to the target definition sequence does not meet the trend judgment condition, continuing to collect a new target definition sequence until the definition trend feature corresponding to the new target definition sequence meets the trend judgment condition; and controlling the imaging device to focus based on the new target definition sequence. The method can get rid of the dependence of a hill climbing algorithm on the definition value, and further improves the focusing speed and focusing accuracy in a low-light and low-contrast scene.

The application provides a communication interference strategy generation method based on reinforcement learning, comprising the following steps: step one, constructing a system model of a communication party and a communication interference party; step two, based on the system model of the communication party and the communication interference party constructed in step one, learning an anti-interference scheme of the communication party by using a win-or-learn policy hill climbing algorithm, and designing a corresponding interference decision model; step three, using the interference decision model obtained in step two, learning an anti-interference strategy of a communication target and implementing interference according to a decision process of 'observation-adjustment-decision-action'. The application comprehensively considers interference basic principles and behavior changes of the communication target, combines a jam-to-noise ratio and anti-interference behaviors such as frequency change and transmission power increase of the communication target after being interfered, and uses the two as measurement indexes of interference effects, so that the purpose of real-time and rapid interference is achieved.

This invention discloses an automatic image focusing method for an imaging device, comprising the following steps: obtaining several target images; randomly generating a parameter set characterizing the lens physical parameters of the imaging device's lens; based on the parameter set as the initial position of the lens, using a neighborhood generation rule to perform operations on a preset step size at the initial position to generate a neighborhood, each neighborhood corresponding to a lens position; evaluating the focusing effect by comparing the focusing effect evaluation values ​​corresponding to different lens positions; obtaining a locally optimal lens position based on a hill-climbing algorithm; and processing the locally optimal solution based on a simulated annealing algorithm to obtain the optimal solution. This invention can effectively find the globally optimal focusing parameters, ensuring the clearest image is captured under this focusing condition, and combines the hill-climbing algorithm and the simulated annealing algorithm to achieve automatic focusing.

The invention relates to a photovoltaic cluster climbing event prediction method based on a satellite cloud picture, in particular to the technical field of photovoltaic power generation power climbing prediction, and the method comprises the steps: obtaining corresponding photovoltaic power data, weather forecast data and satellite cloud picture data of a photovoltaic power station in a photovoltaic cluster within a preset collection time, the preset acquisition time is a time period before the prediction time of the photovoltaic cluster climbing event; determining a forecast meteorological factor corresponding to the photovoltaic cluster from the weather forecast data according to the association strength between the weather forecast data and the photovoltaic power of the photovoltaic power station; determining the extreme weather of the photovoltaic cluster corresponding to the prediction time according to the satellite remote sensing value of the photovoltaic cluster geographic area corresponding to the photovoltaic cluster in the satellite cloud picture data; determining the predicted power of the photovoltaic cluster corresponding to the prediction time based on the extreme weather and the forecast meteorological factor; and determining a photovoltaic cluster climbing event corresponding to the prediction time according to the climbing characteristic quantity corresponding to the prediction power.

The application provides a flexible workshop scheduling optimization method based on a meta-heuristic algorithm, data sets of flexible workshop scheduling are acquired; a differential evolution algorithm based on a trigonometric function mutation is adopted to process the data sets of flexible workshop scheduling and preliminarily form a scheduling scheme by introducing a nonlinear disturbance and a dynamic adjustment mechanism of a companion population; the adaptive hill climbing algorithm is adopted to optimize a current optimal solution and process again to form an optimal scheduling scheme by combining a multiple disturbance mechanism and a disturbance intensity adjustment strategy; and the optimal scheduling scheme achieving the maximum completion time is visually presented; the flexible workshop scheduling optimization method based on the meta-heuristic algorithm can improve the ability to jump out of a local optimum, can significantly reduce the maximum completion time and can improve the solution efficiency of the flexible workshop scheduling problem.

The invention discloses a curvature method-based method for determining an optimal temperature regulation interval in a whole growth stage of greenhouse peppers, which comprises the following steps of: measuring photosynthetic rates of the peppers in different growth stages under different environmental conditions through a nested experiment, and constructing a data set; the environmental conditions comprise temperature, CO2 concentration and photosynthetic photon flux density; taking the growth stage, the CO2 concentration, the temperature and the photosynthetic photon flux density as input vectors, constructing a photosynthetic rate prediction model by adopting a back propagation neural network optimized by a genetic algorithm, and generating a corresponding photosynthetic rate response surface; dispersing the photosynthetic rate response surface into discrete points, calculating the Gaussian curvature of each point, and constructing a Gaussian curvature response surface; and discretizing the Gaussian curvature response curved surface into a temperature-Gaussian curvature response curve, taking a u chord value as a measure, determining a critical boundary of the response curve by adopting a random restart hill climbing method, and finally obtaining an optimal temperature regulation and control interval.

The present application belongs to the field of quantum physics and quantum technology, and particularly relates to an autofocus method for atomic imaging, comprising the following steps: acquiring a fluorescence or absorption image of single atoms in a region to be imaged to obtain two-dimensional gray image data; determining the position of single atoms in the two-dimensional gray image data, and intercepting a region of interest (ROI) of a preset size centered on the position of single atoms; calculating the centroid coordinates of a light spot in the ROI, and taking the centroid as a reference center of a subsequent sharpness evaluation index; calculating a sharpness evaluation index based on the ROI, and combining the sharpness evaluation index according to a preset weight to obtain a sharpness score S; performing a focus search strategy, including an initial scanning stage and a hill climbing search stage, to determine an optimal focus position z best ; and controlling an electric displacement table to adjust the camera to the optimal focus position z best . The present application can improve imaging resolution and signal-to-noise ratio, reduce manual intervention, quickly adapt to different experimental modes, and improve experimental efficiency.

This invention discloses an intelligent fire-fighting patrol vehicle that integrates hazard inspection, fire identification, path planning, and initial fire suppression. The system combines lidar, visual recognition, infrared, and ultrasonic sensors with a YOLO visual model, an improved A* algorithm, and PID control to achieve environmental mapping, fire source location, autonomous navigation, and precise fire suppression. The vehicle uses a Mecanum wheel chassis, enabling omnidirectional movement, obstacle crossing, and hill climbing, making it suitable for complex environments such as factories, shopping malls, hospitals, and schools. The fire suppression module dynamically adjusts the extinguishing agent and spray parameters according to the fire situation, improving fire suppression efficiency and reducing water consumption. This invention effectively enhances the intelligence level of fire inspection, reduces manual intervention, and strengthens emergency response capabilities and safety.

The invention discloses a power distribution network three-phase load balance optimal distribution method, and relates to the field of power distribution network optimization in a power system. The problem that the loss of a power distribution network line is increased due to three-phase imbalance of an existing power distribution network is solved. According to the method, a user clustering-based load three-phase balanced distribution model is adopted, and a mountain climbing algorithm is adopted to optimize a moth flame algorithm, so that the local optimization capability of the moth flame optimization algorithm is enhanced, a three-phase balanced optimal distribution strategy is obtained, the three-phase load is balanced to the greatest extent from the source, and accurate loss reduction of the load side of the power distribution network is realized. The method is suitable for obtaining the three-phase load balance optimal distribution strategy.

The invention discloses a shipyard total group tray site layout optimization method based on simulated annealing in combination with a local hill climbing algorithm, which comprises the following steps: acquiring basic data of a shipyard total group tray site, constructing a multi-target optimization model, and optimizing the multi-target optimization model by taking the maximum space utilization rate, the minimum transportation cost and the minimum process conflict as optimization targets. And solving the multi-objective optimization model by adopting a simulated annealing-local hill climbing hybrid algorithm to obtain an optimal layout scheme, performing constraint verification and effect evaluation on the optimal layout scheme, and outputting a final layout scheme. Compared with a traditional empirical layout, the space utilization rate is improved by 15%-25%, the transportation cost is reduced by 20%-30%, the secondary transfer rate is reduced to 5% or below, and compared with a single simulated annealing algorithm, the convergence rate is improved by 40%-50%; compared with a single local hill climbing algorithm, the method has the advantages that the target function value is reduced by 10%-15%, parameters can be adjusted according to the actual condition of a shipyard, an output scheme can be directly connected with an MES system, and intelligent production decision making is supported.

The application discloses a neural network deployment method of a ReRAM memory-computing integrated array, comprising the following steps: (1) adopting a Reset-Set double sequence verification combined with a non-disturbance conductance reading, subdividing device states of the ReRAM array into three categories of easy-to-use states, low-impact states and high-impact states; configuring differentiated scores, and generating a state matrix and a score matrix consistent with the dimension of the ReRAM array; (2) based on the state matrix and the score matrix, taking the maximum normalized score rate of a subarray as the target, and through the combination of a greedy algorithm, a hill climbing algorithm and a genetic algorithm, a target subarray matched with a target weight matrix is screened out; (3) the target weight matrix is mapped to the target subarray according to rows and columns, the weight is converted into a device conductance through an adaptive programming pulse, and solidification deployment is completed. By using the application, the hardware utilization rate of the ReRAM memory-computing integrated array and the neural network weight deployment efficiency can be significantly improved.

The application discloses an automatic focusing method of a limited far variable conjugate distance variable magnification lens, which comprises the following steps: constructing a measurement two-dimensional lookup table according to a variable magnification group magnification Z, a uniformly coarse divided object distance WD and a focusing group position FP; performing polynomial fitting on each column of data of the measurement two-dimensional lookup table, and constructing a fitting two-dimensional lookup table FP'(WD, Z) by using the fitting values of the uniformly coarse divided WD and FP; moving the variable magnification group and the focusing group to realize variable magnification following according to the current actual WD and the FP'(WD, Z); determining a search range and a single search step in the FP'(WD, Z) according to the current Z, and performing fast automatic focusing by using a hill climbing search method; reversely searching the closest FP by taking the current Z and the current FP in the automatic focusing result as the key, and taking the WD corresponding to the closest FP as a new current WD; and repeatedly performing subsequent variable magnification following and fast automatic focusing according to the new current WD and the FP'(WD, Z). The method reduces the measurement workload, guarantees the accuracy of subsequent variable magnification following, improves the experience of users, and further reduces the time of subsequent automatic focusing.

The application discloses a kind of 8 bit S box construction method based on improved genetic algorithm, comprising the following steps: (1) by Logistic chaotic mapping and Tent chaotic mapping to form a kind of compound chaotic system, generates initialization population;(2) combined with the nonlinearity of S box, difference uniformity and strict avalanche criterion generates a multi-objective segmented weighted fitness function;(3) using the cross operation of guaranteeing S box bijectivity;(4) construct a reverse search method to search mutation point, and combined with hill climbing algorithm, random search method, generate based on 2N round search mutation operation, 2 point mutation point search is carried out in the first N rounds, 3 point mutation point search is used in the last N rounds, and a look-up table statistics method is further used to evaluate the property of optimized S box;(5) based on improved genetic algorithm constructs 8 bit password S box.The method of the application can generate 8 bit S box with excellent cryptographic properties based on improved genetic algorithm.

A method for predicting the health status of spacecraft electromechanical components considering the evolution of performance characteristics is proposed. The steps are as follows: First, collect full lifecycle data of spacecraft electromechanical components and preprocess their multidimensional telemetry parameters. Then, extract explicit and implicit features from the multidimensional telemetry parameters to form a high-dimensional feature set, and perform dimensionality reduction and fusion on the high-dimensional feature set. Next, filter and model the fused curves, and divide the performance degradation stages based on a hill-climbing algorithm using kernel density estimation. Finally, construct a bidirectional LSTM prediction network in stages and predict the health status for different evolution stages. This method effectively integrates massive amounts of telemetry data through feature extraction and feature fusion. Simultaneously, by utilizing a hill-climbing algorithm based on kernel density estimation and a bidirectional LSTM prediction network to consider the evolution mechanism of different degradation stages when predicting the health status, it effectively improves the prediction accuracy of the health status of key spacecraft components.

The present application relates to the technical field of data processing, and more particularly to a power load prediction method and system in a virtual power plant. The method comprises the steps of obtaining reference samples and to-be-predicted samples, obtaining the difference degree of the to-be-predicted samples and the reference samples; obtaining the special degree of the to-be-predicted samples according to the difference degree of the to-be-predicted samples and the reference samples, correcting the number of existing reference samples according to the special degree of the to-be-predicted samples, and obtaining the corrected number of reference samples; combining a random hill climbing algorithm to optimize the corrected number of reference samples to obtain an optimal reference sample number; and constructing a prediction model according to a GBDT algorithm and the optimal reference sample number, inputting the to-be-predicted samples into the prediction model, and obtaining the power load prediction value of the to-be-predicted samples. The present application adapts to the optimal reference sample number, and improves the prediction accuracy of the to-be-predicted samples.

This invention discloses a method and system for improving the continuous uphill power depletion of range-extended electric vehicles (REEVs), relating to the field of automotive powertrain optimization technology. The method includes: acquiring real-time data on factors affecting the uphill operation of the electric vehicle, including vehicle-specific factors and external environmental factors; evaluating the driving state of the electric vehicle during uphill climbing based on the acquired data to determine whether the current driving state meets the uphill requirements; and generating a corresponding optimization strategy based on the determination result. The proposed method for improving the continuous uphill power depletion of REEVs ensures that the electric vehicle can continuously provide sufficient power to meet the uphill requirements through real-time monitoring and adjustment. By adjusting the optimization strategy, it reduces the energy consumption of non-critical systems and adjusts the intensity of the energy recovery system, effectively extending battery life, reducing the occurrence of power depletion problems, and lowering the safety risks caused by insufficient power.

The application provides a fault diagnosis method and device for a GIS disconnector of a motor. The method comprises the following steps: firstly, obtaining a power curve in a motor operation process, and preprocessing the power curve to obtain a first target curve, wherein the preprocessing at least comprises outlier filtering and missing value completion; then, calculating a correlation coefficient based on the first target curve through a hill climbing algorithm to obtain a target coefficient; finally, comparing the target coefficient with a first threshold to determine whether a disconnector of the motor has a fault. The application draws a power curve based on a current and a voltage of the motor, and determines the correlation based on the power curve optimized through the hill climbing algorithm, thereby avoiding the interference of different periods of the power curve, and solving the problem of low accuracy in judging the mechanism blockage and mechanism misplacement faults in the prior art.

This invention relates to the field of power resource dispatching technology, specifically to a method and device for dynamic aggregation of virtual power plants based on two-layer decision-making and satisfaction. The method includes: using the virtual power plant as the upper-layer decision-maker in a two-layer decision-making architecture, and the corresponding entities of the participants as lower-layer decision-makers, to determine the aggregation scheme; wherein, the upper-layer decision-maker uses a two-layer late-acceptance hill-climbing algorithm combined with a proxy objective function to iteratively optimize internal transaction price information, and distributes it to the lower-layer decision-makers through a blockchain network; the lower-layer decision-makers listen to and obtain information through smart contracts, and when trigger conditions are met, call their local participant model to calculate the optimal operating scheme and calculate satisfaction information based on the Frynell satisfaction model, and upload it to the blockchain network for the upper-layer decision-maker to adjust the internal transaction price information of the dispatching cycle. This effectively improves the flexibility of the virtual power plant in dispatching power resources, thereby improving dispatching efficiency.

The invention belongs to the technical field of electric power systems, and discloses a cost evaluation optimization control method and system for a climbing auxiliary service, and the method comprises the steps: building a layered architecture in which a cold storage and heat storage load participates in the climbing auxiliary service, and achieving the uploading and issuing of real-time operation data and a power grid dispatching instruction between an upper layer and a lower layer; on the basis of the cost dimension type of the cold and heat storage load participating in the climbing auxiliary service, variable parameters needed by cost evaluation of all dimensions are obtained, and cost data of different dimensions are calculated respectively; based on a power grid dispatching instruction received by a cold storage and heat storage load aggregator, a climbing instruction optimization controller is constructed in combination with the multi-dimensional cost data and the user comfort factor, power grid climbing instructions are optimally distributed, and the power grid climbing instructions are issued to each cold storage and heat storage load to be executed. The method can guide the cold and heat storage load to participate in the auxiliary service market, support the climbing demand of the load main body in response to the power system, and promote the application demonstration of the large-scale cold and heat storage load in power grid regulation and control.

The application provides an optical lens focusing processing method, device, equipment and medium, and relates to the technical field of automatic focusing. The method comprises the following steps: acquiring environment brightness information of a scene in which an imaging device is currently located, so as to determine a focusing strategy corresponding to the environment brightness information; in the process of focusing search of the imaging device, collecting a target sharpness sequence according to search resource configuration information; in the case that a sharpness trend feature corresponding to the target sharpness sequence does not satisfy a trend judgment condition, continuously collecting a new target sharpness sequence until the sharpness trend feature corresponding to the new target sharpness sequence satisfies the trend judgment condition; and controlling the imaging device to focus based on the new target sharpness sequence. The application can break the dependence of the hill-climbing algorithm on the size of the sharpness value, and thus improve the focusing speed and focusing accuracy in a low-light and low-contrast scene.

The application discloses an improved particle swarm hybrid intelligent agricultural machine path planning algorithm fusing a climbing strategy, and comprises the following steps: S1, inputting path planning data, assuming that each particle represents a path, and initializing particle swarm positions in the improved particle swarm algorithm by using a Tent chaotic mapping algorithm; S2, updating the speed and position of the particle by using a random inertia weight updating strategy and a particle speed updating formula, and dynamically adjusting a learning factor by using an asynchronous dynamic adjustment algorithm; S3, judging whether an end condition of the path planning data is met or not, and re-executing S2 when the end condition is not met; and S4, taking the initial solution meeting the end condition as the input of the climbing algorithm to continue path optimization, and outputting an overall optimal particle, namely an optimal path, after searching by the climbing algorithm. The algorithm has certain improvement in the optimization capability in a complex scene, especially in a curved surface terrain, and the capability of jumping out of a local optimal solution.

The application discloses a kind of MTF template matching measurement methods suitable for multiple focusing lenses, specifically: industrial camera moves to focal plane, and shoots a focal plane image for subsequent template matching scale adjustment;Find the best zoom ratio using hill climbing algorithm, adjust the template size according to the ratio;After adjusting the template size, match the measurement area according to the template;Semantic segmentation detects stripe, after determining the horizontal and vertical stripe area, calculate the contrast of each area;Sort the contrast data according to the area, thereby corresponding to different stripe area number, mark the outer ring and inner ring measurement results and stripe number on the image, and finally draw two MTF curves of horizontal stripe and vertical stripe according to the measurement data.The application can improve the efficiency and accuracy of MTF measurement.

The invention discloses a neural network deployment method for a ReRAM storage and calculation integrated array, and the method comprises the steps: (1) employing Reset-Set double-sequence verification in combination with undisturbed conductive reading, and dividing the device state of the ReRAM array into an easy-to-use state, a low-influence state, and a high-influence state; differentiated scores are configured, and a state matrix and a score matrix which are consistent with the dimension of the ReRAM array are generated; (2) based on the state matrix and the score matrix, with the maximum sub-array normalization score rate as a target, screening out a target sub-array matched with the target weight matrix through combination of a greedy algorithm, a hill climbing algorithm and a genetic algorithm; and (3) mapping the target weight matrix to a target sub-array according to rows and columns, and converting weights into device conductance through adaptive programming pulses to complete curing deployment. According to the invention, the hardware utilization rate of the ReRAM memory-computing integrated array and the neural network weight deployment efficiency can be obviously improved.

The application discloses a transformer fault diagnosis method, device and equipment and a medium, relates to the technical field of transformer fault diagnosis, and adopts a terahertz time domain spectrum technology to acquire terahertz spectrum information of transformer fault insulating oil, decomposes terahertz spectrum sequence data by using a complementary ensemble empirical mode decomposition (CEEMD), calculates the fuzzy entropy of each subsequence after decomposition, so as to form characteristic quantities reflecting the transformer fault state. This process only extracts and processes the terahertz spectrum information of the transformer insulating oil, ignores the influence of the insulating oil, the internal and external environment and the transformer equipment itself, and then a new transformer fault diagnosis model is established according to a regularized extreme learning machine (RELM) and a hill climbing optimization method (HOA). The fault type of the transformer is diagnosed according to the transformer fault diagnosis model, and the accuracy of the transformer fault diagnosis is greatly improved.