Method and system for processing multi-channel thermocouple data of aluminum-silicon alloy eutectic bath electrolyte

Abstract

The application provides a kind of aluminium-silicon alloy eutectoid tank electrolyte multi-path thermocouple data processing method and system, the method is by synchronous acquisition multi-path temperature signal, combined with noise mode dynamic matching filtering strategy multimodal adaptive collaborative filtering, effectively eliminate multi-source noise interference, solved the problem of poor adaptability, data distortion of traditional fixed filtering;Then through the lightweight physical information neural network reconstruction temperature field of fusion tank body physical constraint, give consideration to data fitting degree and physical rationality, solved the problem of low reconstruction accuracy, insufficient real-time of traditional method;According to the multi-hop reasoning of eutectoid process timing knowledge graph, realize the positioning of abnormal type, root cause, solve the defects of high false alarm rate, no causal analysis of traditional detection;Finally, based on abnormal diagnosis generation adaptive control instruction, realize the accurate control of heating, stirring equipment.

Description

Technical Field

[0001] This invention relates to the field of temperature monitoring technology in aluminum alloy production processes, and particularly to a method and system for processing multi-channel thermocouple data in an aluminum-silicon alloy eutectoid tank electrolyte. Background Technology

[0002] Aluminum-silicon alloys, with their low density, high specific strength, good corrosion resistance, and excellent casting properties, are widely used in high-end equipment fields such as aerospace, automotive manufacturing, and electronic packaging. The eutectoid reaction is a core and critical process in the preparation of aluminum-silicon alloys. The uniformity and stability of the electrolyte temperature distribution within the eutectoid tank directly determine the alloy's compositional uniformity, phase composition, and mechanical properties, playing a decisive role in the reliability of the final product.

[0003] To monitor the temperature status of the eutectoid electrolyte in real time, the industry commonly employs the method of deploying multiple thermocouples inside the eutectoid electrolyte to synchronously collect temperature signals from various monitoring points, providing data support for subsequent process control. However, the eutectoid electrolyte operates under complex conditions of high temperature and dynamic stirring. The raw temperature data stream collected by the thermocouples is susceptible to multi-source noise, such as electromagnetic interference, equipment vibration, and electrolyte flow impact. Furthermore, the multi-channel data exhibits coupling correlation characteristics. At the same time, the temperature field distribution during eutectoid processes is nonlinear and strongly correlated in time and space, making the causes of abnormal operating conditions complex and diverse.

[0004] Therefore, how to achieve accurate denoising of multiple temperature signals, efficient reconstruction of temperature fields, rapid identification and causal localization of abnormal operating conditions, and generation of adaptive process control instructions based on analysis results have become core technical requirements for ensuring the stability of the aluminum-silicon alloy eutectoid process and improving product quality. Summary of the Invention

[0005] This invention provides a method and system for processing multi-channel thermocouple data in an aluminum-silicon alloy eutectoid cell electrolyte, which solves the problems of poor temperature data denoising adaptability and insufficient accuracy and real-time performance of temperature field reconstruction in the prior art.

[0006] On one hand, the present invention provides a method for processing multi-channel thermocouple data in an aluminum-silicon alloy eutectoid cell electrolyte, characterized in that it includes: Temperature signals from multiple thermocouple channels deployed inside the eutectoid electrolyte are collected synchronously to form a raw temperature data stream; The original temperature data stream is denoised using a multimodal adaptive collaborative filtering algorithm to obtain denoised data. The multimodal adaptive collaborative filtering algorithm is based on online noise pattern identification and dynamically combines median filtering, Kalman filtering, and collaborative filtering strategies based on adjacent channel data. The denoised data is input into a physical information neural network model to reconstruct the two-dimensional or three-dimensional temperature field distribution of the eutectoid electrolyte. Based on a pre-constructed co-exudative process timeline knowledge graph, anomaly detection and causal reasoning analysis are performed on the temperature field distribution, and anomaly warnings and anomaly cause inferences are output. Based on anomaly warnings and anomaly cause inferences, process optimization parameters are calculated, and adaptive control commands are generated for regulating the heating and stirring equipment in the eutectoid tank.

[0007] Optionally, a multimodal adaptive collaborative filtering algorithm is used to denoise the original temperature data stream to obtain denoised data, including: Frequency domain analysis is performed on each set of temperature information in the original temperature data stream to identify and extract the dominant noise mode features in the temperature information. The dominant noise mode features include pulse-type features, Gaussian features, or periodic features. Based on the dominant noise mode characteristics, a filtering algorithm that matches the dominant noise mode characteristics is dynamically selected and combined to generate a composite filtering operator; The original temperature data stream is filtered and calculated using the composite filter operator to obtain the first intermediate temperature data. A collaborative filtering matrix is ​​constructed based on the spatial topology of multiple thermocouple channels; The first intermediate temperature data is spatially consistent using the collaborative filtering matrix to output denoised data.

[0008] Optionally, inputting the denoised data into a physical information neural network model to reconstruct the two-dimensional or three-dimensional temperature field distribution of the eutectoid electrolyte specifically includes: The denoised data is fused with predefined physical constraint information of the tank to form fused input data. The physical constraint information of the tank includes boundary conditions, heat source location and material thermal property parameters. The fused input data is fed into a pre-trained lightweight physical information neural network. Through forward propagation calculation of the lightweight physical information neural network, a gridded temperature field distribution data covering the entire eutectoid electrolyte region is output. The loss function of the lightweight physical information neural network includes data fitting terms and physical equation constraint terms.

[0009] Optionally, the step of performing anomaly detection and causal reasoning analysis on the temperature field distribution based on a pre-constructed eutectoid process time series knowledge graph, and outputting anomaly warnings and anomaly cause inferences, includes: Extract spatiotemporal feature vectors from the temperature field distribution data; The spatiotemporal feature vector is input into the inference engine of the co-analytic process time series knowledge graph. Based on the connection relationship of the co-analytic process time series knowledge graph, multi-hop matching and inference are performed on the abnormal spatiotemporal features, and the output includes abnormal warning and abnormal cause inference, including abnormal type, confidence level, associated equipment and possible root cause.

[0010] Optionally, extracting the spatiotemporal feature vector from the temperature field distribution data includes: The plane of the eutectoid tank is divided into regularly spaced grid nodes with equal spacing; Spatial dimension analysis is performed on the temperature field distribution data to calculate the temperature difference between each regular grid node and at least two adjacent nodes in spatial directions, so as to form a temperature gradient vector characterizing local spatial changes. The temperature field distribution data is analyzed in the time dimension to calculate the temperature change rate of each regular grid node within a predetermined time window, so as to form a temperature change rate vector representing the temporal change trend. The temperature gradient vector and the temperature change rate vector are concatenated and normalized to generate a spatiotemporal feature vector.

[0011] Optionally, the spatiotemporal feature vector is input into the inference engine of the co-analytic process time-series knowledge graph. Based on the connection relationships of the co-analytic process time-series knowledge graph, multi-hop matching and inference are performed on the abnormal spatiotemporal features, and the output includes anomaly warnings and anomaly cause inferences, including anomaly type, confidence level, associated equipment, and possible root causes. The similarity between the spatiotemporal feature vector and the predefined feature patterns of each node in the co-analysis process time sequence knowledge graph is calculated to obtain one or more initial associated nodes, including equipment entity nodes and process status nodes. Starting from the initial associated node, a directional traversal is performed along the relational edges representing causal, temporal, or control relationships in the co-analysis process timing knowledge graph to obtain all associated nodes and connection paths within a preset number of hops, thus forming a candidate reasoning subgraph. Based on the candidate inference subgraph, the abnormal patterns represented by the spatiotemporal feature vector are classified and attributed according to preset rules, and the abnormal diagnosis report is generated and output. The abnormal diagnosis report includes at least the abnormal type determined by the classification result, the confidence level supported by the matching and inference process, the associated devices corresponding to the device entity nodes involved in the candidate inference subgraph, and the possible root causes pointed to by the key inference path.

[0012] Optionally, the step of calculating process optimization parameters based on anomaly warning and anomaly cause inference, and generating adaptive control commands for regulating the heating and stirring equipment of the eutectoid tank, includes: Analyze the anomaly diagnosis report to identify the target equipment that needs to be regulated and the desired temperature field correction direction; Based on the preset process control rule library and equipment operating status, a set of optimized process setting parameters are calculated, which include at least the target heating power and the stirrer speed. The process setting parameters are encapsulated into adaptive control instructions and sent to the corresponding actuators.

[0013] Optionally, it also includes: The pre-stored calibration parameters of each thermocouple channel are called to perform a secondary calibration calculation on the denoised data, correcting the system error of the thermocouple channel and the influence of ambient temperature, and obtaining the calibrated denoised data. Determine the number of consecutive missing values ​​in the denoised data due to transmission interruption; If the number of consecutive missing values ​​does not exceed a preset first threshold, then linear interpolation is used to complete the missing values.

[0014] Optionally, the anomaly detection and causal reasoning analysis based on the co-analytic process time series knowledge graph further includes: The temperature values ​​at each location point in the temperature field distribution data are compared with the preset upper and lower threshold values ​​for the eutectoid reaction temperature. A Level 1 warning is generated when the temperature at any location point continuously exceeds the upper limit threshold or falls below the lower limit threshold for a preset second threshold number of sampling periods. The sliding window method is used to calculate the rate of temperature change at each location point in the temperature field distribution data per unit time; when the rate of temperature change exceeds a preset rate of change threshold, or when the temperature shows a continuous unidirectional change for a preset third threshold sampling period, a level two warning is triggered. Based on the first-level or second-level early warning, the causal reasoning analysis of the co-analytical process time sequence knowledge graph is initiated, and the abnormal early warning and abnormal cause inference are generated in combination with the early warning information.

[0015] On the other hand, the present invention also provides an aluminum-silicon alloy eutectoid cell electrolyte multi-channel thermocouple data processing system, comprising: The data acquisition module is configured to synchronously acquire temperature signals from multiple thermocouple channels deployed inside the eutectoid electrolyte to form a raw temperature data stream. The data processing module is configured to: use a multimodal adaptive collaborative filtering algorithm to denoise the original temperature data stream to obtain denoised data; the multimodal adaptive collaborative filtering algorithm is based on online identified noise patterns and dynamically combines median filtering, Kalman filtering, and collaborative filtering strategies based on adjacent channel data; The temperature field reconstruction module is configured to: input the denoised data into a physical information neural network model to reconstruct the two-dimensional or three-dimensional temperature field distribution of the eutectoid electrolyte; The anomaly reasoning module is configured to: perform anomaly detection and causal reasoning analysis on the temperature field distribution based on a pre-built co-exudative process timeline knowledge graph, and output anomaly warnings and anomaly cause inferences; The control command generation module is configured to: calculate process optimization parameters based on abnormal early warning and abnormal cause inference, and generate adaptive control commands for regulating the heating and stirring equipment of the eutectoid tank.

[0016] On the other hand, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aluminum-silicon alloy eutectoid electrolyte multi-channel thermocouple data processing method as described above.

[0017] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aluminum-silicon alloy eutectoid cell electrolyte multi-channel thermocouple data processing method as described above.

[0018] On the other hand, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the aluminum-silicon alloy eutectoid cell electrolyte multi-channel thermocouple data processing method as described above.

[0019] The present invention provides a method and system for processing multi-channel thermocouple data in an aluminum-silicon alloy eutectoid tank electrolyte. The method simultaneously acquires multiple temperature signals and combines them with a multi-modal adaptive collaborative filtering strategy based on a dynamic noise pattern matching filter to effectively eliminate multi-source noise interference, solving the problems of poor adaptability and data distortion associated with traditional fixed filtering. Furthermore, it reconstructs the temperature field using a lightweight physical information neural network that integrates the physical constraints of the tank, balancing data fitting accuracy and physical rationality, thus addressing the issues of low reconstruction accuracy and insufficient real-time performance in traditional methods. Based on multi-hop reasoning using the eutectoid process time-series knowledge graph, it locates the anomaly type and root cause, overcoming the shortcomings of high false alarm rates and lack of causal analysis in traditional detection methods. Finally, it generates adaptive control commands based on anomaly diagnosis to achieve precise control of the heating and stirring equipment. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating the multi-channel thermocouple data processing method for an aluminum-silicon alloy eutectoid tank electrolyte provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of the aluminum-silicon alloy eutectoid cell electrolyte multi-channel thermocouple data processing system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0023] Figure 1 This is one of the flowcharts illustrating the multi-channel thermocouple data processing method for the eutectoid electrolyte in an aluminum-silicon alloy eutectoid tank provided in this embodiment of the invention.

[0024] like Figure 1 As shown in the embodiment of the present invention, the method for processing multi-channel thermocouple data in an aluminum-silicon alloy eutectoid tank electrolyte mainly includes the following steps: 101. Simultaneously collect temperature signals from multiple thermocouple channels deployed inside the eutectoid electrolyte to form a raw temperature data stream.

[0025] When acquiring temperature signals, it is necessary to consider the structure and process characteristics of the eutectoid tank and adopt a layered, zoned, and gridded deployment principle to set up thermocouple channels. For example, the electrolyte area of ​​the eutectoid tank is divided into four regions: the core reaction zone, the edge heat exchange zone, the near-heating zone, and the near-stirring zone. In each region, thermocouples are installed in a vertically layered and horizontally distributed manner. The core reaction zone is equipped with 4-6 thermocouples to ensure monitoring density, while the edge heat exchange zone, the near-heating zone, and the near-stirring zone are each equipped with 3-4 thermocouples. The total number of thermocouple channels for a medium-sized eutectoid tank is preferably 16-24. All thermocouple probes must be completely immersed in the electrolyte to a depth of not less than 5 cm, and the minimum distance between the probe and the tank wall, heating element, and stirring paddle must be not less than 3 cm. At the same time, high-temperature resistant ceramic supports are used to fix the thermocouples to prevent probe displacement due to electrolyte flow or equipment vibration.

[0026] After the thermocouple channels are deployed, the signal output terminals of all thermocouples are connected to a multi-channel synchronous data acquisition instrument. Based on the temperature change rate of the aluminum-silicon alloy eutectoid process, the sampling frequency of the acquisition instrument is set to 5-10Hz. The global synchronization trigger mechanism of the acquisition instrument is activated to ensure that all thermocouple channels synchronously acquire the temperature signal of the electrolyte at the same timestamp. According to the thermocouple calibration table, the temperature signal is converted into the corresponding temperature signal. Finally, the temperature data of all channels are integrated and stored in a structured format of channel number-acquisition timestamp-temperature value to form a raw temperature data stream containing spatial location information and time series information.

[0027] 102. A multimodal adaptive collaborative filtering algorithm is used to denoise the original temperature data stream to obtain denoised data; Among them, the multimodal adaptive collaborative filtering algorithm is based on online noise pattern identification and dynamically combines median filtering, Kalman filtering and collaborative filtering strategies based on adjacent channel data.

[0028] Specifically, a multimodal adaptive collaborative filtering algorithm is used to denoise the original temperature data stream to obtain denoised data, including: Frequency domain analysis is performed on each set of temperature information in the raw temperature data stream to identify and extract the dominant noise mode features in the temperature information. The dominant noise mode features include impulse features, Gaussian features, or periodic features. Based on the characteristics of the dominant noise mode, a composite filter operator is generated by dynamically selecting and combining filtering algorithms that match the characteristics of the dominant noise mode. The original temperature data stream is filtered and calculated using a composite filtering operator to obtain the first intermediate temperature data. A collaborative filtering matrix is ​​constructed based on the spatial topology of multiple thermocouple channels; Spatial consistency optimization is performed on the first intermediate temperature data using a collaborative filtering matrix to output denoised data.

[0029] First, a Fast Fourier Transform (FFT) is used to analyze the frequency domain of each group of temperature information in the raw temperature data stream. By comparing the signal energy proportions of different frequency bands, the dominant noise mode features in the temperature information are identified and extracted. These dominant noise mode features include pulse-type, Gaussian-type, or periodic-type features. For example, for the raw data collected by 16 thermocouples in a medium-sized eutectoid cell, if an isolated high-energy spike exists in the frequency domain graph of a certain channel within a certain time interval, its dominant noise is determined to be pulse-type. This feature is often caused by momentary jitter of the thermocouple probe due to equipment vibration. If the frequency domain graph shows a wide-band uniform energy distribution, the dominant noise is determined to be Gaussian-type. This feature often originates from electromagnetic interference from electrical equipment surrounding the eutectoid cell. If the frequency domain graph shows a significant energy peak at a specific frequency, the dominant noise is determined to be periodic-type. This feature is often caused by harmonic interference from the power supply system.

[0030] Subsequently, based on the identified dominant noise pattern characteristics, filtering algorithms that match the dominant noise pattern characteristics are dynamically selected and combined to generate composite filtering operators. For example, if the dominant noise is impulse-type, a composite filtering operator is constructed with median filtering as the core and Kalman filtering as the auxiliary. If the dominant noise is Gaussian-type, a composite filtering operator is constructed with Kalman filtering as the core and median filtering as the auxiliary. If two or more mixed noise characteristics are detected, a composite filtering operator is constructed that integrates median filtering, Kalman filtering, and adjacent channel collaborative filtering. The weights of each filtering algorithm in the operator are dynamically allocated according to the energy proportion of the noise characteristics. For example, when the energy proportion of impulse noise is 60% and the energy proportion of Gaussian noise is 40%, the weight of median filtering is set to 0.6 and the weight of Kalman filtering is set to 0.4.

[0031] Next, the generated composite filter operator is used to perform channel-by-channel and time-stamp-by-time filtering calculations on the original temperature data stream. Median filtering is used to suppress impulse noise spikes, and Kalman filtering is used to smooth and correct Gaussian noise based on the temporal continuity of temperature changes, thus obtaining the first intermediate temperature data with single-channel independent noise initially removed.

[0032] Based on the hierarchical and partitioned spatial topology of multiple thermocouple channels, a collaborative filtering matrix is ​​constructed according to the spatial distance between channels and the correlation of their respective functional areas. For example, for channels 1 and 2 that are horizontally adjacent in the core reaction zone, the spatial correlation weight is set to 0.6; for channel 1 in the core reaction zone and channel 5 in the same column above, the weight is set to 0.3; and for channel 1 in the core reaction zone and channel 12 in the edge heat exchange zone, the weight is set to 0.1. The spatial correlation weights of all channels are integrated in matrix form to form a collaborative filtering matrix with the same dimension as the total number of thermocouple channels. Finally, the collaborative filtering matrix is ​​used to perform spatial consistency optimization on the first intermediate temperature data. The outlier data points in the first intermediate temperature data that deviate from the average temperature of the surrounding channels are corrected by weighted averaging. For example, if the first intermediate temperature value of a certain channel is 1050℃, while the weighted average temperature value of the five adjacent channels is 1020℃, the outlier value is corrected according to the weight of the collaborative filtering matrix. The final output is denoised data that has both temporal smoothness and spatial consistency.

[0033] 103. Input the denoised data into the physical information neural network model to reconstruct the two-dimensional or three-dimensional temperature field distribution of the eutectoid electrolyte.

[0034] Among them, the two-dimensional or three-dimensional temperature field distribution of the reconstructed eutectoid electrolyte can be presented in a visual way, such as by drawing isotherm diagrams, three-dimensional temperature distribution diagrams, etc., so that operators can intuitively understand the temperature distribution in the eutectoid tank.

[0035] Furthermore, the reconstructed temperature field distribution data can be further analyzed and processed. For example, statistical parameters such as the average value and standard deviation of the temperature field can be calculated to assess the temperature uniformity within the eutectoid tank; the gradient change of the temperature field can be analyzed to determine the direction and intensity of heat transfer; and, in conjunction with the requirements of the eutectoid process, it can be determined whether the current temperature field meets the process conditions, providing a basis for subsequent process control.

[0036] Specifically, the denoised data is input into a physical information neural network model to reconstruct the two-dimensional or three-dimensional temperature field distribution of the eutectoid electrolyte, including: The denoised data is fused with the predefined physical constraint information of the tank to form fused input data. The physical constraint information of the tank includes boundary conditions, heat source location and material thermal property parameters. The fused input data is fed into a pre-trained lightweight physical information neural network. Through the forward propagation calculation of the lightweight physical information neural network, a gridded temperature field distribution data covering the entire eutectoid electrolyte region is output. The loss function of the lightweight physical information neural network includes a data fitting term and a physical equation constraint term.

[0037] The process involves deep feature-level fusion of denoised data with predefined physical constraint information of the eutectoid tank to form structured fused input data. This physical constraint information includes the boundary conditions of the eutectoid tank. Boundary conditions include heat transfer coefficients between the tank wall and the external environment, heat dissipation rate of the electrolyte surface, heat source location, and material thermophysical parameters. The fusion process involves tensor concatenation of the channel-time-temperature three-dimensional features of the denoised data with the spatial-attribute two-dimensional features of the physical constraint information of the tank, generating fused input data that matches the dimensionality requirements of the model.

[0038] The fused input data is then fed into a pre-trained lightweight physical information neural network. This network is structurally optimized based on the requirements for reconstructing the temperature field of the eutectoid electrolyte in the eutectoid tank. This is achieved by simplifying the number of neurons in the hidden layers and using small 1×1 convolutional kernels to reduce computational complexity. The pre-training process incorporates a large amount of historical temperature data and heat conduction simulation data from aluminum-silicon alloy eutectoid processes, applying physical constraints simultaneously during model training. Subsequently, through the forward propagation calculation of the lightweight physical information neural network, the input layer extracts features from the fused data, the hidden layers fit the spatial distribution of the temperature field, and the output layer generates gridded temperature field distribution data covering the entire eutectoid electrolyte region. The two-dimensional temperature field distribution data corresponds to a specified cross-section of the eutectoid tank. The temperature grid matrix corresponds to a three-dimensional temperature field distribution data, which includes a vertically layered three-dimensional temperature grid tensor. Each grid node is assigned a temperature value. It is worth noting that the loss function of the lightweight physical information neural network is composed of a weighted sum of a data fitting term and a physical equation constraint term. The data fitting term is used to minimize the mean square error between the model output temperature and the denoised measured temperature, ensuring that the reconstruction result fits the actual monitoring data. The physical equation constraint term is constructed based on Fourier's law of heat conduction, which is used to constrain the temperature field distribution output by the model to satisfy the physical laws of heat transfer within the electrolyte, avoiding reconstruction results that violate the physical mechanism. The final output is a two-dimensional or three-dimensional temperature field distribution that combines accuracy and physical rationality.

[0039] 104. Based on the pre-constructed co-exudative process time sequence knowledge graph, perform anomaly detection and causal reasoning analysis on the temperature field distribution, and output anomaly warnings and anomaly cause inferences.

[0040] The coteritic process time-series knowledge graph stores information on temperature variations, equipment operating status, process parameters, and causal relationships between aluminum-silicon alloy coteritic processes at different time stages. Constructing this knowledge graph requires collecting a large amount of historical coteritic process data, including temperature data, equipment parameters, and process operation records. This data then undergoes cleaning, preprocessing, and feature extraction. Finally, knowledge representation learning methods are used to transform this data into a graph-structured knowledge graph, where nodes represent entities and edges represent causal relationships between entities.

[0041] Specifically, based on a pre-constructed knowledge graph of the eutectoid process timeline, anomaly detection and causal reasoning analysis are performed on the temperature field distribution, outputting anomaly warnings and anomaly cause inferences, including: Spatiotemporal feature vectors are extracted from temperature field distribution data.

[0042] Among them, the spatiotemporal feature vector is extracted from the temperature field distribution data, including: The plane of the eutectoid tank is divided into regularly spaced grid nodes with equal spacing; Spatial dimension analysis is performed on the temperature field distribution data to calculate the temperature difference between each regular grid node and at least two adjacent nodes in spatial directions, so as to form a temperature gradient vector characterizing local spatial changes. Perform time-dimensional analysis on the temperature field distribution data, calculate the temperature change rate of each regular grid node within a predetermined time window, and form a temperature change rate vector that characterizes the temporal change trend; The temperature gradient vector and the temperature change rate vector are concatenated and normalized to generate a spatiotemporal feature vector.

[0043] Specifically, when dividing the plane of the eutectoid tank into regularly spaced grid nodes, the actual tank size parameters and temperature monitoring requirements of the aluminum-silicon alloy eutectoid tank are first combined. The plane of the eutectoid tank is then divided into regularly spaced grid nodes according to a preset regularly spaced standard. For example, for a cylindrical eutectoid tank with a diameter of 1.5m and a height of 0.8m, the cross-section of the cylindrical eutectoid tank can be divided at 5cm intervals along the radial and circumferential directions, and at 10cm intervals along the height direction, thereby forming a three-dimensional regularly spaced grid node array covering the electrolyte area of ​​the tank. Each grid node is assigned unique spatial coordinate information.

[0044] Subsequently, spatial dimension analysis was performed on the temperature field distribution data. Taking each regular grid node as the center, its adjacent grid nodes in at least two spatial directions were selected, and the temperature values ​​corresponding to each node were extracted. The temperature difference between the central node and its adjacent nodes was calculated. For example, the temperature difference between a certain grid node and its adjacent nodes to its right and above was calculated. , The temperature difference values ​​are arranged in an orderly manner to form a temperature gradient vector that can characterize the local spatial temperature change of the node.

[0045] Next, a time dimension analysis is performed on the temperature field distribution data. A predetermined time window is set for each regular grid node, and the temperature data corresponding to different timestamps of the node within the time window is extracted. By calculating the ratio of the temperature difference between adjacent timestamps to the time interval, the instantaneous temperature change rate of the node within the time window is obtained. Then, the multiple instantaneous temperature change rates are averaged to form a temperature change rate vector that can characterize the temporal change trend of the node's temperature.

[0046] Finally, the temperature gradient vector and temperature change rate vector corresponding to each regular grid node are concatenated to obtain a combined vector that integrates spatial and temporal features. Then, the min-max normalization algorithm is used to standardize the combined vector to eliminate the dimensional differences between different feature dimensions, and finally generate a spatiotemporal feature vector that has both spatial local change characteristics and temporal evolution laws.

[0047] The spatiotemporal feature vectors are input into the inference engine of the co-analytic process time series knowledge graph. Based on the connection relationship of the co-analytic process time series knowledge graph, multi-hop matching and inference are performed on the abnormal spatiotemporal features. The output includes abnormal warnings and abnormal cause inferences, including abnormal type, confidence level, associated equipment and possible root causes.

[0048] Specifically, the spatiotemporal feature vectors are input into the inference engine of the co-analytic process time-series knowledge graph. Based on the connection relationships of the co-analytic process time-series knowledge graph, multi-hop matching and inference are performed on the abnormal spatiotemporal features. The output includes anomaly warnings and anomaly cause inferences, including anomaly type, confidence level, associated equipment, and possible root causes. The similarity between the spatiotemporal feature vector and the predefined feature patterns of each node in the co-analytical process time sequence knowledge graph is calculated to obtain one or more initial associated nodes. The initial associated nodes include equipment entity nodes and process status nodes. Starting from the initial associated node, a directional traversal is performed along the relational edges representing causal, temporal, or control relationships in the co-analysis process timing knowledge graph to obtain all associated nodes and connection paths within a preset number of hops, thus forming a candidate reasoning subgraph. Based on the candidate inference subgraph, the abnormal patterns represented by the spatiotemporal feature vectors are classified and attributed according to preset rules, and an anomaly diagnosis report is generated and output. An anomaly diagnosis report should include at least the anomaly type determined by the classification results, the confidence level supported by the matching and reasoning process, the associated devices corresponding to the device entity nodes involved in the candidate reasoning subgraph, and the possible root causes pointed to by the key reasoning paths.

[0049] Specifically, the pre-constructed co-analytic process timing knowledge graph is retrieved first. The co-analytic process timing knowledge graph includes equipment entity nodes, process status nodes, process parameter nodes, and various relation edges that represent the causal, temporal, and control relationships between nodes. At the same time, each node is configured with a predefined feature pattern that matches the spatiotemporal feature vector dimension.

[0050] The extracted spatiotemporal feature vectors are then compared with the predefined feature patterns of all nodes in the knowledge graph using cosine similarity calculation. A similarity threshold of 0.8 is set, and one or more initial associated nodes with similarity higher than 0.8 are selected. For example, if a spatiotemporal feature vector shows abnormal features such as a local grid node temperature gradient > 5℃ / cm and a temperature change rate > 2℃ / s in the core reaction zone, it is matched with the local overheating process state node and the heating element equipment entity node after similarity calculation, which is the initial associated node.

[0051] Starting from the initial associated node, a preset number of jumps is set according to the reasoning requirements of anomaly diagnosis. A directional traversal is performed along the relationship edges in the knowledge graph that represent causal, temporal, or control relationships. For example, starting from the local overheating state node, the traversal is performed along the causal relationship edge caused by ... to the heating element power abnormality equipment state node, and then from the heating element power abnormality node, the traversal is performed along the causal relationship edge originating from ... to the heating element temperature control module failure equipment node. At the same time, all associated nodes and the connection paths between nodes are recorded during the traversal process. These nodes and paths are integrated to form a candidate reasoning subgraph that focuses on the anomaly association logic.

[0052] Then, based on the candidate inference subgraph, a pre-set eutectoid process anomaly diagnosis rule library is invoked. For example, rule 1 in the eutectoid process anomaly diagnosis rule library: if the local temperature gradient is abnormal and associated with the heating element node, it is determined to be a heating anomaly; rule 2: if the temperature change rate is abnormal and associated with the stirring device node, it is determined to be a stirring anomaly.

[0053] Based on rules 1 and 2, the abnormal patterns represented by the spatiotemporal feature vectors are classified. At the same time, the confidence of the abnormal diagnosis results is calculated based on the initial matching similarity and the completeness of the reasoning path. The corresponding associated devices are determined through the device entity nodes in the candidate reasoning subgraph, and the possible root causes of the abnormality are pointed to through the key reasoning path.

[0054] Finally, a standardized anomaly diagnosis report is generated and output. This report includes at least the anomaly type determined by the classification results, the confidence level supported by the matching and reasoning process, the associated devices corresponding to the device entity nodes involved in the candidate reasoning subgraph, and the possible root causes pointed to by the key reasoning paths.

[0055] 105. Based on anomaly warning and anomaly cause inference, calculate process optimization parameters and generate adaptive control commands for regulating the heating and stirring equipment of the eutectoid tank.

[0056] The process optimization parameters include adjustments to heating power, stirring speed, and heating time. When calculating these parameters, the possible directions for parameter adjustments are initially determined based on the anomaly type and potential root causes identified in the anomaly warning, combined with historical data and expert experience regarding the aluminum-silicon alloy eutectoid process.

[0057] Specifically, based on anomaly warnings and anomaly cause inferences, process optimization parameters are calculated, and adaptive control commands for regulating the heating and stirring equipment of the eutectoid tank are generated, including: Analyze the anomaly diagnosis report to identify the target equipment that needs to be regulated and the desired direction of temperature field correction; Based on the preset process control rule library and equipment operating status, a set of optimized process setting parameters are calculated. The process setting parameters include at least the target heating power and the stirrer speed. The process setting parameters are encapsulated into adaptive control instructions and sent to the corresponding actuators.

[0058] The process involves structured parsing of the output anomaly diagnosis report, extracting the specific anomaly type, associated equipment identifier, root cause of the anomaly, and temperature field anomaly characterization data to identify the target equipment requiring parameter adjustment. Simultaneously, the desired temperature field correction direction is determined by combining the temperature field reconstruction results. For example, for a local overheating anomaly caused by a fault in the heating element temperature control module, the target equipment is the fault-related heating element and the stirring device in the surrounding area. The temperature field correction direction is to reduce the temperature of the local overheating area and reduce the temperature gradient between the core area and the edge area, so that the overall temperature field returns to the uniform distribution state set by the process.

[0059] Subsequently, a pre-set eutectoid process control rule library is retrieved. This library includes control logic corresponding to different anomaly types and temperature field deviations. For example, when the local temperature exceeds the set value by 5-10℃, the heating power is reduced by 8%-12%; when the temperature gradient is greater than 3℃ / cm, the stirrer speed is increased by 5%-10%. Simultaneously, real-time operating status data of the eutectoid tank is accessed, and the temperature field correction direction, the root cause of the anomaly, and the equipment operating status are coupled and analyzed to calculate a set of optimized process setting parameters that balance process requirements and equipment safety. These parameters include at least the target heating power and the stirrer speed. For example, for the aforementioned local overheating anomaly, if the current power of heating element No. 3 is 22kW, the process setting power is 20kW, and the temperature exceeds the set value by 8℃, then according to the rule library, the target heating power is adjusted to 18.5kW. Simultaneously, to accelerate heat diffusion, the speed of the bottom stirrer is adjusted from the current 60r / min to 66r / min.

[0060] Finally, according to the communication protocol format of the actuator of the eutectoid tank equipment, the calculated optimized process setting parameters are encapsulated into standardized adaptive control commands, and then the control commands are sent to the corresponding actuators through the industrial control bus to achieve precise adaptive control of the heating and stirring equipment.

[0061] In some embodiments, it also includes: The pre-stored calibration parameters of each thermocouple channel are called to perform secondary calibration calculations on the denoised data, correcting the system errors of the thermocouple channels and the influence of ambient temperature, and obtaining calibrated denoised data. Determine the number of consecutive missing values ​​in the denoised data due to transmission interruption; If the number of consecutive missing values ​​does not exceed the preset first threshold, linear interpolation will be used to complete the missing values.

[0062] Specifically, it calls the pre-stored calibration parameters for each thermocouple channel. The calibration parameters include the factory system error correction coefficient for each thermocouple, the drift compensation coefficient after long-term use, and the correction formula for the influence of ambient temperature.

[0063] Subsequently, by combining the real-time monitoring temperature of the environment surrounding the eutectoid cell, the denoised data was substituted into the calibration formula for secondary calibration calculation. By correcting the systematic errors caused by the performance degradation of the thermocouple itself and the fluctuation of the ambient temperature, the denoised data after calibration with the system deviation eliminated was obtained.

[0064] Next, a full-channel, full-timestamp data integrity scan is performed on the calibrated and denoised data to identify and count the number of consecutive missing values ​​in each thermocouple channel due to signal transmission interruption. At the same time, a preset first threshold is retrieved. If the number of consecutive missing values ​​in a certain channel does not exceed the first threshold, linear interpolation is used to complete the data. Specifically, the effective temperature data points adjacent to the missing segment are extracted, the slope of the temperature change between adjacent effective points is calculated, and then the temperature value of each missing point is calculated and completed based on the timestamp position of the missing value, thereby ensuring the temporal continuity and integrity of the data. If the number of consecutive missing values ​​exceeds the first threshold, the data segment is marked as abnormally missing and recorded for subsequent manual verification.

[0065] In some embodiments, anomaly detection and causal reasoning analysis based on a co-analytic process timing knowledge graph further includes: The temperature values ​​at each location in the temperature field distribution data are compared with the preset upper and lower threshold values ​​for the eutectoid reaction temperature. A Level 1 warning is generated when the temperature at any location point continuously exceeds the upper limit threshold or falls below the lower limit threshold for a preset second threshold number of sampling cycles. The sliding window method is used to calculate the rate of temperature change at each location point in the temperature field distribution data per unit time. When the rate of temperature change exceeds the preset rate of change threshold, or when the temperature shows a continuous unidirectional change for a preset third threshold sampling period, a level two warning is triggered. Based on a Level 1 or Level 2 early warning, causal reasoning analysis of the co-analytical process time sequence knowledge graph is initiated, and abnormal warnings and abnormal cause inferences are generated in combination with the early warning information.

[0066] Specifically, when a Level 1 warning is triggered, such as when the temperature at a certain location continuously exceeds the upper limit threshold of the eutectoid reaction temperature for a preset second threshold sampling period, it indicates a possible abnormality in the heating system, equipment failure, or unreasonable process parameter settings. At this time, relevant information is immediately retrieved from the eutectoid process timing knowledge graph. Starting from the equipment entity nodes and process status nodes associated with that location, a directed traversal is performed along the edges of causal, temporal, or control relationships to construct a candidate reasoning subgraph. For example, if the location is associated with a heating element, then the search begins from the heating element equipment entity node to find possible causes for the continuous temperature exceeding the upper limit, such as abnormal heating element power or a malfunctioning temperature control module.

[0067] When a Level 2 warning is triggered, taking an abnormal temperature change rate exceeding a preset threshold as an example, this may indicate a problem with heat transfer within the tank, such as uneven stirring or localized heat loss. The system will also perform causal reasoning analysis based on the coexisting process time-series knowledge graph. First, it extracts the spatiotemporal feature vector of the location from the temperature field distribution data. Then, it calculates similarity with the predefined feature patterns of each node in the knowledge graph to find the initial associated node. Next, starting from the initial associated node, it performs a directional traversal according to a preset number of hops to construct a candidate reasoning subgraph. For example, if the abnormal temperature change rate at this location is associated with the stirring device, it will traverse along the edges related to the stirring device in the knowledge graph, potentially discovering reasons such as abnormal stirrer speed or damaged stirring blades.

[0068] After completing the causal reasoning analysis, the system combines early warning information with a pre-set eutectoid process anomaly diagnosis rule base to classify anomaly patterns, calculate the confidence level of the anomaly diagnosis results, identify related equipment and possible root causes, and finally generate anomaly warnings and anomaly cause inferences. This information will provide important basis for subsequent process optimization and equipment control, enabling timely measures to ensure the normal operation of the aluminum-silicon alloy eutectoid process.

[0069] Based on the same general inventive concept, this invention also protects an aluminum-silicon alloy eutectoid electrolyte multi-channel thermocouple data processing system. The aluminum-silicon alloy eutectoid electrolyte multi-channel thermocouple data processing system provided by this invention will be described below. The aluminum-silicon alloy eutectoid electrolyte multi-channel thermocouple data processing system described below can be referred to in correspondence with the aluminum-silicon alloy eutectoid electrolyte multi-channel thermocouple data processing method described above.

[0070] See Figure 2 This invention also provides an aluminum-silicon alloy eutectoid electrolyte multi-channel thermocouple data processing system, comprising: The data acquisition module 210 is configured to: synchronously acquire temperature signals from multiple thermocouple channels deployed inside the eutectoid electrolyte to form a raw temperature data stream; The data processing module 220 is configured to: use a multimodal adaptive collaborative filtering algorithm to denoise the original temperature data stream to obtain denoised data; the multimodal adaptive collaborative filtering algorithm is based on online identified noise patterns and dynamically combines median filtering, Kalman filtering and collaborative filtering strategies based on adjacent channel data; The temperature field reconstruction module 230 is configured to input the denoised data into the physical information neural network model to reconstruct the two-dimensional or three-dimensional temperature field distribution of the eutectoid electrolyte. The anomaly reasoning module 240 is configured to: perform anomaly detection and causal reasoning analysis on the temperature field distribution based on a pre-built coexisting process time sequence knowledge graph, and output anomaly warnings and anomaly cause inferences; The control command generation module 250 is configured to: calculate process optimization parameters based on abnormal early warning and abnormal cause inference, and generate adaptive control commands for regulating the heating and stirring equipment of the eutectoid tank.

[0071] Figure 3 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention.

[0072] like Figure 3 As shown, the electronic device may include a processor 310, a communication interface 320, a memory 330, and a communication bus 340. The processor 310, communication interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can call logic instructions from the memory 330 to execute a multi-channel thermocouple data processing method for aluminum-silicon alloy eutectoid tank electrolyte.

[0073] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0074] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the aluminum-silicon alloy eutectoid cell electrolyte multi-channel thermocouple data processing method provided by the above methods.

[0075] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the aluminum-silicon alloy eutectoid cell electrolyte multi-channel thermocouple data processing method provided by the above methods.

[0076] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0077] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for processing multi-channel thermocouple data in an aluminum-silicon alloy eutectoid cell electrolyte, characterized in that, include: Temperature signals from multiple thermocouple channels deployed inside the eutectoid electrolyte are collected synchronously to form a raw temperature data stream; The original temperature data stream is denoised using a multimodal adaptive collaborative filtering algorithm to obtain denoised data. The multimodal adaptive collaborative filtering algorithm is based on online noise pattern identification and dynamically combines median filtering, Kalman filtering, and collaborative filtering strategies based on adjacent channel data. The denoised data is input into a physical information neural network model to reconstruct the two-dimensional or three-dimensional temperature field distribution of the eutectoid electrolyte. Based on a pre-constructed co-exudative process timeline knowledge graph, anomaly detection and causal reasoning analysis are performed on the temperature field distribution, and anomaly warnings and anomaly cause inferences are output. Based on anomaly warnings and anomaly cause inferences, process optimization parameters are calculated, and adaptive control commands are generated for regulating the heating and stirring equipment in the eutectoid tank.

2. The method for processing multi-channel thermocouple data in an aluminum-silicon alloy eutectoid cell electrolyte according to claim 1, characterized in that, The original temperature data stream is denoised using a multimodal adaptive collaborative filtering algorithm to obtain denoised data, including: Frequency domain analysis is performed on each set of temperature information in the original temperature data stream to identify and extract the dominant noise mode features in the temperature information. The dominant noise mode features include pulse-type features, Gaussian features, or periodic features. Based on the dominant noise mode characteristics, a filtering algorithm that matches the dominant noise mode characteristics is dynamically selected and combined to generate a composite filtering operator; The original temperature data stream is filtered and calculated using the composite filter operator to obtain the first intermediate temperature data. A collaborative filtering matrix is ​​constructed based on the spatial topology of multiple thermocouple channels; The first intermediate temperature data is spatially consistent using the collaborative filtering matrix to output denoised data.

3. The method for processing multi-channel thermocouple data in an aluminum-silicon alloy eutectoid tank electrolyte according to claim 1, characterized in that, The step of inputting the denoised data into a physical information neural network model to reconstruct the two-dimensional or three-dimensional temperature field distribution of the eutectoid electrolyte specifically includes: The denoised data is fused with predefined physical constraint information of the tank to form fused input data. The physical constraint information of the tank includes boundary conditions, heat source location and material thermal property parameters. The fused input data is fed into a pre-trained lightweight physical information neural network. Through forward propagation calculation of the lightweight physical information neural network, a gridded temperature field distribution data covering the entire eutectoid electrolyte region is output. The loss function of the lightweight physical information neural network includes data fitting terms and physical equation constraint terms.

4. The method for processing multi-channel thermocouple data in an aluminum-silicon alloy eutectoid tank electrolyte according to claim 1, characterized in that, The pre-constructed eutectoid process time series knowledge graph is used to perform anomaly detection and causal reasoning analysis on the temperature field distribution, outputting anomaly warnings and anomaly cause inferences, including: Extract spatiotemporal feature vectors from the temperature field distribution data; The spatiotemporal feature vector is input into the inference engine of the co-analytic process time series knowledge graph. Based on the connection relationship of the co-analytic process time series knowledge graph, multi-hop matching and inference are performed on the abnormal spatiotemporal features, and the output includes abnormal warning and abnormal cause inference, including abnormal type, confidence level, associated equipment and possible root cause.

5. The method for processing multi-channel thermocouple data in an aluminum-silicon alloy eutectoid tank electrolyte according to claim 4, characterized in that, Extracting spatiotemporal feature vectors from the temperature field distribution data includes: The plane of the eutectoid tank is divided into regularly spaced grid nodes with equal spacing; Spatial dimension analysis is performed on the temperature field distribution data to calculate the temperature difference between each regular grid node and at least two adjacent nodes in spatial directions, so as to form a temperature gradient vector characterizing local spatial changes. The temperature field distribution data is analyzed in the time dimension to calculate the temperature change rate of each regular grid node within a predetermined time window, so as to form a temperature change rate vector representing the temporal change trend. The temperature gradient vector and the temperature change rate vector are concatenated and normalized to generate a spatiotemporal feature vector.

6. The method for processing multi-channel thermocouple data in an aluminum-silicon alloy eutectoid cell electrolyte according to claim 4, characterized in that, The spatiotemporal feature vector is input into the inference engine of the co-analytic process time-series knowledge graph. Based on the connection relationships of the co-analytic process time-series knowledge graph, multi-hop matching and inference are performed on the abnormal spatiotemporal features. The output includes anomaly warnings and anomaly cause inferences, including anomaly type, confidence level, associated equipment, and possible root causes. The similarity between the spatiotemporal feature vector and the predefined feature patterns of each node in the co-analysis process time sequence knowledge graph is calculated to obtain one or more initial associated nodes, including equipment entity nodes and process status nodes. Starting from the initial associated node, a directional traversal is performed along the relational edges representing causal, temporal, or control relationships in the co-analysis process timing knowledge graph to obtain all associated nodes and connection paths within a preset number of hops, thus forming a candidate reasoning subgraph. Based on the candidate inference subgraph, the abnormal patterns represented by the spatiotemporal feature vector are classified and attributed according to preset rules, and the abnormal diagnosis report is generated and output. The abnormal diagnosis report includes at least the abnormal type determined by the classification result, the confidence level supported by the matching and inference process, the associated devices corresponding to the device entity nodes involved in the candidate inference subgraph, and the possible root causes pointed to by the key inference path.

7. The method for processing multi-channel thermocouple data in an aluminum-silicon alloy eutectoid cell electrolyte according to claim 1, characterized in that, The process optimization parameters are calculated based on anomaly warning and anomaly cause inference, and adaptive control commands are generated for regulating the heating and stirring equipment of the eutectoid tank, including: Analyze the anomaly diagnosis report to identify the target equipment that needs to be regulated and the desired temperature field correction direction; Based on the preset process control rule library and equipment operating status, a set of optimized process setting parameters are calculated, which include at least the target heating power and the stirrer speed. The process setting parameters are encapsulated into adaptive control instructions and sent to the corresponding actuators.

8. The method for processing multi-channel thermocouple data in an aluminum-silicon alloy eutectoid cell electrolyte according to claim 1, characterized in that, Also includes: The pre-stored calibration parameters of each thermocouple channel are called to perform a secondary calibration calculation on the denoised data, correcting the system error of the thermocouple channel and the influence of ambient temperature, and obtaining the calibrated denoised data. Determine the number of consecutive missing values ​​in the denoised data due to transmission interruption; If the number of consecutive missing values ​​does not exceed a preset first threshold, then linear interpolation is used to complete the missing values.

9. The method for processing multi-channel thermocouple data in an aluminum-silicon alloy eutectoid cell electrolyte according to claim 1, characterized in that, The anomaly detection and causal reasoning analysis based on the co-analytic process time series knowledge graph also includes: The temperature values ​​at each location point in the temperature field distribution data are compared with the preset upper and lower threshold values ​​for the eutectoid reaction temperature. A Level 1 warning is generated when the temperature at any location point continuously exceeds the upper limit threshold or falls below the lower limit threshold for a preset second threshold number of sampling periods. The sliding window method is used to calculate the rate of temperature change at each location point in the temperature field distribution data per unit time; when the rate of temperature change exceeds a preset rate of change threshold, or when the temperature shows a continuous unidirectional change for a preset third threshold sampling period, a level two warning is triggered. Based on the first-level or second-level early warning, the causal reasoning analysis of the co-analytical process time sequence knowledge graph is initiated, and the abnormal early warning and abnormal cause inference are generated in combination with the early warning information.

10. A multi-channel thermocouple data processing system for an aluminum-silicon alloy eutectoid cell electrolyte, characterized in that, include: The data acquisition module is configured to synchronously acquire temperature signals from multiple thermocouple channels deployed inside the eutectoid electrolyte to form a raw temperature data stream. The data processing module is configured to: use a multimodal adaptive collaborative filtering algorithm to denoise the original temperature data stream to obtain denoised data; the multimodal adaptive collaborative filtering algorithm is based on online identified noise patterns and dynamically combines median filtering, Kalman filtering, and collaborative filtering strategies based on adjacent channel data; The temperature field reconstruction module is configured to: input the denoised data into a physical information neural network model to reconstruct the two-dimensional or three-dimensional temperature field distribution of the eutectoid electrolyte; The anomaly reasoning module is configured to: perform anomaly detection and causal reasoning analysis on the temperature field distribution based on a pre-built co-exudative process timeline knowledge graph, and output anomaly warnings and anomaly cause inferences; The control command generation module is configured to: calculate process optimization parameters based on abnormal early warning and abnormal cause inference, and generate adaptive control commands for regulating the heating and stirring equipment of the eutectoid tank.

Images