313 results about "Information aggregation" patented technology

The invention provides a consensus decision question answering system based on multi-AI agent game, and relates to the technical field of artificial intelligence. The system comprises a multi-domain information aggregation module, an interaction effect deduction module, a strategy fusion calibration unit, a distributed behavior adaptive mechanism and an aggregation strategy discrimination module. The multi-domain information gathering module is used for unifying multi-source strategy information and environment situation data, the interaction effect deduction module is used for analyzing and quantifying the mutual influence relation of strategies between intelligent agents, and the strategy fusion calibration unit generates correction suggestions based on a game deduction and optimization method. The distributed behavior self-adaptive mechanism is used for locally and progressively executing a correction path in an intelligent agent; and the aggregation strategy judgment module dynamically evaluates the overall strategy state. The multi-agent consensus decision-making question-answering method realizes consensus decision-making question-answering of multiple agents in a complex environment, can effectively identify and correct non-collaborative strategy deviation, and improves the coordination, stability and immunity of a system.

The invention relates to the technical field of artificial intelligence and biomedicine, and provides a multi-agent diagnosis planning device and method based on a consultation thinking process. The multi-agent diagnosis planning device comprises a diagnosis planning agent, a pathological section analysis agent, an in-hospital information aggregation agent, an information search agent, a knowledge base search agent, an evaluation agent, an arbitration agent and a risk assessment agent. Each agent realizes data interaction through a dynamic priority message bus, and the evaluation agent and the arbitration agent form a progressive verification closed loop: after the evaluation agent outputs a question evidence chain, the arbitration agent triggers diagnosis correction only when question items are greater than 3 items, and otherwise, final diagnosis is output based on a preset rule. According to the invention, the thinking process of multidisciplinary expert collaboration in clinical consultation is simulated by constructing a multi-agent collaborative diagnosis framework, and the whole process intelligence from medical data acquisition and analysis to diagnosis decision is realized.

The invention discloses an attention perception path reasoning method for a knowledge graph, and the method comprises the steps: firstly, generating an entity representation containing local semantics through employing a graph attention network coding entity and an adjacency relation, and synchronously obtaining an attention weight representing the association intensity between entities; secondly, innovatively providing a target-guided biased random walk path sampling strategy, and adaptively exploring a high-quality multi-hop semantic path related to a target task by taking the attention weight as a bias; then, information aggregation is carried out on the sampled semantic paths through a path encoder, and global path representation is obtained; and finally, carrying out deep fusion on the local entity representation and the global path representation, and jointly inputting the local entity representation and the global path representation into a prediction layer to carry out knowledge graph link prediction. According to the method, local structure perception and global path reasoning are cooperatively optimized through an attention mechanism, so that the link prediction precision and interpretability of the knowledge graph are remarkably improved.

The invention relates to the technical field of material testing, and particularly discloses an intelligent quality detection method and system for a micro-mineral bio-organic fertilizer, local thermal excitation is applied to a fertilizer sample through a micro-area thermal pulse excitation device, a gas release kinetic curve and a spectrum change track are synchronously collected, and a gas spectrum coupling data cube is constructed; performing differential transformation and modal decomposition on the time sequence feature set, extracting transient response feature vectors and generating an activity response distribution diagram; establishing component-activity correlation analysis, and decoupling biological metabolic activity and matrix background interference through feature separation and a comparative learning strategy to obtain a dynamic metabolic fingerprint; constructing a multi-dimensional feature space by using the dynamic metabolic fingerprints and the apparent characteristic parameters, and performing information aggregation and feature reconstruction by using a graph convolutional network to generate a comprehensive quality index; and finally, realizing quality grade judgment based on a quality characteristic pyramid structure, and establishing a dynamic early warning mechanism by analyzing time sequence evolution characteristics of the activity response distribution diagram.

The invention discloses an unmanned autonomous cluster flight control method based on a bionic alert mechanism, which is applied to the field of unmanned autonomous cluster control, and is characterized in that a W-MSR algorithm and a dynamic weighted bionic alert mechanism are fused, the W-MSR filters and eliminates extreme values of neighbor individuals, the dynamic weighted bionic alert mechanism strengthens input from informed individuals, and the dynamic weighted bionic alert mechanism is used for improving the robustness of the unmanned autonomous cluster flight control. And meanwhile, the influence of suspicious individuals is inhibited, so that the local information aggregation degree is effectively improved, and group splitting can be prevented. Different from a scheme depending on explicit attacker recognition, the method can improve the motion accuracy and connectivity of the unmanned autonomous cluster in a confrontation environment by probabilistically suppressing the influence of suspicious individuals and amplifying a consistent signal, provides a new idea for improving the security of a swarm intelligence system, and has a wide application prospect. Damage of malicious individuals to the cluster is effectively defended, and robustness and recovery capability of the cluster in a complex environment are improved.

The invention relates to the technical field of petroleum drilling engineering, and particularly discloses an analysis and prediction system based on petroleum drilling and production data, which realizes virtual reconstruction of a physical drilling system by constructing a digital mapping body, and adopts a multi-source sensor network to collect drill string vortex frequency and wellbore temperature field gradient data. Forming a time sequence feature set through time sequence alignment and fusion processing; performing differential transformation and modal decomposition on the feature set to extract transient response feature vectors and generate a risk distribution diagram; establishing a parameter-risk correlation model, and decoupling mechanical vibration and thermal stress interference by solving a physical equation to obtain a dynamic risk index; the risk indexes and the process parameters are mapped to a three-dimensional grid to construct a multi-dimensional feature space, a feature fusion grid is adopted to achieve information aggregation and reconstruction, and comprehensive risk assessment indexes are generated; a dynamic early warning threshold value is established according to the evaluation indexes, and drilling parameters are optimized in real time through closed-loop control.

The invention discloses a machine learning-based low-voltage transformer area electric leakage risk intelligent identification method and system. The identification method comprises the following steps of 1, multi-dimensional electric leakage feature construction and transformer area information aggregation analysis; 2, dynamically evaluating the electric leakage risk and adaptively adjusting a threshold value; and step 3, an electric leakage type intelligent identification and confidence degree determination mechanism. The electric leakage risk identification and classification method has the beneficial effects that by introducing a graph nerve enhanced gradient boosting tree model (G-GTBoost) and an improved residual error convolution-time sequence neural network (Res-CNN-LSTM) model and cooperating with a multi-modal feature fusion and dynamic threshold adjustment mechanism, the performance is remarkably improved in electric leakage risk identification and classification.

The invention relates to the technical field of artificial intelligence and data processing, and discloses a multi-mode identity relation inference system based on a graph neural network. The system comprises a multi-modal feature extraction module, a cross-modal alignment module, a graph structure construction module, a dynamic relation reasoning module and a decision output module. According to the method, the cross-modal alignment module is introduced to project the image features and the text features to a unified public semantic space, so that the nonlinear distribution difference of heterogeneous modals in an embedding space is effectively eliminated, and cross-modal alignment errors are avoided from the source; by integrating the attention mechanism of modal perception in the graph neural network, the system can dynamically learn the semantic association strength between the nodes in different modals, adaptively adjust the weight distribution in the neighborhood information aggregation process, and significantly improve the accuracy of node characterization.

The invention relates to the field of machine learning, particularly discloses an artificial intelligence-based power system fault analysis and diagnosis method, and effectively solves the problem of information loss caused by neglecting a key waveform form in a transient signal in the prior art through a local feature extraction and serialization module. An original signal is converted into a local feature sequence with more characterization significance. Aiming at the averaging bottleneck of an existing model in an information aggregation stage, a traditional feature compression method is abandoned, and a sequence information aggregation and decision-making mechanism is provided. According to the mechanism, a context sensing sequence is regarded as a probability event, and modeling is carried out on the sequence from three orthogonal dimensions of a content center, time sequence dispersion and distribution uncertainty by calculating feature expectation, time sequence variance and information entropy of the context sensing sequence. The method can deeply insight and quantify the essential difference of different events in the time sequence dynamic evolution mode, thereby fundamentally solving the problem of misjudgment caused by feature confusion.

The invention discloses a regional public opinion propagation mode mining method based on a heterogeneous graph neural network. The method comprises the following specific steps: S1, constructing a heterogeneous graph; s2, weight definition of edge semantics; s3, node feature propagation and updating are carried out based on the graph neural network, and node embedding representation is obtained; and S4, node embedding is clustered, and regional public opinion propagation mode recognition is carried out in an embedding space. According to the method, joint modeling of users, media and regional nodes in a multi-relation heterogeneous graph structure is realized, and cross-semantic-edge information aggregation is completed by using a multi-relation attention mechanism, so that node embedding representation capable of representing multi-level propagation characteristics is obtained; the expressions provide basic support for subsequent public opinion propagation link identification, diffusion range prediction and opinion leader mining.

The invention relates to a Parkinson's disease treatment effect prediction method based on a multi-modal image model and a related device. The method comprises the following steps: acquiring a multi-modal vector of a Parkinson's disease patient; aligning the multi-modal vectors on a time axis, and constructing a time point data element sample sequence; deploying a double-flow cross attention encoder for a sample sequence in each time point data element, and outputting a fused multi-modal feature through the double-flow cross attention encoder; and connecting the fused multi-modal features with digital clinical treatment scheme vectors corresponding to corresponding time points to form time point comprehensive feature vectors, inputting the time point comprehensive feature vectors to a time sequence information aggregation gating circulation unit, outputting final time point aggregation features, and inputting the final time point aggregation features to a multi-task adaptive prediction head. A UPDRS total score or a specific sub-scale score for the patient at a future preset point in time is predicted. According to the method, time sequence modeling is carried out on multi-mode and multi-time-point data, so that the accuracy and interpretability of Parkinson's disease treatment effect prediction are improved.

The invention relates to the field of underwater sound signal processing, in particular to an underwater DOA (direction of arrival) estimation method based on graph nerves and a convolutional neural network, which comprises the following steps: 1, establishing a linear array, and enabling narrow-band signals to simultaneously reach an underwater sound array; 2, performing signal preprocessing to obtain a signal covariance matrix, and performing normalization processing; 3, extracting correlation between array elements and spatial features of array signals, and performing data supplementation on sparse linear array information; 4, forming a double-branch structure, enhancing the information aggregation capability, and extracting features from a space path and a time domain path; and 5, constructing an adjacent matrix, filling node features of damaged array elements, adopting a double-branch structure, extracting spatial features and time domain features, carrying out feature integration, and outputting a DOA estimation result. The spatial correlation between array elements is extracted and the array sparsity problem is processed by using the graph neural network, and the time domain features of the signals are extracted in combination with the convolutional neural network, so that more accurate and more robust DOA estimation can be realized under the conditions of low signal-to-noise ratio and array sparsity.

The invention discloses a local-global information aggregation remote sensing image change detection method based on Version Mama, and belongs to the technical field of remote sensing image change detection. The method adopts a Dual-LGNet network model for prediction, and comprises the following steps: inputting a dual-time-phase remote sensing image map before and after change into a twin encoder to obtain a feature map aggregating global features and local features; inputting the feature map aggregating the global features and the local features into a feature fusion module to obtain a fused difference feature map; and inputting the difference feature map into a progressive decoder to obtain a final prediction result. Wherein the LG-SS2D is introduced into the encoder to carry out feature extraction, and an LG-SS2D block comprises a parallel convolution branch and a Mama-based global branch. Through combination of local-global information aggregation and boundary enhancement, the recognition and detection precision of a remote sensing change area can be improved on the premise of keeping linear complexity, and especially the detection precision of a large building and a small target change area can be improved.

The invention provides a three-dimensional hand posture estimation method fusing hand and object features, relates to the technical field of hand posture estimation, and constructs a three-dimensional hand posture estimation network fusing the hand and object features. Comprising a feature extraction module, a double-flow hand feature pyramid sub-network, a hand feature dynamic adjustment module, a double-flow attention collaborative feature enhancement module and a decoder. Establishing a loss function of the three-dimensional hand posture estimation network fused with the hand and object features, training the three-dimensional hand posture estimation network fused with the hand and object features, and obtaining a hand posture estimation result based on the trained three-dimensional hand posture estimation network fused with the hand and object features. According to the method, local details and global semantic dependence are captured through bidirectional cross-scale information aggregation, and the problem of channel information loss of a traditional feature pyramid is relieved; and in combination with hand-object geometric constraint and semantic complementation characteristics, the cross-modal feature alignment capability is enhanced, and deep fusion of hand-object geometric constraint and semantic complementation is realized.

The invention discloses a federated learning image classification method based on time extension gradient cumulative average. Each client uses a local data set to train a local model to obtain gradient information of a current communication round, and uploads the gradient information to the central server after compression; the client continues to perform local training in a time period from the time when the client starts to upload the compression gradient information of the current communication round to the time when the server completes the compression gradient information aggregation of the current communication round; the central server side aggregates the compression gradient information to obtain average compression gradient information of the current communication round, and issues the average compression gradient information to the client side to update a local model of the (t + beta) th communication round; and repeating the steps until the training is completed, and inputting the to-be-detected image into the local model to obtain an image classification result. The method shows excellent performance under various data sets and different compression levels, has higher convergence speed and model precision, and solves the problem of low communication efficiency caused by network delay and bandwidth limitation in image classification of federated learning.

The invention belongs to the field of geological disaster early warning, and provides a landslide danger prediction system and method, and the method comprises the steps: carrying out the fusion and normalization processing of multi-source disaster-inducing factors, obtaining an impact factor matrix, and carrying out the weighted correction of the impact factor matrix; determining spatial association strength and semantic association degree between the nodes according to the association edges; adopting feature mapping, association weight calculation and information aggregation adaptive learning to obtain association strength and association features of the nodes; according to the association strength and the association features of the nodes, learning by adopting an association graph model to obtain a global prediction model, and optimizing the global prediction model; and predicting the target landslide area through the optimized global prediction model to obtain a prediction result, and carrying out danger grade division on the prediction result according to a preset probability threshold. The beneficial effect of the invention is that the precision of landslide risk prediction is improved.

The invention belongs to the technical field of pipeline safety assessment, and discloses a multi-dimensional information interconnection and autonomous evolution collaborative pipeline full-state safety assessment method. And capturing a high-order relationship of data through double hypergraph reasoning of the instance-level hypergraph and the modal-level hypergraph to realize efficient interconnection. According to the method, mode-level and instance-level hypergraph information features are extracted through hypergraph information propagation, high-order correlation is mined through double-graph information aggregation, cross-mode and cross-instance consistency information and exclusive information are output after feature recombination, multi-dimensional data deep fusion is promoted, and high-quality data support is provided for follow-up pipeline full-state safety assessment. A two-stage autonomous evolution mechanism of intra-class progressive calibration and inter-class knowledge migration is respectively adapted to slight fluctuation and significant change scenes of the deep sea environment: precise adaptation of environment perturbation is realized through dual-branch feature extraction and dynamic weight adjustment in a domain; model parameter dynamic optimization is completed between domains through spatial-temporal feature clustering and cross-domain knowledge migration, and dynamic environment self-adaption can be achieved without manual intervention.

The invention discloses an intelligent calibration method and system for the electric quantity of an instrument, and relates to the technical field of electric energy metering of an electric power system.The intelligent calibration method comprises the steps that an electric meter network diagram with a main meter and sub-meters as nodes is constructed, and robust composite correlation indexes are calculated on the basis of reading increments in a sliding time window; the interference of common background load and instantaneous noise is suppressed by using a threshold clipping and main table purification mechanism, so that the depiction capability of the correlation graph on a physical connection relationship is improved fundamentally, and the phase to which the sub-table belongs can be identified more accurately in a stock area with unknown phase or incomplete topology; a graph neural network model is introduced based on the network graph, electric meter embedding representation with physical meaning is obtained through neighborhood information aggregation, phase grouping is achieved in combination with unsupervised clustering, split-phase energy balance loss is designed in the training stage, and the difference between the reading of a main meter and the sum of the reading of sub-meters in each phase group is used as a constraint. Alignment of the embedding space and the energy conservation relation is pushed in a self-supervised mode.

The invention provides a single-frame infrared small target detection method based on a variant convolutional neural network. The single-frame infrared small target detection method specifically comprises the steps that a picture input module carries out standardized preprocessing and primary feature extraction on an original image; the efficient channel attention module extracts channel descriptors through global pooling, efficiently calculates weights by using a lightweight network, and performs weighted adjustment on input channels; the FPN module realizes efficient fusion of cross-scale features through a space cyclic shift and channel grouping technology; the residual error convolution module adopts a structural design of combining a convolution path and jump connection, so that the problem of gradient disappearance in deep network training is effectively solved; the feed-forward module integrates an efficient channel attention mechanism, and enhances the response capability of a key feature channel through global information aggregation, channel weight generation and re-calibration operation; the model output module converts high-level features into high-precision mask images and calculates detection performance indexes in real time, all the modules work cooperatively to form a complete feature extraction-fusion-prediction processing chain, and the performance of the infrared small target detector is remarkably improved while the real-time reasoning efficiency is guaranteed.

The invention provides a natural image matting method and system based on text and boundary information aggregation, and relates to the technical field of image processing. Splicing the original color image and the corresponding ternary image, extracting initial features, and obtaining enhanced fusion features through multi-scale Laplacian high-frequency extraction and cosine similarity weighted fusion; based on the enhanced fusion feature and the ternary image, generating a gated ternary fusion feature fused with priori knowledge through a trans-attention mechanism; global coding modeling is carried out on the gated three-value fusion features to obtain deep features; text prompt and multi-scale boundary information are introduced based on deep features, adaptive up-sampling is guided through a cross-attention mechanism, semantic difference consistency constraint is adopted between decoding layers, consistency constraint is implemented from pixel appearance, high-level semantics and color component dimensions, and finally a transparency image is output through a prediction header to obtain an image matting result. And the fidelity and the boundary accuracy of high-frequency details in a matting result are effectively improved.

The invention relates to a brain glioma grading method based on multi-instance causal characterization learning, and the method comprises the following steps: S1, data preprocessing, S2, feature extraction through ResNet-50, S3, feature coding of high-dimensional feature representations of instances, S4, decoupling of low-dimensional feature representations of each instance, obtaining causal features and non-causal features, and carrying out classification on the causal features and the non-causal features. S5, constructing an adaptive adjacency matrix A, constructing a sparse adjacency matrix A (k) by adopting a Topk neighbor selection strategy, and carrying out information aggregation through a graph neural network to obtain aggregation features; S6, carrying out local feature aggregation on each scanning plane p, calculating a weight weighted p of each scanning plane by utilizing a full-connection network, and carrying out weighted fusion to obtain an aggregation feature; s7, fusing and grading the aggregated features and the features; the method has the advantage of improving the stability and clinical applicability of the grading model.

The invention discloses an industrial control vulnerability knowledge graph completion method based on negative sample generation, and the method comprises five modules: a vulnerability data crawling module, a named entity recognition module, a graph construction module, a sample enhancement module and a knowledge reasoning module. The vulnerability data crawling module crawls unstructured vulnerability information from a webpage; the named entity recognition module performs keyword extraction on the obtained vulnerability description information; the graph construction module combines the keywords extracted by the named entity recognition module with the information obtained in the data crawling module to construct a vulnerability graph; the sample enhancement module optimizes a data sampling process by applying a sample enhancement method based on sub-sampling, and performs information aggregation on each node of the obtained atlas according to a neighbor relation and attribute information; and finally, dynamically calculating the influence of the head entity on the tail entity by the knowledge reasoning module, and realizing graph completion according to the embedding calculation distance of the node and the relationship. According to the method, the problem of data sparsity in the vulnerability knowledge graph is solved.

Systems and methods for implementing domain-specific agent networks including configuring specialized agents optimized for retrieving information from a respective specific knowledge domain, receiving a user query, analyzing the user query to identify relevant knowledge domains, activating a subset of the specialized agents corresponding to the relevant knowledge domains, retrieving information by the subset of specialized agents, aggregating the retrieved information including information from a plurality of knowledge domains, providing the aggregated information to one or more h-LLMs, receiving a plurality of responses from the one or more h-LLMs, and generating a comprehensive response from the plurality of responses, the comprehensive response incorporating information from the plurality of knowledge domains.

The invention discloses a leakage current intelligent detection and identification method for a low-voltage distribution area, and relates to the technical field of leakage detection of the low-voltage distribution area, and the method comprises the following steps: S1, carrying out the aggregation analysis of the multi-modal information of the area; s2, leakage current dynamic detection and threshold determination; and S3, leakage current intelligent identification and confidence detection. According to the leakage current intelligent detection and identification method for the low-voltage distribution area, high-resolution feature extraction is realized through quantum time-frequency transformation and adaptive neural time-frequency decomposition, in combination with a space-time joint attention model and an uncertainty quantification module, the identification precision of the electric leakage type in measured data reaches 98.7% and is improved by 26.4% compared with that of a traditional SVM method, the average confidence coefficient is 0.93, and the accuracy of the electric leakage type identification is greatly improved. And the reliability under an extreme working condition can be ensured by only needing 6.5% of a re-triggering rate, so that the problems of high false alarm rate and insufficient dynamic response of a traditional method are solved, and an accurate and efficient electric leakage detection scheme is provided for safe operation and maintenance of a low-voltage distribution area.

The invention relates to an automatic news acquisition and submission system based on aggregation retrieval and large model controlled generation, which is characterized in that candidate news is acquired from a third-party news data interface through an information aggregation module, a purified text is extracted through a text extraction module of a multi-level degradation strategy, an optimal input is selected by a preferential retrieval module, and the candidate news is submitted to the third-party news data interface. And generating an abstract and a news brief report through a large model controlled generation module under the constraint of a system guard prompt and an original text fence, and performing automatic typesetting and publishing according to a template after cleaning and labeling. The method supports the bilateral pluggable design of the data source and the generation engine, has the characteristics of low illusion, high consistency, strong traceability, end-to-end automation and the like, remarkably reduces the labor cost, and is suitable for instituted news submission scenes in multiple fields of current politics, economy, science and technology, culture, military and the like.

The invention discloses an aspect emotion triple extraction method for implicit recognition enhanced table filling, and belongs to the field of emotion analysis. Comprising the following steps: modeling an input text and an overall implicit expression thereof by using a pre-trained encoder to obtain a context expression, and deploying a bidirectional information aggregation module to aggregate forward and backward information of two words in a word pair based on the context expression to construct a table expression, all aspect emotion triples are extracted from the tabular representation of the sentence using a decoding module. According to the method disclosed by the invention, the identification of the implicit aspect terms is enhanced, so that the table filling method can extract the aspect emotion triad from the text containing the implicit expression. Besides, the aggregation module fusing forward and reverse information not only integrates the characteristics of a single word, but also integrates the forward and reverse relationships between two words in a word pair, thereby effectively improving the performance of aspect emotion triple extraction.

The invention discloses a thermal power generation carbon emission simulation modeling method and system based on multi-source big data, and relates to the technical field of electric carbon emission calculation, and the method comprises the steps: obtaining thermal power generation data, carrying out the preprocessing of the data, building an electric carbon coupling refined model based on the preprocessed data, and carrying out the simulation modeling of the thermal power generation carbon emission. And generating a dynamic coupling relationship between equipment through time sequence feature mapping and node information aggregation, performing real-time electricity-energy-carbon modeling on coal-electricity and gas-electricity type thermal power generation enterprises under all working conditions, distinguishing data relationships under different operation working conditions, and constructing a hierarchical correlation model. According to the method, a gating mechanism of a graph neural network is introduced into electric carbon coupling modeling, the problem of information screening in dynamic interaction of equipment is solved, high-precision prediction covered by all working conditions is realized through working condition feature extraction and residual iteration, and an integrated carbon management scheme from prediction to regulation is provided for thermal power generation.

The invention relates to the technical field of power grid operation optimization, and discloses a graph convolutional network-based power grid topology analysis model generation method, which comprises the steps of constructing a multi-task learning teacher model which adopts a double-layer graph convolutional network, and performing layer-by-layer neighborhood information aggregation based on an adjacency matrix and a node feature matrix, jointly optimizing a node classification task, an edge prediction task and a graph-level regression task of the power grid topology through a multi-task loss function; training a student model based on teacher model distillation, wherein a total loss function of the student model is formed by weighting knowledge distillation loss, student model cross entropy loss and feature matching loss; and based on newly added power grid topology data, performing incremental training on the teacher model by optimizing a multi-task loss function to obtain an updated teacher model, and performing distillation training on the student model based on the updated teacher model to respond to the dynamic change of the power grid topology. The method has the beneficial effects that the reasoning time delay is reduced, and the dynamic topology adaptive capacity is improved.

The invention discloses a multi-working-condition industrial process soft measurement method based on multi-task learning and probability modeling, and aims to solve the problem of insufficient measurement precision caused by heterogeneous mixing of multi-working-condition process samples. The method comprises three core modules, namely a feature decoupling coding module, a hierarchical feature fusion module and a probability information aggregation module. Firstly, a spatial-temporal feature extractor is designed to explicitly decouple multi-working-condition data into working condition shared features and specific features, and hybrid feature expression and working condition recognition are achieved. Then, a hierarchical feature fusion module is constructed, deep fusion of information between working conditions is realized through a hierarchical expert gating network, and a complex interaction relationship between the working conditions is modeled; and finally, proposing a probability information aggregation strategy, inputting the fusion features into corresponding predictors, and weighting prediction results by using the working condition identification probability to generate final prediction output. According to the method, a classification task and a regression task are incorporated into a unified multi-task learning framework, and the good performance of multi-working-condition process performance index soft measurement is ensured.