Smelting power failure intelligent detection method and system based on multi-modal data fusion

Abstract

This invention discloses an intelligent fault detection method and system for metallurgical power systems based on multimodal data fusion, belonging to the field of data processing technology. The method includes: S1: acquiring multimodal data of the metallurgical power system; S2: performing feature extraction and optimization processing on the multimodal data to obtain optimized feature data; S3: fusing the optimized feature data to obtain fused comprehensive data; S4: performing abnormal fault detection analysis on the fused comprehensive data to obtain detection results; S5: issuing early warnings based on the detection results. The system includes a multimodal data acquisition module, a multimodal data feature extraction and optimization module, a multimodal data real-time fusion module, a multi-level abnormal fault detection module, and a fault early warning module. This invention can monitor the operating status of the metallurgical power detection system and improve the response speed and accuracy of fault detection.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an intelligent detection method and system for power grid faults in metallurgical plants based on multimodal data fusion. Background Technology

[0002] Metallurgical power systems are a crucial component of the modern metallurgical industry, primarily providing stable and reliable power support for the smelting process. Smelting processes typically require high temperatures, high pressures, and complex electrical equipment, making the power system susceptible to faults such as current overload, equipment overheating, and abnormal temperatures. These faults not only affect production efficiency but can also lead to equipment damage, production stoppages, or serious safety accidents. Therefore, fault detection and early warning systems for metallurgical power systems have always been an important research direction in the field of industrial control.

[0003] Currently, fault detection methods in metallurgical power systems largely rely on traditional sensor data acquisition, such as monitoring physical quantities like current, voltage, and temperature. These methods typically employ physical model-based detection or rule-based thresholding for fault diagnosis. However, these methods have certain limitations, primarily their inability to accurately handle complex fault modes and their tendency to lose information or misjudge situations when processing multimodal data. Traditional methods often cannot comprehensively analyze data from multiple sensors and lack sufficient flexibility to address fault modes under different operating environments.

[0004] However, the existing technologies mentioned above still have technical problems such as low accuracy in fault detection and early warning, poor real-time performance, and poor robustness. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides a method and system for intelligent detection of power failures in smelting based on multimodal data fusion, which solves the technical problems of low accuracy, poor real-time performance, and poor robustness in fault detection and early warning.

[0006] The technical solution adopted by this invention to solve its technical problem is: This invention provides an intelligent detection method for power system faults in smelting based on multimodal data fusion, comprising the following steps: S1: acquiring multimodal data of the smelting power system; S2: performing feature extraction and optimization processing on the multimodal data to obtain optimized feature data; S3: fusing the optimized feature data to obtain fused comprehensive data. S4: Perform anomaly detection and analysis on the integrated data after fusion to obtain the detection results; S5: Issue early warnings based on the detection results.

[0007] Preferably, the multimodal data includes one or more of the following: visual data, temperature data, electrical data, and vibration data.

[0008] Preferably, the multimodal data in step S1 is obtained through multimodal data acquisition and multimodal data preprocessing, wherein the preprocessing operation is one or more of the following: noise reduction, alignment and normalization.

[0009] Preferably, the multimodal data feature extraction in step S2 employs one or more of the following feature extraction methods: For visual data, convolutional neural networks are used to extract image features. Basic features in the image are extracted through multiple convolutional layers, and then preliminary dimensionality reduction is performed through pooling layers to extract discriminative features. For temperature data, the time-series features of the temperature data are extracted using statistical feature extraction methods, and the frequency domain features of the temperature data are analyzed using Fourier transform methods to capture potential abnormal fluctuations. For electrical data, a method combining time-domain and frequency-domain features is used for extraction to identify periodic or non-periodic fault modes. For vibration data, time-frequency analysis is used to extract frequency and vibration amplitude features from the vibration data to identify mechanical fault modes.

[0010] Preferably, in step S2, the feature data of the extracted multimodal data are subjected to the following: The spacetime transformation process uses the following formula: ; in, It is the first Mode in time and spatial location Spatiotemporal feature data, using express; It is the first Each mode in time Preliminary characteristic data; Indicates the first The mapping vectors of the preliminary feature data of each modality along the time dimension are used to extract time series features. It is a time series mapping function; Represents convolution or nonlinear operators used to combine temporal and spatial information; It is the first Preliminary feature data of each modality in spatial location and time The spatiotemporal correlation characteristics on It is a spatial correlation function; A multi-level feature complexity optimization mechanism enables features of different modalities to have adaptive weights at different time periods. The feature complexity optimization formula is as follows: ; ; ; in, It is the optimized version of the first Feature data of each modality; It is the first The weighting factor of each mode represents the relative importance of that mode in the feature optimization process; It is the first The modality at time... The rating; It is the first The relative importance of each modality in calculating the score, with values ​​ranging from [value range missing]. Indicates the first The modality at time... The energy level reflects the fluctuation amplitude and activity level of modal data; It refers to the size of the time window, which is determined based on specific needs. It is the first The volatility weighting coefficients for each mode, with values ​​ranging from [value range missing]. It is the first The modality at time... The variance; It is the first Spatiotemporal characteristic data of modalities; It is the first The first mode and the first The interaction weight coefficients between modes reflect the degree of mutual influence between different modes, and their values ​​range from [value range missing]. ; Indicates inner product calculation; It represents the total number of modes.

[0011] Preferably, in step S3, a weighted fusion model is used to fuse the optimized feature data, and the formula for calculating the fused features is as follows: ; in, It is the integrated data after fusion; It is a nonlinear fusion operation used to synthesize weighted optimization features and intermodal interaction terms.

[0012] Preferably, in step S4, the abnormal fault detection and analysis includes: S41: Coarse screening, used to quickly screen and filter out normal data that is unlikely to be faulty; S42: Deep learning-enhanced detection employs deep neural networks for detailed classification to further uncover deeper potential fault features; it uses a deep neural network in the form of a multilayer perceptron to perform multi-level nonlinear mapping on the integrated data after initial screening through multiple hidden layers, thereby capturing complex anomaly patterns. S43: Result fusion, comprehensively considering the detection results of each stage, to arrive at the final fault detection result.

[0013] Preferably, the method further includes step S6: after step S3, constructing a smelting power fault prediction model, performing prediction processing on the fused comprehensive data to obtain fault prediction results, and step S5 also providing early warning based on the fault prediction results.

[0014] This invention also provides an intelligent detection system for smelting power faults based on multimodal data fusion, used to implement the above-mentioned intelligent detection method for smelting power faults based on multimodal data fusion, comprising: The multimodal data acquisition module is configured to execute step S1; The multimodal data feature extraction and optimization module is configured to execute step S2; The multimodal data real-time fusion module is configured to execute step S3; The multi-level anomaly fault detection module is configured to execute step S4; The fault warning module is configured to execute step S5.

[0015] This invention also provides an intelligent detection system for smelting power faults based on multimodal data fusion, used to implement the aforementioned intelligent detection method for smelting power faults based on multimodal data fusion, comprising: The multimodal data acquisition module is configured to execute step S1; The multimodal data feature extraction and optimization module is configured to execute step S2; The multimodal data real-time fusion module is configured to execute step S3; The multi-level anomaly fault detection module is configured to execute step S4; The fault warning module is configured to execute step S5; The fault prediction module is configured to execute step S6.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. By fusing multimodal data, the operational status of the smelting power monitoring system can be comprehensively and accurately monitored. The introduction of spatiotemporal transformation and feature complexity optimization mechanisms allows features of different modes to adaptively adjust according to their weights under different time periods and environments. After optimization, the feature data can dynamically reflect the actual state of the smelting power monitoring system, ensuring high accuracy even in changing smelting environments. The optimized feature data avoids data redundancy and interference, thereby improving the response speed and accuracy of fault detection.

[0017] 2. A multi-level anomaly fault detection algorithm based on deep learning is used to process the multi-modal fused data in multiple stages, gradually screening for potential fault modes. By combining Long Short-Term Memory (LSTM) networks and Convolutional Neural Networks (CNNs), fault prediction of the smelting power detection system can be achieved, enhancing its adaptability and intelligence. Attached Figure Description

[0018] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a structural diagram of a metallurgical power fault intelligent detection system based on multimodal data fusion, as described in this invention. Figure 2 This is a flowchart of an intelligent detection method for power faults in metallurgical plants based on multimodal data fusion, as described in this invention. Detailed Implementation

[0019] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0021] The following description, in conjunction with the accompanying drawings, details a specific scheme for an intelligent detection system for smelting power faults based on multimodal data fusion provided by the present invention.

[0022] See attached document Figure 1 The diagram illustrates a structural diagram of an intelligent detection system for smelting power grid faults based on multimodal data fusion, according to an embodiment of the present invention. The system includes the following components: Multimodal data acquisition and preprocessing module, multimodal data feature extraction and optimization module, multimodal data real-time fusion module, multi-level anomaly and fault detection module, fault prediction module, and fault early warning module; The multimodal data acquisition and preprocessing module acquires multimodal data in real time through data acquisition equipment, preprocesses the multimodal data, and sends the preprocessed multimodal data to the multimodal data feature extraction and optimization module; the multimodal data acquisition module in the system obtains multimodal data through the multimodal data acquisition and preprocessing module.

[0023] The multimodal data feature extraction and optimization module extracts and optimizes features from the preprocessed multimodal data to obtain optimized feature data, which is then sent to the multimodal data real-time fusion module. The multimodal data real-time fusion module fuses the optimized feature data in real time to obtain the fused comprehensive data, and sends it to the multi-level anomaly detection module and fault prediction module. The multi-level anomaly detection module uses a deep learning-based multi-level anomaly detection algorithm to perform multi-level anomaly detection and analysis on the fused comprehensive data, obtain the detection results, and send them to the fault early warning module. The fault prediction module constructs a smelting power fault prediction model based on historical multimodal data in the existing database, and uses the smelting power fault prediction model to process the fused comprehensive data to obtain fault prediction results, which are then sent to the fault early warning module. The fault early warning module issues alarms and warnings based on the detection results and fault prediction results, and notifies relevant personnel to handle the situation in real time.

[0024] See attached document Figure 2 The diagram illustrates a flowchart of an intelligent detection method for smelting power grid faults based on multimodal data fusion, provided by an embodiment of the present invention. The method includes the following steps: Note: In this embodiment, for ease of description, the steps "acquiring multimodal data of the smelting power system; performing feature extraction and optimization processing on the multimodal data to obtain optimized feature data; fusing the optimized feature data to obtain fused comprehensive data" are combined into one step, namely step S1. Compared with the invention content, the only difference is in the description method, which does not affect the disclosure of the invention. Step S2 below is similar and will not be described in detail.

[0025] S1. Real-time acquisition of multimodal data, preprocessing of multimodal data, feature extraction and optimization processing to obtain optimized feature data, real-time fusion of optimized feature data to obtain fused comprehensive data; Multimodal data is collected in real time using data acquisition devices such as sensors and cameras. The multimodal data includes visual data (images of the appearance of smelting equipment), temperature data, electrical data (such as current, voltage, etc.), vibration data, etc. The multimodal data is then preprocessed using methods such as noise reduction (such as filtering), alignment (such as interpolation), and standardization to obtain preprocessed multimodal data. The preprocessing methods used are all well known to those skilled in the art and will not be described in detail here.

[0026] Furthermore, existing feature engineering techniques are used to extract features from the preprocessed multimodal data to obtain preliminary feature data. Specifically, for visual data, convolutional neural networks (CNNs) are used to extract image features. Multiple convolutional layers are used to extract basic features such as edges, textures, and shapes from the image, followed by preliminary dimensionality reduction using pooling layers to extract the most discriminative features. For temperature data, statistical feature extraction methods (such as mean, standard deviation, maximum, and minimum values) are used to extract time-series features of the temperature data. Simultaneously, methods such as Fourier transform are used to analyze the frequency domain features of the temperature data to capture potential abnormal fluctuations. For electrical data, a combination of time-domain and frequency-domain features is used for extraction. Time-domain features include mean, variance, and peak value, while frequency domain features are extracted using Fourier transform to help identify periodic or non-periodic fault modes. For vibration data, time-frequency analysis methods (such as wavelet transform) are used to extract frequency and vibration amplitude features from the vibration data to help identify mechanical fault modes.

[0027] Furthermore, the preliminary feature data undergoes spatiotemporal transformation processing, introducing a spatiotemporal transformation formula for the preliminary modal features, as follows: ; in, It is the first Mode in time and spatial location Spatiotemporal feature data, using Indicates spatial location. This refers to the physical installation location or observation area coordinates of the multimodal data acquisition equipment in the smelting power system. It determines the spatial location attribute of the modal feature data, reflects the spatial influence of equipment distribution on the fault state, and is an important dimension for constructing spatiotemporal feature modeling. It is the first Each mode in time Preliminary characteristic data; Indicates the first The mapping vector of the preliminary feature data of each modality along the time dimension is used to extract time series features, representing the time features. It is a time series mapping function, determined according to the specific application scenario, such as linear regression, Fourier transform, smoothing, etc. It is a general-purpose time series modeling processing function in this field, which can select different time modeling methods according to modal characteristics. It is a common method in this field and has a clear implementation path. It refers to convolution and nonlinear operators, specifically a feature interaction mechanism for fusing temporal and spatial features. The implementation method can be selected according to specific application requirements to achieve effective combination of modal spatiotemporal features. For example, convolution operations (one-dimensional convolution, two-dimensional convolution) are used to model spatial propagation or local diffusion; Hadamard product operations, i.e., element-wise multiplication operations, are used to express local influence weights; nonlinear interaction mapping (such as multilayer perceptron combination) is used to extract complex nonlinear spatiotemporal coupling features, and is used to combine temporal and spatial information. It is the first Preliminary feature data of each modality in spatial location and time The spatiotemporal correlation features describe the influence of spatial location and time on the preliminary modal feature data, representing spatial features. This is a spatial correlation function, constructed using sensor locations and the physical characteristics of the smelting environment. Examples include spatial interpolation functions based on spatial data, physical modeling functions such as thermal diffusion functions and electromagnetic response functions, and functions learned through deep learning models. It has a sufficiently standardized modeling approach in the field of spatial modeling. The above formula introduces joint modeling of spatial location and timestamps, capturing spatiotemporal relationships through nonlinear functions to ensure that features accurately reflect the operating status of the smelting power system in a dynamic environment. Additionally, there is a spatiotemporal mapping function. , As a function for compressing temporal features and constructing spatial features from multimodal raw data, its selection is based on the physical characteristics and data distribution patterns in the actual smelting process (e.g., if the sampling frequency is high and the data contains periodic disturbances, Fourier transform is preferred; if the data changes slowly with time and the signal is less affected by sudden changes, moving average is preferred; if the spatial distribution shows a linear decay trend, linear interpolation can be selected). The method employed is a feature processing technique commonly used in data mining and machine learning. Different selections of these functions do not affect the overall technical approach and effectiveness of the proposed "fault detection based on spatiotemporal feature optimization and fusion". The outputs of all functions will be uniformly mapped to a vector space for downstream deep learning models to extract fault modes; differences can be automatically learned and compensated during model training.

[0028] After obtaining the spatiotemporal features, a multi-level feature complexity optimization mechanism is introduced to give adaptive weights to features of different modes at different time periods, thus solving the feature redundancy problem caused by fluctuations in the smelting environment. The feature complexity optimization formula is as follows: ; ; ; in, It is the optimized version of the first Feature data of each modality It is the first The weighting factor of each mode represents the relative importance of that mode in the feature optimization process; It is the first The modality at time... The rating; It is the first The relative importance of each modality in calculating the score is determined using the modal entropy weighting method based on information theory, with a reference range of values. Indicates the first The modality at time... The energy level reflects the fluctuation amplitude and activity level of modal data; It refers to the size of the time window, which is determined according to specific needs, such as 30 seconds or 1 minute; It is the first The volatility weighting coefficients for each mode are obtained based on historical statistical methods, such as normalizing the historical volatility of the mode characteristics. A reference value range is provided. It is the first The modality at time... The variance; It is the first Spatiotemporal characteristic data of modalities; It is the first The first mode and the first The interaction weight coefficients between modalities reflect the degree of mutual influence between different modalities. Based on statistical correlation calculation methods, such as the Pearson correlation coefficient and mutual information, the reference value range is [insert range here]. ; Indicates inner product calculation; It represents the total number of modes.

[0029] This step dynamically adapts to changes in the smelting environment through a spatiotemporal transformation formula and a multi-level feature complexity optimization mechanism, demonstrating scene adaptability. The spatiotemporal transformation formula simultaneously maps the features of each mode to both the time and spatial dimensions, allowing the features to change in tandem with dynamic factors such as temperature gradients, electromagnetic environment variations, equipment position shifts, and sensor placement differences at the smelting site. Simultaneously, the multi-level feature complexity optimization mechanism updates the importance of different modes in real time based on modal energy, variance volatility, and interaction weights between modes. This enables the system to autonomously adjust the contribution of each modal feature under changing furnace conditions (such as variations in heat intensity, current surges, changes in charge loading, and smoke obstruction at different smelting stages), thus maintaining the stability and effectiveness of feature representation. Therefore, the fused features dynamically reflect real-world conditions, are independent of fixed scene configurations, and are unaffected by external interference such as sensor noise, changes in smelting operations, and equipment state drift. The system as a whole exhibits significant scene adaptability.

[0030] After feature optimization, the optimized features from each modality are fused to form a comprehensive multi-dimensional feature set. A weighted fusion model is used, which comprehensively considers the weights of each modality and the nonlinear interaction effects. The formula for calculating the fused feature set is as follows: ; in, It is the integrated data after fusion; It is a nonlinear fusion operation used to synthesize weighted optimization features and intermodal interaction terms. Additionally, " "Nonlinear fusion operation" uses nonlinear functions in neural networks (such as ReLU, Sigmoid, multilayer perceptron, or gating mechanisms) to fuse weighted features of different modalities and interaction information between modalities. The aim is to capture the complex nonlinear interaction relationships between multimodal data, enhance the expressive power and discriminative power of fused features, and enable the system to more accurately identify potential faults and improve prediction accuracy when facing the variable and coupled abnormal modes in the smelting process.

[0031] S2. Based on the integrated data after fusion, perform multi-level abnormal fault detection and analysis to obtain the detection results. At the same time, construct a smelting power fault prediction model, perform prediction processing on the integrated data after fusion, and obtain the fault prediction results.

[0032] A deep learning-based multi-level anomaly detection algorithm is used to perform multi-level anomaly detection and analysis on the fused comprehensive data to obtain detection results. The algorithm performs multi-stage anomaly detection on the fused comprehensive data, progressively screening for potential fault modes and providing a judgment result for each fault level. Specific content includes: The first stage, coarse screening (basic detection), is mainly used to quickly screen and filter out normal data that is unlikely to be faulty. Existing basic anomaly detection algorithms, such as Support Vector Machines (SVM), are used to initially assess the fused composite data and identify any obvious anomalies. SVM is a binary classification algorithm suitable for classifying data through a hyperplane in a high-dimensional space. Specifically, the fused feature vectors are input into the SVM model for classification. If the feature vector falls within the normal range, it is classified as "normal"; if it falls within the anomaly range, it proceeds to the next stage of detection.

[0033] The second stage, Deep Learning Enhanced Detection (Detailed Classification), employs Deep Neural Networks (DNNs) for detailed classification to further uncover deeper potential fault features. A multilayer perceptron (MLP)-based deep neural network is used, with multiple hidden layers performing multi-level nonlinear mapping on the initially screened and fused comprehensive data to capture complex anomaly patterns. Specifically, the data initially screened by SVM is input into the DNN for further feature mining. The DNN model uses multiple perceptron layers (including an input layer, multiple hidden layers, and an output layer) and nonlinear activation functions (such as ReLU) to perform deep feature learning on the data. Combined with the softmax activation function, it performs multi-class classification to achieve detailed classification. The third stage, the results fusion stage, comprehensively considers the detection results from each stage to arrive at the final fault detection result, including fault type (such as current overload, equipment overheating, image abnormality, etc.) and fault level (such as minor fault, serious fault, etc.). This step adopts a hierarchical detection structure of SVM+DNN, which significantly improves the recognition accuracy of complex faults, unlike single-model detection.

[0034] Based on historical multimodal data in an existing database, a smelting power fault prediction model is constructed. This model employs a combined architecture of Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN). LSTM is a recurrent neural network specifically designed for processing time-series data, effectively handling long-term dependencies in time series data. In smelting power detection systems, time-series data such as current and voltage exhibit significant temporal dependencies. LSTM can capture long-term trends in this data for prediction. The CNN demonstrates significant advantages in image data processing. In smelting power detection systems, image data (such as furnace appearance, smoke, and sparks) is crucial for fault identification. The CNN extracts low-level features (e.g., edges, textures) from images through convolutional layers, while pooling layers reduce dimensionality and computational cost, preventing overfitting. Finally, fully connected layers classify high-dimensional features to identify potential faults in the smelting process (e.g., overheating, sparks), yielding fault prediction results.

[0035] Furthermore, the integrated data is processed using a metallurgical power fault prediction model to obtain prediction results, including fault types such as current overload, equipment overheating, and abnormal temperature. Fault probability: This indicates the confidence level or likelihood that the predicted result is abnormal. For example, the fault probability for current overload is 95%. Fault level: Based on the severity of the fault and combined with a pre-set threshold using expert experience, the fault is classified into categories such as minor fault, severe fault, and critical fault.

[0036] Finally, the detection and prediction results are sent to the fault early warning module, which then issues alarms and warnings based on these results, notifying relevant personnel to handle the situation in real time. In case of current overload, the current is reduced to prevent equipment damage.

[0037] In summary, a smart fault detection system for metallurgical power systems based on multimodal data fusion has been developed.

[0038] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0039] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. 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. Such 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, and should all be included within the protection scope of the present invention.

Claims

1. A method for intelligent detection of power faults in metallurgical plants based on multimodal data fusion, characterized in that, Includes the following steps: S1: Acquire multimodal data of the smelting power system; S2: Perform feature extraction and optimization on the multimodal data to obtain optimized feature data; S3: Fuse the optimized feature data to obtain the fused comprehensive data; S4: Perform anomaly detection and analysis on the fused integrated data to obtain the detection results; S5: Issue warnings based on test results; In step S2, the feature data of the extracted multimodal data are processed as follows: The spacetime transformation process uses the following formula: ； in, It is the first Mode in time and spatial location Spatiotemporal feature data, using express; It is the first Each mode in time Preliminary characteristic data; Indicates the first The mapping vectors of the preliminary feature data of each modality along the time dimension are used to extract time series features. It is a time series mapping function; Represents convolution or nonlinear operators used to combine temporal and spatial information; It is the first Preliminary feature data of each modality in spatial location and time The spatiotemporal correlation characteristics on It is a spatial correlation function; A multi-level feature complexity optimization mechanism enables features of different modalities to have adaptive weights at different time periods. The feature complexity optimization formula is as follows: ； ； ； in, It is the optimized version of the first Feature data of each modality; It is the first The weighting factor of each mode represents the relative importance of that mode in the feature optimization process; It is the first The modality at time... The rating; It is the first The relative importance of each modality in calculating the score, with values ​​ranging from [value range missing]. Indicates the first The modality at time 1 The energy level reflects the fluctuation amplitude and activity level of modal data; It refers to the size of the time window, which is determined based on specific needs. It is the first The volatility weighting coefficients for each mode, with values ​​ranging from [value range missing]. It is the first The modality at time... The variance; It is the first Spatiotemporal characteristic data of modalities; It is the first The first mode and the first The interaction weight coefficients between modes reflect the degree of mutual influence between different modes, and their values ​​range from [value range missing]. Indicates inner product calculation; It represents the total number of modes.

2. The intelligent detection method for smelting power faults based on multimodal data fusion according to claim 1, characterized in that, The multimodal data includes one or more of the following: visual data, temperature data, electrical data, and vibration data.

3. The intelligent detection method for smelting power faults based on multimodal data fusion according to claim 1, characterized in that, The multimodal data in step S1 is obtained through multimodal data acquisition and multimodal data preprocessing. The preprocessing operation is one or more of the following: noise reduction, alignment and normalization.

4. The intelligent detection method for smelting power faults based on multimodal data fusion according to claim 2, characterized in that, In step S2, multimodal data feature extraction employs one or more of the following feature extraction methods: For visual data, convolutional neural networks are used to extract image features. Basic features in the image are extracted through multiple convolutional layers, and then preliminary dimensionality reduction is performed through pooling layers to extract discriminative features. For temperature data, the time-series features of the temperature data are extracted using statistical feature extraction methods, and the frequency domain features of the temperature data are analyzed using Fourier transform methods to capture potential abnormal fluctuations. For electrical data, a method combining time-domain and frequency-domain features is used for extraction to identify periodic or non-periodic fault modes. For vibration data, time-frequency analysis is used to extract frequency and vibration amplitude features from the vibration data to identify mechanical fault modes.

5. The intelligent detection method for smelting power faults based on multimodal data fusion according to claim 1, characterized in that, In step S3, a weighted fusion model is used to fuse the optimized feature data. The formula for calculating the fused features is as follows: ； in, It is the integrated data after fusion; It is a nonlinear fusion operation used to synthesize weighted optimization features and intermodal interaction terms.

6. The intelligent detection method for smelting power faults based on multimodal data fusion according to claim 1, characterized in that, In step S4, the abnormal fault detection and analysis includes: S41: Coarse screening, used to quickly screen and filter out normal data that is unlikely to be faulty; S42: Deep learning-enhanced detection employs deep neural networks for detailed classification to further uncover deeper potential fault features; it uses a deep neural network in the form of a multilayer perceptron to perform multi-level nonlinear mapping on the integrated data after initial screening through multiple hidden layers, thereby capturing complex anomaly patterns. S43: Result fusion, comprehensively considering the detection results of each stage, to arrive at the final fault detection result.

7. The intelligent detection method for smelting power faults based on multimodal data fusion according to any one of claims 1-6, characterized in that, It also includes step S6: after step S3, construct a smelting power fault prediction model, perform prediction processing on the integrated data after fusion, and obtain fault prediction results. Step S5 also provides early warning based on the fault prediction results.

8. A smart detection system for smelting power grid faults based on multimodal data fusion, used to implement the smart detection method for smelting power grid faults based on multimodal data fusion as described in any one of claims 1-6, characterized in that, include: The multimodal data acquisition module is configured to execute step S1; The multimodal data feature extraction and optimization module is configured to execute step S2; The multimodal data real-time fusion module is configured to execute step S3; The multi-level anomaly fault detection module is configured to execute step S4; The fault warning module is configured to execute step S5.

9. A smart detection system for smelting power grid faults based on multimodal data fusion, used to implement the smart detection method for smelting power grid faults based on multimodal data fusion as described in claim 7, characterized in that, include: The multimodal data acquisition module is configured to execute step S1; The multimodal data feature extraction and optimization module is configured to execute step S2; The multimodal data real-time fusion module is configured to execute step S3; The multi-level anomaly fault detection module is configured to execute step S4; The fault warning module is configured to execute step S5; The fault prediction module is configured to execute step S6.

Images