68 results about "Incremental learning algorithm" patented technology

The invention relates to a thermal power plant fault early warning and diagnosis method based on a nebula system, and the method comprises the steps: collecting the multi-dimensional operation time sequence data of a thermal power plant, carrying out the feature extraction and lexical element processing of the multi-dimensional operation time sequence data through an encoder, and obtaining a unified equipment state lexical element sequence; based on the equipment state lexical element sequence, constructing a dynamic star map representing the operation state of the whole power plant; inputting the dynamic star map into a space-time fusion backbone network; the space-time fusion backbone network performs iterative processing on the dynamic star map and generates a health degree attenuation trajectory; when the slope of the health degree attenuation trajectory exceeds a preset threshold value, dynamic early warning and system diagnosis are triggered, and a natural language diagnosis report containing a causal reasoning chain is generated; new multi-dimensional operation time sequence data are collected in real time, and an incremental learning algorithm is used to update the encoder and the space-time fusion backbone network online; compared with the prior art, the system can continuously adapt to working condition changes and has high self-optimization capacity.

The invention discloses a water body apparent spectrum synchronous acquisition and water quality parameter inversion system based on an unmanned ship, and relates to the technical field of intelligent sensing, and the system comprises the following steps: acquiring environmental parameters of the unmanned ship, constructing a deep reinforcement learning model, inputting the environmental parameters, and outputting a phase control instruction; constructing a water quality parameter inversion model, inputting noise-free spectrum data, and outputting water quality parameters; when the water quality turbidity in the water quality parameters exceeds a dynamic turbidity sudden change threshold value, triggering a laboratory to collect water quality verification data, comparing the water quality verification data with the water quality parameters to output a turbidity error, and when the turbidity error exceeds an inversion verification error threshold value, updating the weight of the water quality parameter inversion model; and performing spatial interpolation calculation based on the updated water quality parameter inversion model, outputting water quality parameter distribution, and displaying a water quality parameter thermodynamic diagram and a traceability pollution path through an electronic map. According to the method, the PLSR weight is updated online through the incremental learning algorithm, and the model adaptation speed is increased.

The invention provides a three-coordinate measuring machine adaptive dynamic error compensation method, and relates to the field of three-coordinate measuring machines, and the method comprises the steps: obtaining multi-source state data of a three-coordinate measuring machine, and calculating a real-time dynamic error, the state data comprising a motion state, a dynamic response and environment disturbance data; based on the real-time dynamic error, using a recurrent neural network to construct a virtual measuring machine model, and performing offline training to obtain an initial error prediction model; and acquiring real-time error feedback data, and performing fine adjustment on the initial error prediction model by using an incremental learning algorithm and a sliding time window mechanism to obtain a prediction model capable of dynamically evolving and aging adaptive parameters. The method is used for overcoming the defect that in the prior art, a linear model or a fixed compensation parameter is difficult to accurately describe and compensate all dynamic errors sometimes.

The invention provides a centralized heating optimal scheduling method and system based on multi-heat-source dynamic matching, and the method comprises the steps: constructing a thermal load prediction model, and fusing various types of data to generate a thermal load demand dynamic curve in the next 24 hours; establishing a dynamic priority ranking model, and generating a heat source priority sequence; generating a multi-heat-source cooperative operation strategy; constructing a multi-heat-source dynamic matching optimization model to obtain an optimal scheduling strategy; heat sources are adjusted in real time through the central control system, and dynamic correction is conducted by combining pipe network feedback data; and periodically updating the thermal load prediction and dynamic priority ranking model. According to the method, a bidirectional LSTM neural network is fused with data to construct a high-precision thermal load prediction model, and a curve is generated in combination with a feature pyramid network; meanwhile, related data are periodically combined, the weight of the LSTM neural network is updated through an incremental learning algorithm, the weight of the priority ranking model is adjusted through a gradient descent method, the adaptability of the prediction model is continuously optimized, and the prediction precision is improved.

The invention discloses a physical examination suggestion intelligent adaptation and recommendation system based on a personalized health portrait. The method aims to solve the problems of static portraits, general suggestions, lack of continuous tracking, insufficient safety and the like in existing health management services. Through real-time acquisition and fusion of multi-source heterogeneous data, the incremental learning algorithm is constructed and adopted to dynamically update the health portrait of the user, and the timeliness of the portrait is ensured. Based on the dynamic portrait, the system carries out deep traceability and quantitative attribution on abnormal indexes through a multi-path reasoning decision tree, and carries out comprehensive health risk assessment. In a recommendation stage, the system comprehensively considers a user portrait, a risk assessment result, cost and feasibility preference, and performs strict medical suggestion conflict resolution by using a medical knowledge graph based on an OWL ontology and an SWRL rule, so that a highly personalized, safe and feasible physical examination suggestion and health management scheme is generated.

The invention relates to an intelligent power grid load demand dynamic monitoring and evaluation method, and the core scheme of the method comprises the steps: carrying out the data synchronous collection and structural processing of a multi-protocol sensing network, carrying out the normalization and anomaly filtering to enhance the data quality, and analyzing the influence of a dynamic quantification external factor on a load through a sliding window and a correlation coefficient. Dimensionality reduction and dynamic weight adjustment are carried out through principal component analysis, and self-adaptive modeling of multiple factors on load changes is achieved. The incremental learning algorithm supports real-time optimization and compression of model parameters, the online updating efficiency is improved, model self-correction is achieved in combination with a prediction deviation feedback and calibration mechanism, and follow-up initialization optimization is supported through a knowledge base. According to the method, the response capability of the load prediction model to external environment change, the prediction precision and the stability of engineering application are remarkably improved.

The invention discloses a complete-cycle health maintenance method, system and device for secondary circuit equipment and a medium, and the method comprises the steps: obtaining the operation data of the secondary circuit equipment, and dividing the operation data into an initial data set and an incremental data set; constructing an initial health prediction model by using the initial data set, and predicting the residual service life of the secondary circuit equipment; according to an incremental learning algorithm, updating the initial health prediction model by using an incremental data set to obtain an updated health prediction model; and performing health state evaluation and residual service life prediction on the secondary circuit equipment by using the updated health prediction model, thereby realizing complete-cycle health maintenance of the secondary circuit equipment, automatically adjusting an incremental learning strategy according to the change of an equipment operation environment, and improving the adaptability of the system.

The invention discloses a bedridden patient sign measurement and medication assisting system based on multi-sensor fusion, which belongs to the technical field of medical instruments and comprises a multi-sensor fusion measurement module, a self-adaptive calibration compensation module, an intelligent evaluation decision module, a medication assisting calculation module and a closed-loop feedback optimization module. A pressure sensor array, an ultrasonic sensor and an acceleration sensor are used for cooperatively measuring the height and weight of a patient, a deformation compensation model and a body position correction model are adopted for accurate calibration, a BMI index is automatically calculated, a VTE risk score, a falling risk score and a self-care ability score are generated, and the medication dosage is intelligently recommended according to medicine characteristics and patient physique. And continuously optimizing system parameters through an incremental learning algorithm to form a measurement-evaluation-medication-optimization deep coupling closed-loop system, so that the physical sign measurement precision, clinical evaluation efficiency and medication safety of the bedridden patient are remarkably improved, and the nursing labor intensity and the medical error risk are reduced.

The invention provides a power grid anti-bird multichannel twitter sound source positioning and intervention method, which comprises the following steps: acquiring sound signals from a power grid area through a multichannel microphone array equipped with a laser radar calibration module, optimizing a separation threshold by adopting Fourier transform and combining a preset bird voiceprint feature library, separating twitter components and noise components, and performing interference on the twitter components and the noise components; pure birdsong signals are obtained; according to the pure birdsong signal, calculating the time difference between channels by adopting a time delay estimation method, dynamically correcting a preset threshold value by combining a real-time environment sensor, if the time difference exceeds the corrected threshold value, marking as an effective birdsong event, and adopting a centimeter-level differential GPS module to assist in determining the position coordinates of the birds; interference execution confirmation is received through an edge computing node and returned to a system log, a D-S evidence theory data fusion method is adopted to integrate the positioning coordinates and the risk level, if it is judged that the fusion result is consistent in a preset confidence interval, the bird damage monitoring model is updated based on an incremental learning algorithm, and real-time closed-loop feedback is obtained.

The invention provides an automobile foot mat multi-style mixed flow production scheduling method based on customer order data, and the method comprises the steps: carrying out the standardization and feature vector generation of order process parameters, combining with the real-time state collection of production equipment, building a static process similarity and dynamic equipment response cost evaluation model, and fusing multi-source data to analyze the dynamic switching cost. The method comprises the following steps: establishing a scheduling sequence, realizing scheduling sequence optimization by using an improved NSGA-II multi-target genetic algorithm, comprehensively considering switching cost, equipment utilization rate and production line rhythm, finally dynamically selecting an optimal scheduling scheme through a TOPSIS decision method and a workshop load rate, and continuously correcting and evaluating model parameters by using an online incremental learning algorithm, thereby realizing scheduling sequence optimization. According to the method, the production scheduling efficiency and the resource utilization rate are improved, the equipment switching loss is remarkably reduced, and flexible production self-adaptive optimization is realized.

The invention discloses a photographing, meter reading and power saving method based on AI, and belongs to the technical field of meter reading, the photographing, meter reading and power saving method comprises the following steps: deploying an embedded meter reading robot at a water meter end, automatically photographing a water meter reading picture, and uploading the collected picture to a cloud end; deploying an AI reckoning module at the cloud end, and reckoning a predicted meter reading value interval of the next meter reading period; the meter reading robot accurately shoots a water meter dial plate according to a preset meter reading period, a local data comparison engine is called, and deviation verification is carried out on current meter reading data and a locally cached predicted value interval; when the meter reading data is consistent with the predicted value, the meter reading robot does not upload the collected meter reading picture to the cloud, and when the meter reading data is not consistent with the predicted value, the current photographed picture is uploaded to a cloud data storage library; the cloud AI calculation module is dynamically updated based on an incremental learning algorithm; the photographing, meter reading and power saving method realizes a power saving target through algorithm optimization and cloud collaboration, and is high in practicability and wide in application range.

This invention belongs to the field of campus management and discloses a method for monitoring campus personnel behavior trajectories based on multi-source data fusion, including the following steps: acquiring multi-source data within the campus; preprocessing the multi-source data using a data weighted fusion algorithm; performing differentiated scene optimization processing on the collected facial images to output clear facial image frames and effective feature data; constructing real-time behavior trajectories of campus personnel based on a temporal modeling model using clear facial image frames, effective feature data, and preprocessed multi-source data; analyzing the real-time behavior trajectories through an anomaly detection mechanism to determine the trajectory anomaly status and classify the anomaly level; executing multi-channel alarm push according to the anomaly level; visually restoring the real-time behavior trajectories and historical behavior trajectories based on a campus spatial model; and absorbing new data and anomaly judgment results through an incremental learning algorithm to update the parameters of the temporal modeling model, anomaly detection mechanism, and differentiated scene optimization processing related models.

The application relates to the field of display voice interaction, in particular to a display voice interaction system. Voice data in a real-time environment is acquired through a microphone array, the voice data is converted into text data by using an improved RNN-T voice recognition model, a timestamp and a confidence level are marked, and an initial operation instruction is obtained; historical dialogue data is acquired, the initial operation instruction and the historical dialogue data are associated, a context weight is dynamically adjusted through an incremental learning algorithm, a display is taken as a master node, a semantic protocol is established with smart home equipment, the target operation instruction is analyzed, is distributed to associated equipment and is synchronized with a context; and feedback logs in the display and the smart home equipment are analyzed. Settings of the equipment can be automatically adjusted, the display and the equipment are individually and intelligently controlled, and the use experience of a user is improved.

This invention relates to a self-learning method and system for home environment regulation based on multimodal feedback and scene awareness. The method collects the user's physiological characteristics and body movement signals through non-contact sensing devices to identify the user's current activity scene. The system integrates the user's explicit intervention operations with implicit comfort feedback based on physiological stability, dynamically assigning fusion weights for explicit and implicit feedback according to the activity scene to generate target feedback labels. An online incremental learning algorithm is used to update the local environmental control model, and a privacy-preserving model aggregation is performed in the cloud through a federated learning mechanism. This invention solves the problems of traditional home control relying on single commands and lacking physiological feedback mechanisms, achieving intelligent closed-loop control that balances user privacy and personalized comfort.

This invention discloses a system for simultaneous acquisition of apparent spectra and inversion of water quality parameters based on an unmanned surface vessel (USV), belonging to the field of intelligent sensing technology. The system includes: acquiring environmental parameters from the USV; constructing a deep reinforcement learning (PLSR) model and inputting the environmental parameters, outputting phase control commands; constructing a water quality parameter inversion model and inputting noise-free spectral data, outputting water quality parameters; when the turbidity in the water quality parameters exceeds a dynamic turbidity mutation threshold, triggering the acquisition of laboratory water quality verification data and comparing it with the water quality parameters to output turbidity error; when the turbidity error exceeds an inversion verification error threshold, updating the weights of the water quality parameter inversion model; performing spatial interpolation calculations based on the updated water quality parameter inversion model, outputting the water quality parameter distribution, and displaying a water quality parameter heatmap and tracing pollution source paths via an electronic map. This invention uses an incremental learning algorithm to update the PLSR weights online, improving the model's adaptation speed.

The application relates to the technical field of work order management, in particular to a work order closed-loop management system based on PDCA-SDCA double circulation, which comprises a data acquisition module, a hidden danger analysis unit, a rule reasoning module, a work order distribution module, a verification feedback module and a knowledge optimization module. Through the structured feature deconstruction algorithm of the 5M1E, i.e. man (Man), machine (Machine), material (Material), method (Method), environment (Environment) and measurement (Measurement), and the safety rule knowledge base of graph structure storage, accurate identification and classification of hidden dangers are realized; the unmanned aerial vehicle rephotographing and image comparison technology is used to complete automatic verification of rectification effect; and the safety rule knowledge base is dynamically optimized through an incremental learning algorithm, for example, an online random forest algorithm is used to realize dynamic updating of the knowledge base. The application can improve the comprehensiveness and accuracy of hidden danger identification, ensure dynamic adjustment and closed-loop management of rectification task priority, and effectively improve the efficiency and reliability of safety management of a railway engineering construction site.

The invention relates to the technical field of model predictive control, and provides a decomposing furnace outlet temperature model predictive control method comprising the following steps: S1, carrying out data standardization preprocessing on a cement plant data set; s2, performing principal component analysis processing on the preprocessed cement plant data, and reducing the number of obtained observation variables from K to F; s3, training a width learning network model, if the verified R2 decision coefficient is greater than a preset value, keeping the trained width learning network model and then ending training, otherwise, training by using an incremental learning algorithm; s4, taking the predicted output of the model as a predicted value of the outlet temperature of the decomposition furnace in the future, and constructing a model prediction control MPC optimization problem; s5, solving a model predictive control MPC optimization problem by adopting a rolling optimization strategy; and S6, inputting the optimal control quantity into an actual control system of the decomposing furnace, executing corresponding adjustment operation, and obtaining newest observation data again at the next sampling moment. And a reliable technical support is provided for daily operation regulation and control and production decision making of a cement plant.

This invention discloses a system and method for psychotherapy that integrates virtual reality scenes, belonging to the interdisciplinary field of psychotherapy and virtual reality. It addresses the problems of fixed scenes, delayed adjustment, weak interactivity, and lack of cultural adaptability in traditional psychotherapy. The method includes using non-contact biosensors to capture facial blood oxygen fluctuations, combining this with a hemodynamic model to remove artifacts, and then quantifying these fluctuations into emotional energy values ​​based on an individual emotional sample set and an incremental learning algorithm to generate scene seeds containing parameters such as color temperature. A virtual physics engine is then driven to reconstruct elements such as light and shadow, and fluids, forming a physically-level dynamic scene. Inertial sensors capture user movements and convert them into virtual pressure gradients, constructing a coupled feedback loop between movement and scene. Psychological trends are predicted through hierarchical Fourier analysis of the seed sequence. This invention achieves real-time adaptation and proactive intervention between psychological states and virtual scenes, enhancing user immersion and cultural and emotional resonance, and improving the accuracy and effectiveness of psychotherapy.

The invention discloses a dynamic course generation and self-adaptive teaching system based on multi-modal AI, and relates to the technical field of intelligent education, the system comprises a multi-modal data acquisition module, a multi-modal data alignment fusion module, a dynamic learner portrait module, a self-adaptive course generation module and a course push module, and all the modules cooperate to form a closed loop mechanism; through the multi-modal data alignment fusion module, a cross-modal alignment algorithm fusing contrast learning and an attention mechanism is adopted, accurate mapping and deep fusion of multi-modal data in a feature space are achieved, the problems of multi-modal data sparsity and semantic gaps in a traditional technology are solved, high-quality data support is provided for learner state analysis, and the learning efficiency is improved. The dynamic learner portrait module adopts an incremental learning algorithm to realize real-time iterative updating of the learner portrait, so that the problem of updating lag of the traditional portrait is solved, and the portrait is ensured to be highly matched with the current state of the learner.

The present application relates to the technical field of medical devices, and discloses an intestinal fluid intermittent return method and system based on pressure self-adaptation, wherein multi-modal sensor network is used to collect patient situation state, physiological rhythm and intestinal pressure data to obtain multi-dimensional monitoring data flow; time series signal processing technology is used to extract the corresponding phases of each data and calculate synchronization metrics; a synchronous state classification model is used for collaborative state classification; based on the classification result, a return decision instruction is generated to control the return device and adjust the return parameters according to the intestinal pressure; in addition, the present application also uses an incremental learning algorithm to adjust the model threshold and decision parameters based on individualized data sets, which can accurately control the return timing and parameters, and improve the return effect and safety of intestinal fluid.

The invention provides an AI engine user demand analysis method for multi-source data input, relates to the field of artificial intelligence, and solves the technical problems that in the prior art, relevance and dynamic insight of user demands are insufficient, and user motivation is difficult to mine. The method comprises the following steps: acquiring multi-modal data from different data sources, and carrying out preprocessing and semantic fusion to generate a multi-modal data stream with unified representation; wherein the data source comprises a public network data source and an Internet of Things data source; based on the multi-modal data flow, through a graph neural network, constructing a dynamic demand graph with demand entities as nodes and association relationships as edges; based on the dynamic demand graph, identifying user demands and user motivations by using graph reasoning and causal discovery algorithms, and generating a structured demand analysis report; and based on real-time monitoring data, updating the dynamic demand map through an incremental learning algorithm, and carrying out continuous tracking of demand evolution.

The invention belongs to the technical field of tourism e-commerce, and particularly relates to an artificial intelligence-based tourism e-commerce big data mining method, which comprises the following steps of: obtaining user behaviors, cross-platform product supply, historical transaction characteristics and social public opinion emotion, and constructing a tourism interest dynamic evolution graph; quantizing a time sequence drift and community diffusion rule of user interests through a time sequence diagram neural network model, constructing a supply-demand matching degree prediction model based on cross-platform product supply and user behavior data, and combining with the updated evolution graph and through a generative adversarial network to obtain a supply-demand matching degree prediction model. And generating a personalized tourism product recommendation list and a dynamic pricing strategy, carrying out intelligent diversion and resource pre-distribution, synchronously collecting user interaction feedback and market conversion efficiency data, and through an online incremental learning algorithm, optimizing a time sequence diagram neural network model, and generating a tourism demand mining and market response evaluation report. Therefore, the problems of insufficient user interest dynamic feature capture, weak user demand mining ability and the like in the prior art are solved.

This invention discloses a chip wire bonding inspection method, apparatus, electronic device, and storage medium, relating to the field of wire bonding inspection technology. It includes employing a hardware-triggered synchronization mechanism to connect a high-resolution optical microscope, X-ray computed tomography, and ultrasonic imaging system to acquire multimodal images, and integrating a high-precision vibration sensor to collect environmental vibration data, dynamically correcting the blur of the multimodal images; using a convolutional neural network and attention mechanism to extract multi-scale features from each modality of data, employing a cross-modal Transformer architecture for feature fusion, dynamically weighting the feature contributions of different modalities through a self-attention mechanism to highlight defect-related signals, introducing a graph neural network to model the wire bonding topology, capturing the spatial relationships between wire bondings, and generating a unified defect feature map; constructing an integrated model based on a gradient boosting decision tree and an online learning mechanism, and introducing an incremental learning algorithm.

The invention discloses an employment service knowledge graph intelligent recommendation system, and relates to the technical field of employment service and artificial intelligence, and the system comprises a multi-source data fusion module which is used for integrating a resume library, a recruitment website, an industry report and social media data; the dynamic knowledge graph construction module comprises an ontology modeling sub-module and a real-time updating sub-module; the user portrait engine module supports multi-dimensional feature extraction and dynamic weight adjustment; the intelligent recommendation engine module adopts a mixed recommendation algorithm to combine knowledge graph reasoning and collaborative filtering; and the visual interaction module is used for providing a force-oriented graph to display the position relation network. The method has the advantages that real-time updating of graph node attributes is achieved through the incremental learning algorithm, and the problem of data lag of a traditional system is solved. For example, when an enterprise newly adds a'meta universe architecturist 'post, the system automatically identifies skills (Unity3D, 3D modeling) required by the post and updates the atlas.

An incremental learning-based optimization control method and system for modified energy systems includes: acquiring on-site data from the original energy system and processing it using multiple data processing methods to obtain an original energy system dataset; training a pre-selected data model using the normalized original energy system dataset, and selecting a system energy consumption and temperature model by comparing the fitting effects of different types of data models; for the modified new energy system, using an incremental learning algorithm to form a new energy system energy consumption and temperature model applicable to both old and new knowledge; and using the new energy system energy consumption and temperature model, according to different optimization objectives, combining appropriate optimization algorithms to solve for the optimal operating mode, operating state, and optimal system control parameters under different operating conditions to achieve energy saving. This invention utilizes the forward transfer of knowledge from the old energy system to quickly form a new energy system model, avoiding the problem of catastrophic forgetting.

The invention discloses a numerical mode initialization efficiency evaluation method and system based on sea wind meteorological data, and belongs to the technical field of real-time meteorology. The method comprises the steps that intermediate data in the numerical mode initialization process and short-time forecast data based on an initial field are acquired, and the intermediate data and the short-time forecast data are used for generating a lightweight performance index data stream in real time through a recursive calculation rule; the lightweight performance index data stream comprises a real-time deviation coefficient for quantifying the deviation degree of an initial field and observation data, and a consistency score for evaluating the time-space coordination of the multi-source data; carrying out continuous learning by adopting an online incremental learning algorithm based on the lightweight performance index data stream, dynamically updating parameters of the performance evaluation model, and outputting a dynamic evaluation result of the initialization performance of the current numerical mode; and constructing the dynamic evaluation result into a time sequence, inputting the time sequence into a long-short-term memory network model fused with an attention mechanism, and outputting a numerical mode initialization efficiency attenuation probability in a specific time window in the future.

The invention relates to the technical field of air combat intelligent aid decision making, in particular to an aid decision making recommendation model adaptive method and system based on a Bayesian neural network, and is particularly suitable for man-machine collaborative decision making in a 2V2 air combat game scene. By designing a self-adaptive framework and an incremental learning algorithm, the model can be dynamically updated based on real-time data in the use process of a user, so that the recommendation accuracy and robustness are improved. The method comprises the following steps: constructing an incremental learning algorithm of a Bayesian neural network, analyzing the time complexity and the space complexity of the incremental learning algorithm, evaluating the real-time performance of the incremental learning algorithm in an online reasoning process, and researching the robustness performance under the condition of data uncertainty. The method can be widely applied to strategy selection in an intelligent recommendation system, an auxiliary decision-making platform and a multi-agent game.

The invention discloses an industrial data mining and interaction method and system, and the method comprises the steps: firstly obtaining original data from industrial data, carrying out the preprocessing, classification and rating, obtaining credit evaluation data, carrying out the standardization of the credit evaluation data, constructing a knowledge graph model, forming an initial knowledge base, and achieving the timing updating of the knowledge base through an incremental learning algorithm; constructing an optimization prediction model to predict the original data to obtain prediction data, and calculating an optimization reward value through a multi-task confidence scoring formula and a demand quality scoring formula based on the prediction data; and finally, according to the optimized reward value and the updated knowledge base, constructing a feedback understanding model in combination with similarity calculation and a context enhancement principle, obtaining an optimized understanding model through meta-parameter optimization, generating user understanding data for controlling industrial system equipment based on the model, and completing mining and interaction of industrial data. The efficiency and accuracy of industrial data mining and interaction are improved.

The invention relates to a ponding detection method and system. The method comprises the following steps: acquiring an incremental training data set of a ponding scene of a target area; adopting an incremental learning algorithm based on regularization, and utilizing the incremental training data set to update and train a pre-trained ponding detection model to obtain an updated and trained ponding detection model; and inputting the to-be-detected image of the target area into the updated and trained ponding detection model to obtain a corresponding ponding detection result. Compared with the prior art, the incremental learning algorithm is adopted to update and train the model, the model can be updated without total retraining, and the dynamic change characteristics of the ponding scene are adapted; meanwhile, the size of the model is not remarkably increased, and the lightweight deployment requirement in a ponding detection scene is met; on the premise that old knowledge is not forgotten, the model flexibly learns new tasks, and the plasticity and stability of the model are balanced; the model after training can detect the ponding area more accurately and effectively.

The application provides an electric remelting smelting endpoint element dynamic prediction and process decision optimization method and system, the method comprises the following steps: obtaining historical data sets and real-time data streams in the smelting process, and preprocessing the data; introducing an online stochastic gradient descent and incremental learning algorithm, constructing an element dynamic prediction model, receiving sensor data in the smelting process in real time, and dynamically optimizing the element concentration prediction model; based on the deep deterministic policy gradient algorithm, an optimization decision model is constructed, the element composition, smelting conditions and other environmental parameters predicted by the element dynamic prediction model are used to adjust the smelting operation strategy in real time; a real-time feedback mechanism is set, so that the element dynamic prediction model and the optimization decision model work together to realize the real-time synchronous adjustment of the element concentration prediction and the control strategy. The application can adjust the smelting parameters in real time, improve the element prediction accuracy, reduce the energy consumption, and reduce the manual intervention, and significantly improve the smelting efficiency and quality stability.