34 results about "Neural network regression" patented technology

The invention provides a grounding grid intelligent operation and maintenance system based on digital twinning and computer equipment. A multi-factor coupled corrosion kinetic equation is generated based on a laboratory grounding grid steel sheet accelerated corrosion test, and the corrosion kinetic equation is used for predicting the instantaneous corrosion rate of a galvanized steel sheet and estimating the mass corrosion loss of the steel sheet within specified time; combining the corrosion kinetic equation with the simulation model to establish a steel sheet corrosion data set, and training a physical information neural network to construct a corrosion prediction physical model; training the corrosion prediction physical model by adopting a neural network regression algorithm; constructing a grounding grid digital twin model by taking the corrosion prediction physical model as a base and combining field environment monitoring data; defining a Markov decision process in the digital twin environment of the grounding grid; and based on the feedback and reward function of the digital twin model of the grounding grid, training by adopting a near-end strategy optimization algorithm to obtain an intelligent decision-making body, and continuously optimizing the decision-making of the intelligent decision-making body through an online fine tuning function.

The invention provides an artificial intelligence grading classification-regression combination prediction method for multi-class materials, and belongs to the technical field of material informatics and artificial intelligence. According to the method, a historical sample data set of material chemical composition and crystal structure information is obtained, a band gap value is divided into a plurality of intervals based on an equal width principle, and interval labels are generated. Key features are extracted, an SVM classification model is trained to predict a band gap interval, and then a multi-layer neural network regression model is utilized to predict a band gap value. If the highest confidence coefficient is higher than the threshold value, directly calling the multi-layer neural network regression model of the corresponding interval; and if the value is lower than the threshold value, multi-interval parallel regression and weighted fusion prediction are carried out. According to the artificial intelligence grading classification-regression combination prediction method for the multi-class materials, the precision and robustness of band gap prediction can be effectively improved, particularly, the performance is prominent when high and low band gap samples are distributed imbalanced, and a precise and reliable solution is provided for material band gap prediction.

The invention provides a small sample AI proxy model construction method based on sensitivity analysis and supplementary sampling, and the method comprises the steps: carrying out the global sensitivity analysis through the automatic simulation process of parameterized parts of wind power equipment in combination with test design, and removing insensitive variables, thereby guaranteeing the accuracy and standardization of a sample, avoiding the invalid consumption of irrelevant variables, and achieving the automatic simulation of the parameterized parts of the wind power equipment. Latin hypercube sampling is adopted based on key variables, a core design area is uniformly covered, the number of initial simulation times is greatly reduced, the cost is controlled, the problem that the generalization ability is poor due to uneven small sample distribution is solved, a deep neural network regression model is trained through a training set, the strong nonlinear fitting ability of the deep neural network regression model adapts to a complex mapping relation, the core law is efficiently learned, and the robustness is high. Based on initial model prediction precision and sensitivity information, samples are accurately supplemented in weak areas, samples are not increased blindly, investment is reduced, prediction blind areas are made up, errors are gradually reduced through iterative training closed-loop optimization, and finally the small sample AI proxy model meeting preset requirements is obtained.

The invention discloses a valve life test method based on multi-modal sensing and deep learning, which comprises the following steps: acquiring high-frequency original data of valve operation through a multi-modal sensing system consisting of a broadband acoustic sensor array, a vibration accelerometer and a high-precision micro-flowmeter; the method comprises the following steps: automatically extracting features sensitive to early degradation from original data by using a deep learning model fused with a one-dimensional convolutional neural network and a Transform encoder; the high-dimensional features are mapped into single health index values through neural network regression to quantify the performance degradation degree; and a historical sequence of the health indexes is learned by using a time sequence convolutional network, and a future evolution trajectory is predicted, so that the residual life of the valve is accurately calculated. Sensitive detection and non-destructive life prediction of early weak faults of the valve are achieved, and the defects that a traditional physical testing method is long in period, high in cost and incapable of early warning are overcome.

The invention discloses a solid waste raw material activity prediction method based on oxide composition characteristics, and the method comprises the steps: firstly, constructing a standard data set, obtaining oxide component data, D50 particle size data and corresponding actually measured activity indexes of silicon-aluminum solid wastes such as fly ash, slag, steel slag and coal gangue, and carrying out the normalization and abnormal value processing to form training data; secondly, training an oxide-particle size and activity nonlinear mapping model by adopting a neural network regression model, analyzing and quantifying the contribution degree of each input feature to the activity in combination with SHAP, and ensuring the interpretability of the model; and finally, inputting the oxide of the solid waste to be detected and D50 data into the trained model, and outputting an activity predicted value. According to the method, the solid waste activity can be rapidly predicted within 10 minutes, the prediction precision R2 is larger than or equal to 0.9, the relative error is smaller than or equal to 10%, the detection cost is greatly reduced, the method can be directly used for solid waste purchase front-end pre-judgment, cement-based material formula design and raw material screening, and efficient technical support is provided for solid waste resource utilization.

The invention relates to the technical field of additive manufacturing, and discloses a calibration method of a high-precision inherent strain coefficient calibration model, comprising an original inherent strain coefficient calibration model; the heat conduction correction structure changes the heat transfer rate and the heat dissipation path in the printing process; and the topological hollowed-out structure is arranged in the heat conduction correction structure, a local heat conduction path is regulated and controlled through topological form changes, and the heat history in the original inherent strain coefficient calibration model printing process is adjusted. Acquiring the thermal history of the original inherent strain coefficient calibration model in the model in the actual printing process through data acquisition; accumulating the obtained inherent strain coefficients corresponding to the thermal history in a database, supervising and training a deep neural network regression model by using the basic database, calculating a temperature field of a complex model needing stress simulation, applying the corresponding inherent strain coefficients according to the thermal history of each point of the model, and performing stress simulation on the complex model. And the refined calculation of the inherent strain method is realized.

The invention relates to a facial acupoint real-time recognition model construction and recognition method based on facial key points and a neural network regression model. The method comprises the following steps: acquiring a face video containing various head postures and expressions and with mark points attached to target acupoints, and extracting a facial image frame; the face key point detection model is used for extracting two-dimensional coordinates of key points, the target detection model is used for recognizing two-dimensional coordinates of acupuncture points, and the two-dimensional coordinates are combined into a data set after normalization. A data set is divided into a training set and a test set, and the training set is amplified by a principal component analysis method. And training a neural network regression model by using the amplified training set, inputting key point coordinates, outputting acupuncture point coordinates, obtaining an identification model by using a root mean square error of predicted and actual acupuncture point coordinates as a loss function, and evaluating by using a verification set. The framework takes a key point as input and a perceptron as a core, compared with a complex structure of empirical rule fitting or mixed features, the framework is simpler, the prediction precision is higher, and the performance reaches or is superior to the optimal level of the existing end-to-end scheme.

The application provides a SPR connection quality tracing method and system based on pruning light weight, which comprises the following steps: training a multi-output neural network regression model by using standardized SPR curves and standardized process parameters; applying local disturbance to the standardized process parameters, inputting the disturbed samples into the trained multi-output neural network regression model to obtain predicted SPR curves; training a simplified linear model based on the relationship between the disturbed samples and the predicted SPR curves to obtain a regression coefficient matrix and a bias term; training an inverse regression model by using the standardized SPR curves and the process parameters, the regression coefficient matrix and the bias term; and inputting an abnormal SPR curve detected in real time into the inverse regression model to obtain process parameters of the abnormal SPR curve. The application improves the prediction accuracy and reasoning speed of the model, increases the interpretability of the model, and ensures the accuracy and stability of the SPR connection process.

The invention discloses a high-toughness nodular cast iron design method fused with a physical information neural network, and belongs to the technical field of material design and intelligent manufacturing. The method comprises the following steps: firstly, systematically constructing a nodular cast iron data set covering chemical components, microstructures and mechanical properties, and carrying out preprocessing and feature dimension reduction, so as to construct a deep neural network regression model, and taking a physical law of nodular cast iron as a constraint condition; forming a physical information neural network through a composite loss function formed by weighting data loss and physical loss and embedding the composite loss function into a training process; and the trained physical information neural network is used as a fitness predictor of a genetic algorithm, multi-target optimization is carried out on the chemical components, and an optimal parameter combination is reversely deduced. According to the method, deep fusion of data driving and a physical mechanism is achieved, the problems that a traditional trial and error method is long in period and a pure data model is poor in extrapolation are solved, and an innovative solution is provided for rapid and accurate design of high-performance nodular cast iron.

The invention relates to a commercial vehicle energy consumption attribution and optimization method and related equipment based on a neural network and interpretability, and the method comprises the steps: extracting multi-dimensional features and energy consumption values from commercial vehicle operation data, and constructing a training data set; training a neural network regression model by taking the multi-dimensional features as training samples and the energy consumption values as labels to obtain an energy consumption prediction model; performing SHAP interpretability analysis on the energy consumption prediction model to obtain an SHAP influence result; carrying out statistics on the distribution of each feature of the multi-dimensional features in the training data set to obtain a feature distribution reference; comparing the multi-dimensional features of the target commercial vehicle with the feature distribution reference to obtain a single vehicle feature distribution offset result; determining a vehicle optimization index based on the SHAP influence result and the single vehicle feature distribution offset result; and according to the vehicle optimization index and the single vehicle feature distribution offset result, matching energy consumption optimization measures in the knowledge base, and generating a vehicle energy consumption optimization suggestion. According to the method, the transparency, individualization and engineering landing capability of energy consumption analysis are improved.

The present application relates to a fusion method and system based on interactive iteration of sensor data and satellite data. The method obtains sensor data of one or more sensor observation points and satellite data of satellite observation points in a long period, to form a sample data with short period data and long period data in the same long period, to construct a neural network regression model with short period data as input value and long period data as output value, to obtain a prediction model by using the actual obtained long period data as expected output value and adopting a back propagation algorithm to iteratively train the model. The system comprises a satellite remote sensing system, a plurality of ground sensors and a neural network system / platform, and the neural network system / platform is used to implement the method. The present application can construct a prediction model based on fusion of sensor data and satellite data, and can be used for modeling, prediction / analysis of land subsidence and other suitable occasions / purposes.

The present application belongs to the field of oil and gas exploration, and discloses a horizontal well formation pressure artificial intelligence prediction method and system, which comprises: horizontal well logging data sensitive response feature construction; original logging response feature data preprocessing; softening treatment of pressure coefficient; pressure prediction model training based on neural network regression; and horizontal well pressure coefficient prediction. The present application is a cross-border innovation of the existing artificial intelligence method in the field of oil and gas exploration. Through comparative analysis of the measured pressure coefficient and the predicted pressure coefficient, the rationality and obvious advantages of the present application can be proved. As an effective horizontal well pressure coefficient prediction method, the present application significantly improves the performance accuracy of the traditional formation pressure prediction method, makes the formation pressure prediction method widely applicable to different lithology conditions, ensures the performance stability of the formation pressure prediction method in the application of horizontal wells, and accurately solves the 'pain points' of the traditional formation pressure prediction method.

The invention provides a probabilistic neural network mineral product prediction method based on spatial awareness, and relates to the technical field of mineral product prediction, and the method comprises the steps: extracting mineral control elements from multi-source geological data as feature data, extracting mineral production area data as label data, and randomly dividing the feature data and the label data into a training set and a test set; constructing a mineral product prediction model based on the neural network, introducing a space kernel function to perform output category weighting on a neural network output layer, outputting a neural network regression coefficient and bias, and introducing Dirichlet distribution to model output; training the model by using the training set, constructing a loss function through negative logarithm likelihood of Dirichlet distribution and KL divergence regularization, and testing the trained model by using the test set; and performing mineral product prediction by using the tested model to obtain a prediction probability, and calculating the uncertainty of each sample. According to the invention, non-stationary prediction and uncertainty measurement and simulation of a predicted target space are realized.

The invention discloses a laser welding seam key point automatic detection method based on a deep neural network, and the method comprises the steps: collecting a welding seam image in real time through a binocular industrial camera, carrying out the regression of the three-dimensional coordinates of a welding seam key point through an end-to-end multi-task deep neural network, and enabling the network to comprise a residual backbone, multi-scale feature fusion, a confidence coefficient branch and a coordinate regression branch. And an attention mechanism is introduced to inhibit arc light and splashing interference. The MLLA-YOLO novel multi-layer linear attention network structure of the method integrates a multi-scale information guiding mechanism and a dynamic receptive field control capability, has an excellent cross-scale key point sensing capability, and can maintain high-precision and high-robustness key point detection performance and output a final key point track even for V-shaped welding seam structures with different geometrical shapes. According to the method, manual participation or external targets are not needed, full-automatic, real-time and high-precision detection of the weld joint key points can be achieved in a complex industrial environment, the single-frame reasoning time is smaller than or equal to 10 ms, the average absolute error is smaller than or equal to 0.05 mm, the MLLA-YOLO can achieve the average positioning error of smaller than or equal to 0.05 mm in various weld joint styles, good universality and engineering adaptability are achieved, and the method is suitable for large-scale popularization and application. And the laser welding quality control precision can be obviously improved.

The application discloses a missile guidance control integration method based on a neural network mechanism modeling, which comprises the following steps: in a continuous time system, a vertical plane missile guidance control integration model is established; according to the dynamic equation of the vertical plane missile guidance control integration, a state transition equation in a discrete time system is established; a data set for fitting the dynamic model of a neural network is established, and a neural network parameter model obtained through neural network regression calculation is used as a non-mechanism fitting model of a missile control system; and according to the neural network parameter model, a guidance control integration mechanism model used in a control cycle is obtained. The method uses the input and output data of a control process to perform self-modeling of a time-varying control system, meanwhile, the model is equivalent to a mechanism method, and the modeling result can be used for the control of a discrete time system, and the modeling principle has interpretability and reliability.

The invention discloses a rapid radio storm waterfall plot adaptive feature extraction and classification method and system based on deep learning. The method comprises the steps of obtaining waterfall plot data; outputting an estimated dispersion measure (DM) value through regression of the first neural network; waterfall plot data are input into a second neural network containing a deformation convolution layer, and the sampling point offset of a deformation convolution kernel is calculated by the estimated DM value according to the dispersion law, so that adaptive feature extraction is carried out; and finally, performing classification judgment according to the extracted features. The system comprises a DM estimation module, a DM-guided adaptive feature extraction module and a classification judgment module, and is used for realizing the method. According to the method, the physical law of the signal is embedded in the neural network structure, so that the convolution kernel receptive field can be adaptively matched with the dispersion curve form of the FRB, and the problems of low detection sensitivity and poor model generalization ability caused by mismatching of the receptive field and the signal form in the prior art are solved.

The invention discloses a hydrogen leakage diagnosis method and device based on multi-modal information combination, and relates to the technical field of industrial safety monitoring and fault diagnosis, and the method comprises the steps: determining whether hydrogen concentration data exceeds a set concentration threshold value or not; if the hydrogen concentration data exceeds a set concentration threshold value, acquiring multi-modal information data in real time; if the hydrogen concentration data does not exceed the set concentration threshold value, processing is not carried out; determining an acoustic confidence map, an optical confidence map and an infrared confidence map; determining a fusion weight parameter based on the environment information data and the memory cell bank, and further obtaining a fusion confidence map and a weighted fusion feature vector; determining a hydrogen leakage position based on the fusion confidence map; adopting a multi-output neural network regression model to obtain hydrogen leakage quantitative parameters based on the weighted fusion feature vectors; and updating the memory cell bank based on the environmental information data, the hydrogen leakage position and the hydrogen leakage quantitative parameter. The accuracy, reliability and long-term stability of hydrogen leakage diagnosis can be improved.

This invention discloses an anti-skid navigation control method for agricultural machinery based on historical driver data, comprising: S1, collecting operational data of the driver operating the agricultural machinery on slippery road sections, using the Pearson correlation coefficient method to screen key dependent variables affecting steering control, and constructing training samples; S2, constructing an attention-BP-LSTM neural network regression prediction model for desired wheel angle and steering wheel torque; S3, model training; S4, evaluating and verifying the model's prediction results; S5, deploying the model on the terminal controller of the agricultural machinery to acquire real-time operational data of the agricultural machinery. When slippage is detected, the model outputs the desired wheel angle and steering wheel torque, and, combined with the safety protection mechanism of output limiting and gradual control, automatically adjusts the steering mechanism of the agricultural machinery to control the agricultural machinery to escape slippage. This invention generates precise and real-time control commands through deep learning of the driver's historical sideslip correction data, solving the problems of traditional methods relying on physical models and lacking experience utilization.

The application provides a particle concentration prediction method and a cooking fume exhaust device, comprising: acquiring a target prediction feature for predicting a particle concentration of a target breathing area; inputting the target prediction feature into a pre-trained neural network regression model to obtain a particle concentration prediction value of the target breathing area; the neural network regression model is used to represent a nonlinear mapping relationship between the target prediction feature and the particle concentration of the target breathing area; and outputting the particle concentration prediction value. The method can predict the particle concentration of the target breathing area where the user's mouth and nose are located based on the local detection result of the preset detection position and the actual operation state of the cooking fume exhaust device, thereby avoiding the numerical distortion caused by directly displaying the local measured value of the sensor, and further improving the accuracy of the particle concentration feedback result.

The invention relates to a tactile perception positioning method and system based on a motor reducer signal. The method comprises the following steps: constructing a robot three-dimensional model by using three-dimensional modeling software, and carrying out gridding point location coverage to establish a point location coordinate system; a position signal vector and a torque signal vector of a motor speed reducer generated when a robot is touched are collected, a motor speed reducer signal data set is constructed, feature vectors are extracted and mapped to a point coordinate system, coordinate labels corresponding to the feature vectors are obtained, a neural network regression model is trained, and the robot is touched. And outputting the predicted three-dimensional coordinates of the touch point through the neural network regression model. Under the condition that no extra force / touch sensor is installed, external contact is sensed only through position signal vectors and torque signal vectors generated by a motor and a speed reducer arranged in the robot joint, the contact position is accurately positioned, the cost and complexity can be greatly reduced, the mechanical structure is simplified, and the reliability is improved.

The invention provides a training method, a prediction method and system of an energetic material gas-solid two-phase formation enthalpy prediction model, equipment and a storage medium. The training method comprises the following steps: S1, constructing a data set: respectively constructing a gas phase data set and / or a solid phase data set; s2, molecular graph structure coding; and S3, model training: based on the gas phase data set and / or the solid phase data set, independently training a gas phase diagram neural network regression model and / or a solid phase diagram neural network regression model. The prediction model obtained by the training method of the gas-solid two-phase formation enthalpy prediction model of the energetic material provided by the invention is used for predicting the gas-solid two-phase formation enthalpy of the energetic material, and the end-to-end prediction of structure-property can be accurately and reliably realized.

The application provides a real-time road safety evaluation method based on vibration sensing of a driving vehicle, a medium and equipment, first, the road to be monitored is divided into regions according to a specified size, and the road to be monitored is divided into intervals according to region attribution and road type; then a neural network regression model is established, which is used for predicting vibration difference values according to road type, temperature data, weather conditions and road surface obstacles; finally, the vibration sensing values generated by the automobile in the driving process are compared with the historical standard vibration sensing values of the automobile after a series of standardization treatments, to help judge whether abnormal conditions such as cracks, depressions and changes in the structure below the road occur. The application can effectively reduce the cost of artificial detection, discover risks in time, take countermeasures and avoid the occurrence of road accidents.

The application provides a fuel cell thermal management method based on a neural network, which comprises the following steps: operating a test fuel cell stack on a test bench, obtaining an initial data set in a time sequence, and forming a training data set and a test data set; establishing an echo state neural network regression model, training the echo state neural network regression model by using the training data set, and obtaining an echo state neural network prediction model; testing by using the test data set to verify the prediction accuracy of the model, and obtaining a trained neural network prediction model; collecting data of a fuel cell stack in a running state in real time, obtaining a prediction signal by using the trained neural network prediction model, and controlling the running state of a cooling water pump and a radiator according to the prediction signal. The application further discloses corresponding devices and systems. By implementing the application, the accuracy and efficiency of fuel cell thermal management can be improved.

The invention discloses a submerged arc welding equipment effective productivity calculation system and method based on multi-view visual identification, and belongs to the technical field of steel structure intelligent manufacturing and industrial efficiency evaluation. According to the method, a multi-view industrial camera is deployed to collect an H-shaped steel member image, geometric parameters of the H-shaped steel member image are regressed by using a pre-trained lightweight convolutional neural network, and member identity perception without physical identification is realized; machining starting and ending time is accurately recorded through a virtual detection line, and working condition data are collected from a submerged arc welding machine PLC in real time; identifying effective welding segments based on strict multi-condition screening rules (equipment operation state, parameter deviation, duration and no serious fault); calculating the theoretical weight of the member qualified in quality inspection; and finally, the accumulated qualified weight is divided by the accumulated effective welding time to obtain an effective productivity index of a standardized unit (ton / hour), and the effective productivity index is transmitted back to the MES system. According to the method, the problems that a traditional method depends on vulnerable identification, statistics is extensive, and units are disordered are solved, and accurate, automatic and comparable productivity evaluation and production management closed loop are achieved.

The invention provides a day-ahead low-carbon economic production optimization scheduling method and system for a park multi-energy complementary system, and relates to the technical field of industrial park production energy optimization scheduling. A deep neural network regression model is constructed by using the nonlinear fitting regression capability of a deep convolutional neural network; according to the model, the total output predicted value of the time-sharing multi-energy complementary system under the energy supply control strategy of each energy conversion device of the multi-energy complementary system in at least one day in the future is predicted, and nonlinear factors of input, conversion and output links in the actual energy conversion process of the multi-energy complementary system are comprehensively considered. Establishing an optimization objective function for production optimization scheduling by taking minimization of the sum of the total operation cost and the carbon transaction cost as an objective, and obtaining an optimal hourly energy supply control strategy of each energy conversion device of the multi-energy complementary system in at least one day in the future after optimization; accurate day-ahead low-carbon economic production optimization scheduling is carried out on an actual existing industrial park comprehensive energy system.

The invention relates to a thermal power generating unit operation index determination method and device, equipment, a medium and a product, relates to the technical field of operation management in the power industry, and can realize dynamic adjustment of operation indexes of a thermal power generating unit under different working conditions. The method comprises the following steps: acquiring multiple groups of real-time operation data of the thermal power generating unit under different working conditions; the multiple sets of real-time operation data are input into a dynamic reference function, actual values corresponding to each set of real-time operation data output by the dynamic reference function are obtained, and the dynamic reference function is generated based on polynomial fitting or neural network regression; according to the actual value and a preset standard value, determining a scoring result of the real-time operation data corresponding to the actual value; and according to the scoring result of each group of real-time operation data, target operation indexes of the thermal power generating unit are determined, and the target operation indexes comprise working conditions and operation data.

This invention relates to a method and system for evaluating the online reading quality of students. The method includes: Q1. During online reading, collecting data on the user's viewing distance, blink frequency, and heart rate at the same moment, and performing normalization processing to obtain normalized data on the user's viewing distance, blink frequency, and heart rate at the same moment. Then, using an ELMAN neural network regression prediction algorithm based on the sardine optimization algorithm, the method predicts the user's viewing distance, blink frequency, and heart rate at the next moment, obtaining data on the user's viewing distance, blink frequency, and heart rate at the next moment. This invention not only evaluates the user's reading quality from multiple aspects, ensuring the accuracy of the reading quality assessment, but also assesses and reminds users of their reading quality in advance, thereby improving the user's enthusiasm and stability in reading.

The application provides a kind of artificial intelligence hierarchical classification-regression combination prediction method for multi-class materials, belongs to material informatics and artificial intelligence technical field.The method obtains the historical sample data set of material chemical composition and crystal structure information, based on the principle of equal width, the band gap value is divided into multiple intervals, and the interval label is generated.By extracting key features and training SVM classification model to predict the band gap interval, then use multilayer neural network regression model to predict the band gap value.If the highest confidence is higher than the threshold, directly call the corresponding interval multilayer neural network regression model;if it is lower than the threshold, then multiple interval parallel regression and weighted fusion prediction are carried out.The application adopts the above-mentioned artificial intelligence hierarchical classification-regression combination prediction method for multi-class materials, which can effectively improve the precision and robustness of band gap prediction, especially when the high and low band gap sample distribution is uneven, and provides a precise and reliable solution for material band gap prediction.

The invention provides an SPR connection quality tracing method and system based on pruning lightweight. The method comprises the following steps: training a multi-output neural network regression model by using a standardized SPR curve and standardized process parameters; local disturbance is applied to the standardized process parameters, disturbance samples are input into the trained multi-output neural network regression model, and a predicted SPR curve is obtained; training the simplified linear model based on the relationship between the disturbance sample and the predicted SPR curve to obtain a regression coefficient matrix and a bias term; using the standardized SPR curve, the process parameters, the regression coefficient matrix and the bias terms to train a reverse regression model; and inputting the abnormal SPR curve detected in real time into the reverse regression model to obtain the process parameters of the abnormal SPR curve. According to the invention, the prediction precision and reasoning speed of the model are improved, the interpretability of the model is improved, and the precision and stability of the SPR connection process are ensured.

The invention relates to the technical field of electronics, in particular to a chip packaging test method, which comprises the step of performing high-precision identification and long-term reliability prediction on a defect area by adopting a deep learning model through combining a chip thermal stress distribution map and electrical characteristic data of the defect area. In the detection process of a packaging defect area, normalization processing is carried out on a chip thermal stress distribution map, principal component analysis and dimension reduction are carried out on electrical characteristic data of the defect area, standardization and dimension reduction processing of input data are ensured, the standardized map and the reconstructed electrical characteristic data are input into a pre-trained convolutional neural network, and the electrical characteristic data of the defect area are detected. And extracting spatial features and carrying out feature fusion to generate a fused feature vector. And outputting predicted values of the thermal cycle life, the electromigration life and the mechanical fatigue life through a neural network regression layer based on the fusion feature vector. According to the invention, efficient and accurate detection of the packaging defect area is realized, and improvement of chip packaging quality and estimation of product reliability are facilitated.