110 results about "Online model" patented technology

The invention relates to a weak supervision online video moment positioning method and system based on memory perception, and belongs to the technical field of artificial intelligence, and the method comprises the steps: carrying out the multi-modal feature fusion of a given video and a text query thereof, and obtaining the unified frame level representation at each stage; inputting the fused features into an offline module and an online module in an integral and frame-by-frame manner by using an offline guide online model architecture; in the off-line module, generating a Gaussian mask to reconstruct query of a covered part of words, and obtaining a proposal of an action starting moment; in the on-line module, the long-term historical memory in the window is used for enhancing the score, the attention weight of the score in the window is dynamically generated, and the score of the current frame is calculated in a weighted mode; taking the proposal obtained by the offline module as a pseudo tag, and providing supervision information for the score sequence of the online module; and high-performance weak supervision on-line moment positioning can be completed only by independently deducing the on-line module. The expansion capability and the application value of the model are remarkably improved.

The invention discloses a water pump frequency optimization control method and system based on flow-energy consumption feedback, and the method comprises the steps: obtaining a target demand flow of a water pump and a current pipe network resistance parameter, inputting the target demand flow and the current pipe network resistance parameter into a preset energy efficiency optimization model, and obtaining an initial optimal frequency set value and a corresponding predicted energy efficiency value; the actual flow and the actual total power consumption of the water pump operating at the value are obtained, and the actual energy efficiency value in the current operating state is calculated; whether the energy efficiency error between the energy efficiency value and the predicted energy efficiency value is smaller than a preset error threshold value or not is judged; if yes, taking the initial optimal frequency set value as a target frequency set value; if not, correcting the initial optimal frequency set value according to the absolute value of the energy efficiency error to obtain a target frequency set value; and controlling the water pump to operate according to the target frequency set value. By combining real-time energy efficiency feedback, iterative closed-loop correction, multi-stage fine optimization, online model calibration and other means, the effects of remarkable energy saving, high self-adaption and quick and accurate optimization are achieved.

The invention relates to the technical field of network fault maintenance, in particular to a network fault self-healing and prediction maintenance method based on artificial intelligence, and the method comprises the steps: S1, constructing a multi-source data real-time collection framework; s2, deploying a lightweight A I model at an edge node; s3, introducing an interpretable AI technology; s4, constructing a causal reasoning module; s5, designing a dynamic self-healing strategy library; s6, establishing an online model learning mechanism; s7, developing a simulation verification environment; and S8, realizing a man-machine cooperative operation and maintenance workflow. According to the scheme, data is subjected to streaming preprocessing and intelligent dimension reduction at the source, the transmission load is greatly reduced, the processing efficiency is improved, real-time, near-real-time and batch processing tasks are further distinguished through the edge side parallel assembly line technology, it is ensured that key indexes are preferentially processed, and compression and acceleration are achieved on the model level through knowledge distillation, quantification and pruning technologies.

The invention relates to a precision agriculture monitoring system and method based on multispectral imaging. The system and method are applied to real-time monitoring of crop physiological parameters and variable fertilization decision making. The system comprises an unmanned aerial vehicle imaging module, an edge computing unit and a cloud analysis server. The unmanned aerial vehicle module is provided with a multispectral filter wheel, a three-axis holder and an RTK positioning device and is used for acquiring a high-resolution crop image; the edge calculation unit integrates a radiation correction module, an image splicing module and a canopy segmentation module to realize on-site preprocessing; the cloud server runs a deep learning model and a feature fusion mechanism, outputs estimation of parameters such as nitrogen, chlorophyll and moisture, and generates a high-resolution fertilization prescription map. In the aspect of the method, dynamic monitoring of the nitrogen content of crops is realized through route planning, data synchronization, radiation normalization, multi-source feature fusion and time sequence prediction. The system supports online updating and ground verification of the model, has high precision, low delay and large-area operation capability, and is suitable for intelligent agriculture and precise fertilization scenes.

The invention belongs to the technical field of aircraft control, particularly relates to a range extender control system for an unmanned aerial vehicle and a controller, and aims to solve the problems of limited endurance, poor flight stability and the like caused by discontinuous energy supply and power response lag. The system comprises a range extender power source module, an electric energy conversion and distribution module, a flight state sensing module, a load power prediction module, a multi-target optimization scheduling module and a closed-loop feedback execution module. Through flight state real-time perception and load power prediction based on a recurrent neural network, an optimal power generation instruction is generated in combination with multi-target optimization scheduling, and precise rotating speed tracking is realized through adaptive PID control. A fuel consumption, power supply smooth switching and battery health combined cost function is introduced, dynamic weighting is carried out according to task types, and the cruising ability and the system robustness are improved; the system has the functions of fault emergency response, power battery hot plug and model online updating, and safety and maintainability are enhanced.

The invention provides a production line parameter real-time scheduling method based on reinforcement learning, and belongs to the technical field of production lines, and the method comprises the steps: collecting sensor data to form an original state vector, reducing the dimension of the original state vector into a low-dimensional feature state vector through a state compression encoder, inputting the low-dimensional feature state vector corresponding to a microcosmic scheduling unit into a parameter decision model, and carrying out the real-time scheduling of the microcosmic scheduling unit; the attention weight coefficient of the model is determined by the product of the historical scheduling success rate, the element value of the difference degree matrix and the maximum characteristic value of the inter-stage sensitivity matrix, after a scheduling instruction is output, simulation evaluation is performed in a digital twin platform, and after the scheduling instruction passes, a real production line is issued for execution and multi-time-scale deviation indexes are collected; and calculating a comprehensive reward value according to the indexes, and storing an experience sample to an experience playback buffer area for model online update training, thereby solving the technical problem that production line parameter scheduling is difficult to consider multi-level state feature recognition and dynamic decision weight optimization at the same time.

The invention relates to the technical field of wind energy resource assessment, and discloses a wind resource assessment method based on a millimeter wave wind finding radar. The method comprises the steps of collecting wind field data of a target area through a millimeter wave wind measurement radar and performing preprocessing to generate a standardized data set; inputting the data into a prediction model to obtain a wind energy distribution characteristic graph; generating a parameter set by adopting threshold segmentation based on the feature map, and dynamically constructing an evaluation function for simulation evaluation; synchronously collecting actual wind energy data, and calculating a deviation matrix between the actual wind energy data and a predicted value to quantify an evaluation error; performing parameter correction on the prediction model by using the error to obtain an optimized model; and finally, regenerating a feature map by adopting the optimization model and completing resource evaluation. According to the method, through online model correction and dynamic evaluation function construction, closed-loop optimization and data driving in the evaluation process are realized, and the accuracy of an evaluation result and the adaptability to a specific site are improved.

The invention relates to the technical field of intelligent clamping control, in particular to an online prediction method and system for the clamping stability of a flexible manipulator, and the method specifically comprises the following steps: collecting a clamping operation signal in a simulation experiment platform through a sensor network, and constructing a data set marking the clamping instability probability; performing alignment processing on unequal-length signals by adopting dynamic time warping integrated with physical constraints; then constructing an online model including multi-modal feature adaptive extraction fusion, time sequence feature enhancement and key frame dynamic detection and stability probability prediction, and completing model training optimization by using mean square error loss and small-batch gradient descent; and finally, deploying the model to a manipulator system to realize real-time data processing and clamping instability probability online output. The method can effectively adapt to a biochemical vessel clamping scene, the time sequence signal alignment precision and the risk prediction reliability are improved, and a real-time guarantee is provided for the operation stability of the flexible manipulator.

The invention belongs to the field of real-time monitoring, particularly relates to a chemical process real-time monitoring method and system based on digital twinning, and aims to solve the technical problem that an existing method lacks an online model correction and updating mechanism. The monitoring method comprises the following steps: S1, constructing a state snapshot data set, and synchronously collecting real-time measurement data; s2, identifying a process fluctuation interval, and selecting to-be-selected state snapshots to form an optimal snapshot matrix; s3, calculating a reduced-order truncation error energy ratio, and constructing a reduced-order model based on a reduced-order basis function; s4, in the real-time monitoring stage, if the norm of the prediction residual error is smaller than a monitoring threshold value, a Kalman filtering algorithm is adopted to correct the state coefficient of the reduced-order model; otherwise, calculating the projection error of the prediction residual on each primary function in the standby primary function library; and outputting a full-order state vector reconstructed based on the corrected or updated reduced-order model. The monitoring method provided by the invention ensures the continuous effectiveness and accuracy of monitoring when the chemical process changes.

The invention relates to the technical field of motor testing, and discloses a motor dynamic characteristic compensation test method and platform based on multi-modal data, the method collects motor data through a multi-modal synchronous test system, and a processing module executes the following steps: constructing a state observer model to estimate winding temperature and loss torque; in the online test, the accuracy of the model is monitored by calculating the residual error between the model prediction and the measured value; and when the residual dynamic characteristic exceeds a threshold value, generating and applying an adaptive excitation sequence according to the residual characteristic, collecting high-resolution data to update model parameters on line, forming a self-optimization closed loop, after the test is finished, adopting a fixed interval smoothing algorithm to be combined with a final model to obtain a global optimal state track, and calculating a dynamic characteristic compensation result according to the global optimal state track. According to the invention, through adaptive excitation of residual error driving and online model updating, accurate identification and compensation of dynamic characteristics of the motor are realized, and the test automation level and accuracy are improved.

The invention relates to the technical field of traffic signal control, and discloses a signal timing optimization method and system based on vehicle-road cooperation and traffic flow prediction, and the method comprises the steps: obtaining the track and road side detection data of a network-connected vehicle, dynamically estimating the permeability through Kalman filtering, and reconstructing the full traffic state of an intersection; predicting future traffic flow parameters by using a long-short-term memory network based on the total state historical sequence; establishing an optimization model aiming at minimizing delay and queuing, and solving by adopting a deep reinforcement learning algorithm to obtain a signal timing scheme; and issuing and executing the scheme, broadcasting a green light speed guide message, and feeding back online update model parameters based on an execution result. According to the method, the problem of state perception in a low permeability environment is solved, active prediction control of traffic signals and vehicle-road collaborative closed-loop optimization are realized, and the traffic efficiency of the intersection is effectively improved.

The invention provides a gait evaluation method and system based on a human body nonlinear system analysis technology, and the method comprises the steps: collecting a gait cycle six-channel high-precision time sequence signal outputted by a wearable inertial measurement unit, eliminating noise and gait difference through wavelet threshold denoising and Z-score standardization, extracting a chaotic feature vector composed of a Lyapunov index spectrum and a Kolmogorov entropy value, and carrying out the recognition of a gait signal, a wavelet neural network is adopted to realize nonlinear mapping of chaotic features and phase-space reconstruction parameters, and a self-adaptive feedback mechanism is introduced to dynamically optimize modeling parameters, so that the accuracy and personalized matching capability of gait pattern recognition and stability evaluation are effectively improved; quantitative characterization of the gait chaos level and real-time online model optimization can be achieved, and high-robustness support is provided for rehabilitation training and exercise aided decision making.

The application relates to the technical field of system-in-package, and discloses a functional chip SIP (System in Package) system-in-package method and system. The method constructs a reduced-order discrete thermal state predictor through multi-order exponential attenuation fitting, generates a DVFS gear-noise hazard degree spectrum mapping table based on a synchronous switching noise spectrum and an ADC noise sensitivity curve, generates a multi-DVFS working condition robust grounding network topology by optimizing a narrow bridge connection structure parameter by using a sequential quadratic programming, quantizes equivalent noise interference amounts of each DVFS gear on an ADC chip, and jointly optimizes DVFS gear selection and power consumption upper limit distribution under a rolling time domain mixed integer quadratic programming framework, simultaneously performs online model correction through exponential weighted moving average and Kalman filtering, and realizes cooperative satisfaction of thermal constraints and noise constraints.

This invention discloses a method and system for inverting mechanical parameters and intelligently analyzing the stability of tunnel surrounding rock, relating to the field of intelligent construction technology for tunnels and underground engineering. The method includes: collecting tunnel monitoring and measurement data and constructing a displacement field observation matrix; constructing a physical information neural network embedded with the geotechnical mechanics control equations to invert the mechanical parameters and stress field of the surrounding rock; automatically calling the finite element kernel through a programming interface and calculating the safety factor of the surrounding rock using the strength reduction method; using evidence theory to fuse multi-source analysis results and output the stability level; and driving online model updates and support optimization through prediction-monitoring comparison verification. This invention achieves the integration of parameter inversion, automated numerical simulation, and closed-loop verification, improving the accuracy, efficiency, and intelligence level of surrounding rock stability analysis.

This invention discloses a rubidium atomic clock frequency adaptive prediction method and system based on LSTM network, belonging to the field of time and frequency technology. The invention includes the following steps: Step 1. Data acquisition; Step 2. Filtering and denoising preprocessing; Step 3. Normalization processing; Step 4. LSTM model training and deployment; Step 5. Frequency prediction; Step 6. Online model adaptation. This invention employs a Long Short-Term Memory (LSTM) network, a special type of recurrent neural network, which can effectively learn the complex time dependencies and noise patterns in rubidium atomic clock time and frequency data. The inherent gating mechanism of the LSTM model makes it adept at capturing long-term dependencies in time series, thus enabling high-precision prediction of short-term frequency fluctuations. Through an online fine-tuning mechanism, this invention allows the LSTM model to continuously adapt to the individual drift and aging characteristics of a single rubidium atomic clock, achieving precise optimization for each clock and model, significantly improving the practicality and long-term stability of the method.

The invention discloses an online model service updating method and device and electronic equipment. The method comprises the following steps: classifying a first image through a first auditing result of a target object on the first image and a second auditing result of a first large model on the first image, and respectively updating a first to-be-trained set and a first list through obtained classification results, the first list is used for recording reasons for failing to check the target object; under the condition that the accuracy rate of the first large model does not meet a first set condition, updating a first cue word of the first large model according to a second list obtained by updating to obtain a second cue word; and under the condition that the accuracy of the first large model adopting the second cue word does not meet a second set condition, training the first large model through a second to-be-trained set obtained by updating and the second cue word, and updating an online model service through the trained first large model.

This invention discloses an intelligent inverter control method based on an adaptive algorithm, belonging to the field of intelligent inverter control technology. It employs a five-step approach: multi-source temperature acquisition and environmental identification, machine learning-based electro-thermal coupling prediction, adaptive inverter parameter optimization, collaborative thermal management, and fault diagnosis and self-repair. Addressing the risk of high or low temperature failures in motors and inverters under extreme climates, it acquires real-time temperature distribution and load information, predicts temperature rise trends in advance, and proactively adjusts inverter current, voltage, and modulation strategies to achieve safe derating or torque compensation. Furthermore, it performs multi-sensor cross-validation and observer fusion when sensors drift or components age, maintaining stable system operation and efficient energy utilization, significantly improving driving performance and vehicle reliability in extreme environments. During this process, online model updates further refine fault diagnosis and self-repair, enhancing the durability and economy of the electric drive system.

This invention belongs to, but is not limited to, the field of industrial visual inspection technology, and discloses a visual inspection system and method based on automatic defect sample annotation and online model updating. It includes a multi-station parallel inspection unit, an intelligent loading and unloading unit, a product flow and rejection unit, a sample processing and model optimization unit, a data interaction unit, and a core control module. The multi-station parallel inspection unit enables multi-dimensional parallel defect detection; the intelligent loading and unloading unit completes precise loading and unloading; the product flow and rejection unit achieves graded sorting; the sample processing and model optimization unit achieves online model updating through sample screening, automatic annotation, and incremental training; the data interaction unit achieves data closure; and the core control module ensures system stability. The method includes seven steps, such as initial configuration and intelligent loading and unloading. This invention improves inspection efficiency and accuracy, reduces annotation costs, and adapts to the dynamic inspection needs of industry.

The invention discloses a central heating and refrigerating system power adjusting method, terminal equipment and a storage medium. All available time data are mapped into a hidden space. Data points located on the manifold between the points associated with the available training time data are selected so that new data can be generated that complies with a new time dependency. And the enhanced data is input into the graph structure network, so that the model prediction performance is improved. And lagging relationship representation is obtained by performing lagging relationship calculation on the time series data, future values of the time series data are predicted in an auxiliary manner, and the capturing capability of the model on the large inertia characteristic of the system is enhanced. The improved ST-GNN prediction model and an MPC optimization controller are deeply integrated to form a complete closed loop from perception, prediction, optimization, execution and feedback. A model online updating mechanism is introduced, when the prediction deviation is increased due to the change of system characteristics, the model can be quickly adjusted based on a small amount of new data, and the model performance is prevented from attenuating along with time.

The invention relates to the technical field of automatic control of valves, and discloses a self-adaptive adjusting method of an intelligent pressure balance valve, which comprises the following steps of: performing robust preprocessing on valve front negative pressure and valve position signals in real time locally in a valve control cabinet, and identifying a valve-pipe network model online by using a normalized recursive least square method; residual low-frequency components are extracted through short-time spectrum analysis, disturbance estimation is generated in a self-adaptive mode, and meanwhile statistics is conducted on historical commands and position response quantization dead zones and friction indexes; the reverse opening degree is calculated based on an online model, a final control command is synthesized by combining disturbance feed-forward and dead zone reverse compensation, amplitude limiting and abnormity reporting are implemented, and therefore self-adaptive adjustment is conducted at each intelligent pressure balance valve of a bag-type dust collector pipe network, the fluctuation of the valve front negative pressure within the set target range is minimum, and the control accuracy is improved. And meanwhile, local self-adaptive compensation of sensor drift, slow disturbance and valve friction is realized, and the system is matched with a DCS (Distributed Control System) when an abnormal alarm is given.

The present application relates to a memory-aware based weakly supervised online video moment localization method and system, belonging to the field of artificial intelligence technology, comprising: multi-modal feature fusion on a given video and its text query to obtain unified frame-level representation at each stage; using an offline-guided online model architecture, the fused features are input into the offline and online modules in the form of the whole and frame by frame; in the offline module, a Gaussian mask is generated to reconstruct the hidden part of the query, obtaining the proposal of the action starting moment; in the online module, the long-term historical memory in the window is used for enhancement, and the attention weight in the window is dynamically generated, and the score of the current frame is calculated by weighting; the proposal obtained by the offline module is used as a pseudo label to provide supervision information for the score sequence of the online module; only the online module needs to be inferred separately, and the weakly supervised online moment localization with high performance is completed. The present application significantly improves the expansion capability and application value of the model.

The application relates to the field of artificial intelligence and discloses a target recognition method and system based on deep learning, which comprises the following steps: acquiring visible light and infrared images of the same scene, generating illumination invariance features through an adversarial training network; modeling the feature sequence by using a space-time joint network to generate an initial recognition result; performing space-time verification on the initial recognition results of continuous multiple frames by using a Markov random field model to generate a calibrated recognition result; and performing online model calibration on the space-time joint network based on the correction deviation between the initial recognition result and the calibrated recognition result. The application also provides a recognition system for executing the method. Through the synergistic mechanism of multi-modal feature decoupling, space-time joint modeling and verification, and online feedback calibration, the application effectively improves the all-weather adaptability, recognition accuracy and trajectory continuity of the system in complex scenes such as illumination changes and target occlusions.

The invention discloses a channel time-frequency joint extrapolation method based on a multi-layer perceptron hybrid architecture. Aiming at the problems of high acquisition cost of actual communication scene channel data, high time consumption of model online training and difficulty in covering the actual communication scene by simulation data, the method comprises the following steps of: making a pre-training model by using a large-scale simulation channel data set to extract channel change characteristics; and carrying out online rapid adaptation according to the trace data of the actual measurement scene. Aiming at the characteristics that the time delay spread of an indoor wireless channel is relatively small and the number of antennas is relatively small, a multi-layer perceptron hybrid architecture is applied to the time-frequency domain extrapolation of a single-carrier broadband channel. The model can quickly adapt to an actual communication scene through fine tuning of a small amount of data. The method has the advantages of being high in adaptation speed, small in data dependence, high in scene migration capacity, high in calculation parallelism degree and the like, and is suitable for edge deployment and rapid adaptation to actual scenes.

This invention belongs to the field of open-pit coal mine safety monitoring and data fusion technology, and specifically provides a multi-source data fusion method for open-pit coal mines based on BP neural networks. This method includes five steps: data acquisition, preprocessing, BP neural network fusion model construction, fusion inference and early warning application, and online model updating. Data acquisition covers geological, equipment status, and environmental parameters. Preprocessing includes data cleaning, noise suppression, feature selection, and standardization. The model construction adopts an input layer-double hidden layer-output layer structure, trained using the Adam optimizer and physical constraint loss function. Fusion inference outputs equipment status or slope displacement results and provides early warnings. The model is updated incrementally on a daily / weekly basis, and a full retraining is performed when the evaluation index degrades by more than 10%. This method solves the problems of weak adaptability and limited fusion accuracy in existing algorithms, and features high accuracy, strong robustness, good real-time performance, and practical applicability. It can improve the safety monitoring level of open-pit coal mines and reduce the incidence of disasters.

The application relates to the technical field of artificial intelligence, and discloses a real-time voice order placing method and system based on multi-modal artificial intelligence, aiming to solve the problems of poor robustness, high delay and weak generalization capability of existing voice order placing technology in complex environments. The method comprises the following steps: collecting user voice and synchronously obtaining text history; performing feature extraction on multi-modal data and realizing semantic alignment through a cross-modal attention mechanism; generating a fused intention representation by dynamically weighting through a gated fusion network, and outputting a structured task label through a hierarchical intention classifier; calling a task template engine to generate a compliant order; distributing the order through a high-priority message queue with low delay and constructing a feedback loop to support online model optimization. The above scheme realizes the unification of high robustness, low delay and strong generalization capability, and significantly improves the accuracy and real-time performance of voice order placing in noisy, dialectal and high-concurrency scenarios.

This invention discloses a prediction-aided channel estimation method suitable for TDD Massive MIMO systems, including an online model training step and an aging channel prediction step. This invention uses a Latent ODE model to continuously model the channel, solving the problem of not being able to directly obtain the CSI at aging time points within a time slot, thus breaking away from the paradigm of "discrete time series prediction model + interpolation function". The online training prediction architecture proposed in this invention avoids the model aging problem caused by the high latency of complex model training by decoupling the MIMO channel using SISO. Furthermore, online training enables the prediction model to track the dynamic changes in the propagation environment in real time, thereby improving model inference performance.

The application discloses an industrial visual detection system and belongs to the technical field of industrial image processing and artificial intelligence. The system comprises a multi-source heterogeneous data fusion acquisition module, a multi-scale attention feature extraction module, a contrast distillation defect semantic analysis module and an incremental online model evolution module. Through multi-modal sensor fusion acquisition, spatial pyramid expansion residual network multi-scale feature extraction, teacher-student contrast distillation defect positioning and classification, and the combination of knowledge distillation and elastic weight solidification incremental learning strategy, the application constructs a double closed-loop collaborative mechanism with model evolution closed loop and quality control closed loop, realizes a defect detection rate of more than 99.2%, a micro defect detection rate of 97.8% and a new defect adaptation time of 4 to 8 hours.

The present application relates to an online model fine-tuning method and apparatus based on edge intelligence, and a computer device. The method comprises: receiving an online model fine-tuning request uploaded by at least one terminal node to which an edge node belongs, and on the basis of the at least one terminal node, generating an online model fine-tuning node group; on the basis of a model fine-tuning configuration file list, broadcasting model fine-tuning feature signaling to each terminal node in the online model fine-tuning node group; acquiring subscription request signaling sent by any terminal node; and transmitting a subscribed model parameter configuration file to said terminal node, such that said terminal node performs online model fine-tuning on a pre-trained model on the basis of the subscribed model parameter configuration file, so as to obtain an updated model.

The invention provides a precise reduction control method for achieving bloom continuous casting static soft reduction based on a first-level mode, and belongs to the technical field of continuous casting. The method comprises the following steps: determining the rolling reduction of each reduction roller according to the requirements of a reduction process; determining the total rolling reduction of each rolling reduction roller based on the rolling reduction of each rolling reduction roller; and each reduction roller performs static soft reduction based on the total reduction amount. According to the method, a dynamic reduction online model can be replaced, the reduction process is accurately executed, the dependence of reduction operation on a secondary model is reduced, and the maintenance workload of a reduction system is reduced. According to the method, the casting blank internal defect that the execution effect of the pressing process is uncertain and unsatisfactory due to the fact that calibration is not timely or inaccurate can be eliminated, and the execution effect of the pressing process can be completed by maintaining the minimum normal pressing action. The method can improve the production efficiency and reduce the production cost. In addition, the method is simple, convenient and easy to implement, and can be popularized, applied and implemented on similar bloom continuous casting machines in a large scale.

This invention provides a deep learning-based crowdsourced database query optimization and trusted data processing system, relating to system software and database fields. It includes a query optimization engine module, a task allocation module, a data trust assurance module, and a trusted blockchain center module. The query optimization engine module generates an optimization execution plan and enables online model evolution through asynchronous incremental training triggered on demand by a verification submodule. The task allocation module achieves accurate task distribution through graph model task modeling and iterative matching. The data trust assurance module adopts an on-chain and off-chain collaborative architecture, detects data tampering through hash consistency comparison, and completes database recovery based on time-slice replicas and legitimate transaction replay. The trusted blockchain center module provides privacy computing and automatic verification and evidence storage services for multi-stream data. This invention effectively solves the problems of low query efficiency, inaccurate task matching, and data tamper-proofing and compliance self-verification in high-concurrency environments in crowdsourced database scenarios.