72 results about "Unsupervised learning algorithm" patented technology

The invention discloses a network security threat research and judgment method, system and device and a storage medium, and the method comprises the following steps: S1, obtaining network traffic, terminal logs, application program interface calling records and threat intelligence data in real time, carrying out the standardized cleaning and format conversion of the data, and building a unified data lake; s2, matching, identifying and determining threats through a preset known threat feature library, constructing a normal behavior baseline by using an unsupervised learning algorithm, and marking suspicious events deviating from the baseline; s3, for the suspicious event marked in the step S2, mining a potential attack path and an attack intention by combining knowledge graph technology associated asset information, a historical attack chain and a homologous IP address; and S4, based on the attack success probability, the influence asset importance and the diffusion speed, calculating a threat level by adopting a fuzzy comprehensive evaluation model, and generating a research and judgment report containing disposal suggestions.

The invention is suitable for the technical field of agricultural intellectualization, and provides an intelligent blueberry disease detection method and system based on multi-modal unsupervised learning, and the method comprises the following steps: carrying out the feature extraction and clustering of preprocessed multi-modal data based on an adaptive contrast deep clustering framework, and obtaining a feature extraction result; obtaining a multi-modal preliminary feature and a preliminary clustering result; based on a multi-modal complementary feature fusion mechanism, according to the multi-modal preliminary features and the preliminary clustering result, carrying out adaptive weighted fusion on the multi-modal preliminary features to obtain multi-modal fusion features; performing unsupervised clustering optimization and disease type identification on the multi-modal fusion features to obtain an unsupervised learning model; and performing deployment and incremental learning on the unsupervised learning model, and detecting the blueberry diseases. According to the method, early-stage accurate detection of blueberry diseases is realized through an unsupervised learning algorithm, a new normal form is provided for intelligent accurate management of blueberries, and the disease prevention and control efficiency and industrial economic benefits are effectively improved.

The invention discloses an intelligent detection method for product defects on an automatic production line and a detection system based on machine vision, and relates to the technical field of industrial product quality detection. According to the method, a multi-waveband imaging technology is combined with temperature and chemical component information to generate a multi-dimensional feature data set, and an unsupervised learning algorithm is utilized to perform clustering analysis, so that the limitation of traditional single-waveband imaging is broken through, product features are comprehensively captured, the defect detection range and accuracy are remarkably improved, and a foundation is laid for subsequent analysis and classification; further utilizing an attention mechanism and a deep learning technology to accurately position and classify defects, and triggering deep scanning through a priority index to improve the detection precision and the system adaptability; and finally, combining with a random forest algorithm to analyze defect influence, and feeding back and optimizing production parameters through a dynamic adjustment mechanism, thereby realizing production optimization closed-loop management, and improving production efficiency and product quality detection.

The invention discloses a defect classification and segmentation method and system based on unsupervised and weak supervised combination, and the method comprises the steps: employing a training data set only containing a defect-free sample, extracting the feature representation of the defect-free sample through a feature extractor, and storing the feature representation in a feature memory library; comparing the mixed data set without labels including defect and defect-free samples with the features in the feature memory bank to determine false classification labels; designing a class activation graph-based weak supervision network for extracting features in the image, a semantic segmentation network and a loss function thereof, and training the semantic segmentation network by using a training data set of a defect-free sample and pseudo-classification label data; according to the method, the advantages of weak supervision and unsupervised learning algorithms are combined, and efficient defect detection and positioning can be realized.

The invention discloses a deep learning-based lung cancer early screening system, and particularly relates to the technical field of medical assistance, which comprises an intelligent data ecological module, an intelligent analysis and diagnosis module, an intelligent decision support module and a user interaction management module. According to the invention, cross-mechanism data sharing and cooperative training are realized by using data semantic understanding and federated sharing technologies, transparency and traceability of data are ensured through a block chain technology, and in an intelligent analysis and diagnosis module, a deep learning model comprehensively analyzes a multi-modal image and identifies molecular features of cancer cells, so that the accuracy of cancer diagnosis is improved. Potential early-stage lung cancer signs are found through an unsupervised learning algorithm, individual risks are dynamically evaluated, focus development is tracked, and an intelligent decision support module combines personalized treatment suggestions, optimizes clinical pathways and multidisciplinary collaboration, provides accurate decision suggestions and assists doctors in making scientific and effective treatment schemes.

The invention discloses a teaching evaluation method and system based on artificial intelligence, and relates to the technical field of teaching evaluation, and the method comprises the steps: constructing a student learning analysis time series data set based on an LMS learning management system integration platform; constructing a dynamic student learning behavior evaluation model by using an unsupervised learning algorithm, and generating a student personalized learning portrait; based on the personalized learning portrait of the student, combining the historical score data of the student and the learning progress of the student, utilizing a deep neural network optimization model to realize dynamic real-time prediction and self-adaptive adjustment of the learning state of the student, and generating a continuously updated student learning state evaluation map; and based on the continuously updated student learning state evaluation map, analyzing the relationship between the student learning progress and the teacher teaching effect, dynamically adjusting the student learning strategy and the teaching plan, and generating an artificial intelligence teaching evaluation scheme. The method has the beneficial effects that personalized and scientific teaching evaluation and optimization are realized, and the teaching effect and the learning achievement of students are improved.

The invention provides an AI-based shipbuilding man-hour unification method and system, a medium and a terminal, and the method comprises the steps: data collection and preprocessing, feature extraction, clustering algorithm application, model generation and the like. Three-dimensional model data and attribute information of ship parts of design end production design modeling are analyzed, an unsupervised learning algorithm is used for automatic classification, the parts with similar characteristics are classified into one class, and the average working hours of the parts of the same class are statistically analyzed in combination with currently recorded part working hour information. Part working hours of similar categories are unified into one datum, part categories in actual construction working hour management are reduced, working hour data are standardized, and field production is guided.

The invention provides a tobacco leaf curing environment parameter optimization method based on computational fluid mechanics, which belongs to the technical field of tobacco leaf curing, and comprises the following steps: firstly, collecting multi-point environment data of a tobacco leaf curing room to establish an initial data set, constructing a three-dimensional grid model and setting boundary conditions; secondly, importing data into fluid mechanics simulation software, setting physical property parameters and a turbulence model, and obtaining distribution of a flow field, a temperature field and a humidity field; then, an unsupervised learning algorithm is applied to identify an environment non-uniform region, and a temperature and humidity gradient model is constructed to generate an optimization curve; then calculating the similarity of temperature and humidity curves of adjacent time sequence frames to determine a key gradient point, and establishing a tobacco leaf quality and environmental parameter multivariate correlation model; and finally, performing comprehensive optimization by applying an environmental parameter fusion optimization function, calculating an optimal baking time sequence control scheme, screening an optimal parameter combination after verification in the model, applying the optimal parameter combination to an actual baking process, monitoring physical and chemical indexes of tobacco leaves, realizing accurate control of baking environmental parameters, and facilitating stable quality of the tobacco leaves.

The invention discloses an audio signal anomaly detection method based on an isolated forest algorithm. The audio signal anomaly detection method comprises the following steps: step 1, acquiring an audio signal; 2, performing feature extraction on the audio signal, generating a feature vector, and obtaining an audio feature vector; 3, inputting the audio feature vectors into an isolated forest model, and training an isolated forest by taking the audio feature vectors as training data to obtain a trained isolated forest model; and 4, receiving a to-be-detected target audio signal, performing feature extraction on the target audio signal to obtain a target audio feature vector, and calculating an abnormal score of the target audio feature vector through the isolated forest model to judge whether the target audio signal is abnormal or not. According to the unsupervised learning algorithm based on the isolated forest algorithm, rules do not need to be set manually, only one preset threshold value for abnormal score comparison needs to be set, and other threshold values do not need to be set.

The invention discloses a gas power plant combined heat and power generation real-time optimization method based on SIS data. The method comprises the steps of obtaining real-time SIS data of a gas power plant; based on the real-time SIS data and the historical SIS data, a dynamic safe operation boundary matched with the current operation condition is dynamically determined through an unsupervised learning algorithm, and the dynamic safe operation boundary is a variable constraint set representing the safe operation limit of gas power plant equipment and changing along with operation condition parameters; and by taking the dynamic safe operation boundary as a constraint condition of optimization solution, generating an operation control instruction for controlling the operation of the cogeneration unit. The safety boundary matched with the operation condition is dynamically identified through an unsupervised learning algorithm, and real-time optimization is performed by taking the safety boundary as a constraint, so that the conservative property of a traditional fixed boundary method is effectively overcome, and the economical efficiency of cogeneration operation is remarkably improved on the premise of ensuring the safety of a unit.

The invention discloses a big data analysis-based reduction furnace feeding method, a reduction furnace feeding device, reduction furnace feeding equipment and a computer readable storage medium, and the method comprises the following steps: carrying out data cleaning processing on historical batch data to obtain cleaned batch data; performing data feature processing on the cleaned batch data to obtain feature data; performing clustering analysis on the feature data by adopting an unsupervised learning algorithm to obtain an operation mode corresponding to each batch; training the initial prediction model based on the historical batch data of each batch and the corresponding operation mode to obtain an optimal parameter prediction model; performing prediction processing on the current operation state and the identified current operation mode based on the optimal parameter prediction model to obtain current optimized charge table data; and the current optimized charge table data are sent to a reduction furnace control device, so that the reduction furnace device carries out feeding control based on the current optimized charge table. Feeding adjustment is achieved in real time, and the feeding accuracy of the reduction furnace is improved.

The invention is suitable for the technical field of sodium ion batteries, and provides a digital twinning-based intelligent operation and maintenance method for a sodium ion battery energy management system, and the method comprises the steps: constructing and operating a virtual model corresponding to a sodium ion battery system in a cloud digital twinning platform based on collected multi-source heterogeneous operation data; the virtual model simulates the internal state and evolution trend of the sodium ion battery system by coupling the electrochemical model, the thermal model and the life model; tracking and predicting the state of charge and the state of health of the sodium ion battery system, generating a charging and discharging power instruction and a thermal management strategy through a dynamic optimization control algorithm, and issuing and executing the instruction; and performing anomaly detection by applying an unsupervised learning algorithm, and generating predictive maintenance suggestions in combination with the prediction result of the health state. By constructing digital twinborn body and multi-model coupling, multi-dimensional accurate perception and multi-target dynamic optimization of the sodium ion battery system are realized, and the full life cycle value and safety level of the energy storage system are improved.

The application belongs to the field of numerical simulation, and particularly relates to a modeling method of a shock wave light field in a hypersonic flow field, comprising the following steps: adopting an unsupervised learning algorithm to perform gradient clustering on flow field data, and determining the position of a shock wave and the inner and outer boundaries of a shock wave domain in the flow field domain; constructing a hybrid dimension numerical continuous and discrete strategy, determining the main direction of a light ray based on the strategy, performing continuous medium approximation processing on the main direction of the light ray, and performing structured grid discrete processing on the non-main direction of the light ray; based on the analogy of Fermat's principle and Maupertuis' principle, the transmission of the light ray in the refractive index field is analogized to the movement of equivalent real particles in the refractive index potential field, and the dynamic relationship between the equivalent real particles and the refractive index potential field is established; based on the dynamic relationship, the transmission path of the light ray in the shock wave is simulated, and the modeling of the shock wave light field in the hypersonic flow field is realized. The application significantly improves the efficiency and accuracy of the modeling of the shock wave light field in the hypersonic flow field.

The present invention provides a method for mining and analyzing security abnormal behaviors in the industrial Internet of Things, comprising the following steps: collecting multi-dimensional data in real time through data collection devices installed on each node of the industrial Internet of Things; preprocessing the collected multi-dimensional data to generate a preprocessed data set; extracting features from the preprocessed data set to obtain key features reflecting the operating state of devices and network behaviors, and these key features include temporal features and frequency features; analyzing the extracted key features through an unsupervised learning algorithm to establish a baseline model for the normal operation of the industrial Internet of Things; during the actual operation of the industrial Internet of Things, inputting the extracted features into the baseline model for comparison to detect abnormal behaviors deviating from the baseline model. This method effectively improves the detection accuracy and response speed of security abnormal behaviors in the industrial Internet of Things through multi-level data processing and analysis, and enhances the overall security of the system.

Systems and methods of recommending replacement of printheads. In an embodiment, a system trains a first neural network to generate anomaly scores for printheads using an unsupervised learning algorithm based on first training samples of conforming printhead data from a pool of conforming printheads. The system generates a training dataset for a recurrent second neural network by identifying training printhead data for a pool of training printheads, inputting second training samples of the training printhead data into the first neural network to generate training anomaly scores for the training printheads over a plurality of time units, and formatting third training samples for the training printheads. The system trains the recurrent second neural network to generate scaled anomaly scores for printheads using a supervised learning algorithm based on the second training dataset.

The application provides a data processing method, device, equipment and medium. Data of multiple heterogeneous systems is collected multiple times, and the data of the multiple heterogeneous systems collected each time is transmitted to an analysis system respectively. Each analysis system runs different unsupervised learning algorithms and outputs corresponding analysis results. Each analysis result includes a first interworking heterogeneous system matched by each heterogeneous system. Then, each analysis result is integrated into a classification matrix. Rows of the classification matrix include matching results of each heterogeneous system under different analysis systems, and columns of the classification matrix include matching results of each heterogeneous system under each analysis system. Secondary analysis is performed on data in each row of the classification matrix to analyze analysis results of each heterogeneous system under different analysis systems, and a second interworking heterogeneous system matched by each heterogeneous system is obtained, so that an interworking heterogeneous system can be effectively obtained.

The application provides a kind of dry reactor sound signal visual detection method, medium and system, belong to dry reactor sound signal detection technical field, include: first, the sound signal of dry reactor is collected, and pretreatment is carried out to eliminate environmental noise.Then the time-frequency analysis is carried out to the sound signal after pretreatment, and the time-frequency spectrum is obtained.Next, adopt the way of bayesian probability inference and adaptive threshold increase, and highlight the small change in time-frequency spectrum, and obtain the increased time-frequency spectrum.Subsequently, energy distribution, peak frequency and harmonic structure are extracted from the increased time-frequency spectrum, and combined into a multi-dimensional feature vector.Apply dimension reduction algorithm, map high-dimensional feature vector to two-dimensional or three-dimensional space, and obtain the second feature vector.Finally, use unsupervised learning algorithm to carry out cluster analysis on the second feature vector, and assign color or label according to the clustering result for different categories, generate visual classification image output.

The invention discloses an intelligent accounting auditing system based on AI and a block chain, and particularly relates to the technical field of accounting auditing and computer application. The system comprises a data acquisition module, a data processing module, an AI auditing module and a block chain storage module, wherein the data acquisition module acquires original accounting data from an enterprise internal information system through a RESTful API interface and an ETL technology; the data processing module cleans the original accounting data by using a large language model and an unsupervised learning algorithm, and generates audit data in combination with a machine learning classification algorithm and an accounting data standardization rule base; the AI auditing module constructs an AI auditing model based on a graph neural network in combination with reinforcement learning, and processes an auditing graph constructed through the auditing data through the AI auditing model to generate an auditing result; and the block chain storage module adopts AES-256 to encrypt and store the original accounting data, the audit data and the audit result. According to the invention, intelligent auditing and safe storage of the accounting data are realized, and auditing efficiency and data credibility are improved.

The invention provides a wind turbine blade early damage identification method and system, and belongs to the technical field of wind turbine generator set state monitoring, and the method comprises the steps: carrying out the two-time data cleaning of the wind speed-power real-time data of an SCADA system; comparing the wind speed-power real-time data after the last cleaning with a standard wind speed-power curve to obtain an initial wind power blade damage state; and inputting the real-time acoustic emission signals into the wind power blade early damage prediction model to obtain the type and degree of early damage identification of the wind power blade, the wind power blade early damage prediction model, classifying the acoustic emission signals collected by a fatigue loading test through an improved K-means algorithm, and constructing a convolutional neural network. According to the method, the problem that an unsupervised learning algorithm excessively depends on experience to set a threshold value is avoided, the defects of poor data cleaning quality and low algorithm generalization are overcome, and the reliability, accuracy and usability of SCADA data are improved.

The invention provides a capacity management method and device. The method comprises the following steps: determining a first feature tensor according to a preset time window, business event features and environment features; wherein the first feature tensor comprises a plurality of data dimensions; inputting the first feature tensor into a hybrid neural network model, and predicting the resource demand quantity at a specific time point in the future to obtain prediction information corresponding to the resource demand quantity at the specific time point in the future; performing abnormal data identification on the real-time data flow according to an unsupervised learning algorithm to obtain abnormal information; and determining an optimal capacity management strategy according to the prediction information and the abnormal information. Through the embodiment of the invention, the problem of low capacity management efficiency caused by the fact that the traditional CMP passively responds to resource alarm, depends on manual decision and execution and cannot predict future demands or global optimization in the related technology is solved.

The invention relates to the technical field of building nondestructive testing, and discloses a building leakage point detecting, identifying and positioning method for building engineering, which comprises the following steps: firstly, adaptively identifying a resonant frequency set of a to-be-detected building structure; then, on the basis of the resonant frequency set, normal excitation and tangential excitation are sequentially applied through a composite orthogonal bimodal excitation unit, and surface micro-motion video data of the structure are synchronously collected by a high-frame-rate optical imaging unit; thirdly, resolving the video data to generate a P-wave velocity field and an S-wave velocity field, and constructing a wave velocity ratio field sensitive to leakage characteristics; and finally, multi-dimensional features are fused to construct a high-dimensional feature vector, an unsupervised learning algorithm is utilized to calculate an abnormal score, and automatic identification and accurate positioning of the leakage point are realized. According to the method, the sensitivity and the accuracy of detection are remarkably improved by utilizing a resonance enhancement mechanism and a wave velocity ratio diagnosis basis sensitive to a water medium, and automation of a whole detection process and objectification of result interpretation are realized by introducing unsupervised learning.

The invention discloses a chemical emergency hazard prediction method, device, equipment and medium, and relates to the technical field of hazard prediction, the method is combined with an unsupervised learning algorithm to assist accident analysis to complete generation of a labeled sample, and when a supervised learning algorithm is used, the labeled sample can be rapidly generated. And an integrated learning thought is adopted to complete accident hazard prediction, so that standard, scientific and accurate accident analysis and risk level evaluation are achieved.

The present invention is applicable to the field of intelligent agricultural technology and provides a method and system for intelligent blueberry disease detection based on multimodal unsupervised learning. The method comprises the following steps: based on an adaptive contrast deep clustering framework, feature extraction and clustering are performed on preprocessed multimodal data to obtain multimodal preliminary features and preliminary clustering results; based on a multimodal complementary feature fusion mechanism, the multimodal preliminary features are adaptively weighted and fused according to the multimodal preliminary features and preliminary clustering results to obtain multimodal fusion features; unsupervised clustering optimization and disease type identification are performed on the multimodal fusion features to obtain an unsupervised learning model; and the unsupervised learning model is deployed and incrementally learned to detect blueberry diseases. The present invention achieves early and accurate detection of blueberry diseases through an unsupervised learning algorithm, providing a new paradigm for intelligent and precise blueberry management, effectively improving disease prevention and control efficiency and the economic benefits of the industry.

The invention provides a shared vehicle scheduling method and system and related equipment. The method comprises the following steps: acquiring operation data and external environment data of a shared vehicle; performing preprocessing and time-space fusion on the operation data and the external environment data to obtain multi-source fusion data; based on the multi-source fusion data, a shared vehicle hot spot area is mined through an unsupervised learning algorithm; inputting the multi-source fusion data into a trained machine learning prediction model, and outputting the vehicle demand number of each station / region; in combination with the shared vehicle hot spot area and the vehicle demand number of each station / area, differential scheduling tasks of the pile stations and the pile-free areas are generated; carrying out cost quantitative evaluation on the differentiated scheduling tasks and obtaining a priority sequence of each scheduling task; and converting the sorted scheduling tasks into standardized instructions, and sending the standardized instructions to the scheduling terminals. According to the method, vehicle operation and environment data are taken into consideration, and pile-free differentiated scheduling is realized, so that a complex scheduling scene can be better coped with.

The invention belongs to the field of numerical simulation, and particularly relates to a modeling method for a shock wave light field in a hypersonic flow field, which comprises the following steps: carrying out gradient clustering on flow field data by adopting an unsupervised learning algorithm, and determining the position of a shock wave in a flow field domain and the inner and outer boundaries of a shock wave domain; constructing a mixed dimension numerical value continuity and discretization strategy, determining the main direction of the light based on the strategy, performing continuous medium approximation processing on the main direction of the light, and performing structured grid discretization processing on the non-main direction of the light; on the basis of the analogy of the Fermat principle and the Mopedor principle, transmission of light in a refractive index field is analogous to movement of equivalent physical particles in a refractive index potential energy field, and a dynamic relation between the equivalent physical particles and the refractive index potential energy field is established; based on the dynamic relationship, the transmission path of light in the shock wave is simulated, and modeling of the shock wave light field in the hypersonic flow field is achieved. According to the method, the modeling efficiency and precision of the shock wave light field of the hypersonic flow field are remarkably improved.

The invention discloses an intelligent fault diagnosis method and device for an energy storage battery, electronic equipment and a storage medium. The method comprises the following steps: continuously acquiring operation parameter time sequence data of each monomer in the battery pack; preprocessing the operation parameter time sequence data; extracting characteristic parameters used for representing the state of the battery from the preprocessed operation parameter time sequence data in the set time window to form a characteristic set; identifying and positioning abnormal monomers in the battery pack through an unsupervised learning algorithm based on the feature set, and performing abnormal identification on the abnormal monomers; inputting the operation parameter time sequence data of the abnormal monomer into a pre-trained lightweight fault diagnosis model to determine the fault category and the fault level of the abnormal monomer; and generating corresponding early warning and operation suggestions based on the fault category and the fault level. According to the invention, the early-stage accurate diagnosis of the fault of the energy storage battery, the stable adaptation of the extreme environment, the lightweight real-time response of the ground station and the intelligent operation and maintenance can be realized, and the flight safety and reliability are greatly improved.

This invention discloses a dynamic response analysis method for AC / DC hybrid power grids. The method first searches for a fault set in the near-field region of the DC landing point of the AC / DC hybrid power grid. Then, it uses electromechanical-electromagnetic transient simulation to analyze the faults, obtaining the system dynamic response curve corresponding to each fault in the fault set. An unsupervised learning algorithm is then used to cluster the system dynamic response curves and label the categories. Finally, a random forest model is trained using the power flow dynamic information of the fault and the fault information corresponding to the fault as model inputs, and the system dynamic response curves corresponding to the faults as model outputs. By inputting real-time fault information and power flow dynamic information into the trained random forest model, the category of the system dynamic response curve corresponding to the real-time fault can be output. This analysis method can deeply explore the correlation between power flow dynamic information and fault information and the dynamic response characteristics of power grid voltage and frequency, and quickly determine the voltage and frequency response characteristics of the power grid.

The invention relates to a refrigerator environment testing method, device and equipment and a storage medium. The method comprises the steps that refrigerator sample data of multiple sets of refrigerator equipment is obtained; performing data preprocessing on the refrigerator sample data to obtain optimized sample data; calculating statistical parameters of each feature in the optimized sample data, and determining a first environment test standard according to the statistical parameters; performing unsupervised cluster division on the refrigerator sample data based on a DBSCAN clustering algorithm to obtain a clustering result, and determining a second environment test standard according to the clustering result; correcting the first environment test standard according to the second environment test standard to obtain a target environment test standard; and carrying out environment testing on the to-be-tested refrigerator equipment according to the target environment testing standard. The refrigerator environment test standard is automatically formulated through the unsupervised learning algorithm and is applied to the environment test, the good product screening accuracy of the refrigerator can be effectively improved, and the production efficiency is improved.

This invention provides an OPC model training method, system, and terminal based on unsupervised learning. It constructs a simplified OPC training dataset by performing multi-dimensional lithographic feature extraction, feature preprocessing, clustering, classification, and hierarchical filtering on multiple lithographic mask test patterns. An OPC model is then trained based on this dataset to perform optical proximity correction on the lithographic mask patterns. This invention combines multi-dimensional lithographic characteristics to accurately cover different pattern scenarios, avoiding the omission of key samples and ensuring the accuracy of OPC model correction. It employs an unsupervised learning algorithm to lightweight sample data and classifies differentiated pattern categories through clustering, effectively improving the process coverage of the OPC model. It fully adapts to the changing lithography process requirements under advanced process nodes and requires no customized annotation rules, allowing for rapid migration to various lithography processes. It combines high precision, full coverage, and strong versatility.

The present invention discloses a method for optimizing energy consumption of a refrigeration unit based on an unsupervised learning algorithm, comprising: collecting operation and energy consumption parameters of equipment in the refrigeration unit; using the data in the refrigeration unit to perform real-time equipment energy consumption feature extraction using an unsupervised learning center, wherein the real-time equipment energy consumption feature extraction performed by the unsupervised learning center includes the operation status information of the refrigeration unit; using the unsupervised learning center to perform real-time equipment energy consumption optimization efficiency feature extraction; calculating whether the optimization efficiency of the real-time equipment energy consumption feature extraction performed by the unsupervised learning center is greater than a preset energy consumption threshold; when the optimization efficiency of the real-time equipment energy consumption feature extraction performed by the unsupervised learning center is calculated to be greater than the preset energy consumption threshold, the real-time equipment energy consumption feature extraction performed by the unsupervised learning center is calculated separately to the data control platform of each refrigeration unit module. The present invention can improve the accuracy of calculation and the optimization efficiency of refrigeration unit data.