Multi-source heterogeneous data processing method, electronic equipment, storage medium and program product

Abstract

The invention provides a multi-source heterogeneous data processing method fusing physical priori knowledge, which comprises the following steps: acquiring multi-source heterogeneous data subjected to time-space synchronization and alignment, the multi-source heterogeneous data having a time dimension, a space dimension and a channel dimension; cross-modal tensor fusion is carried out on the multi-source heterogeneous data to obtain a fusion tensor, and the fusion tensor has a time dimension, a space dimension and a feature dimension; according to an interpretable primary function and a fusion tensor which are designed based on a physical process, sparse coding is used for iterative solution, and a sparse coefficient matrix is obtained; inputting the sparse coefficient matrix into a dynamic causal network, activating causal edges based on physical constraints, and outputting a node feature matrix and a causal adjacency matrix; performing weighted aggregation according to the node feature matrix and the causal adjacency matrix, and outputting an aggregated node feature matrix; and projecting the aggregation node feature matrix to the interpretable subspace to obtain interpretable features. The invention further provides electronic equipment, a storage medium and a program product.

Description

Technical Field

[0001] This application mainly relates to the field of intelligent security monitoring technology, and in particular to a multi-source heterogeneous data processing method, electronic device, storage medium and program product. Background Technology

[0002] For safety monitoring, the system's main control room generates tens of thousands of channels of real-time monitoring data per second, covering various physical parameters such as temperature, pressure, flow rate, and equipment status signals. This data is characterized by its multi-source nature, heterogeneity, dynamism, and strong noise interference. Traditional methods, employing filtering techniques and threshold alarm mechanisms, struggle to extract patterns in the state evolution of monitored objects from real-time monitoring data. Recent improvements have utilized deep learning and neural network technologies. However, existing technologies suffer from the following drawbacks: First, heterogeneous data fusion fails; current technologies only perform simple data alignment for data with different sampling rates and physical dimensions, leading to the loss of high-frequency details and broken physical correlations. Second, feature adaptation capability is lacking; current technologies use feature extraction models with fixed parameters, making it difficult to adapt to the dynamic evolution of data characteristics and lacking feedback optimization mechanisms. Third, physical interpretability is weak; current technologies often employ black-box models, making it difficult to correlate features extracted with the system's causal logic. These shortcomings result in discrepancies between the feature representations extracted by existing technologies and the true state of the system, thus affecting the reliability of subsequent feature-based analysis and diagnosis. Summary of the Invention

[0003] The technical problem to be solved by this application is to provide a multi-source heterogeneous data processing method that integrates physical prior knowledge, which enables the features obtained from multi-source heterogeneous data processing to reflect the true state of the object.

[0004] To address the aforementioned technical problems, this application provides a method for processing multi-source heterogeneous data that integrates prior physical knowledge, comprising the following steps: acquiring multi-source heterogeneous data that has undergone spatiotemporal synchronization and alignment, wherein the multi-source heterogeneous data has a time dimension, a spatial dimension, and a channel dimension, wherein the spatial dimension represents the monitoring position of the data in the object space, and the channel dimension represents the type of the physical quantity corresponding to the data; performing cross-modal tensor fusion on the multi-source heterogeneous data to obtain a fused tensor, wherein the fused tensor has the time dimension, the spatial dimension, and the feature dimension; using sparse coding iteratively to solve for a sparse coefficient matrix based on interpretable basis functions designed based on physical processes and the fused tensor, wherein the sparse coefficient matrix has the time dimension, the spatial dimension, and the sparse coefficient dimension; inputting the sparse coefficient matrix into a dynamic causal network, and activating causal edges based on physical constraints, outputting a node feature matrix and a causal adjacency matrix; performing weighted aggregation based on the node feature matrix and the causal adjacency matrix to output an aggregated node feature matrix; and projecting the aggregated node feature matrix onto an interpretable subspace to obtain interpretable features.

[0005] In one embodiment of this application, acquiring spatiotemporally synchronized and aligned multi-source heterogeneous data includes: sampling first data with a first frequency range using a variable time window to obtain a first multi-source heterogeneous data component, wherein the width of the variable time window is related to the instantaneous frequency of the first data; performing timestamp dynamic correction and physical constraint verification on second data with a second frequency range to obtain a second multi-source heterogeneous data component; and using a probabilistic graphical model to supplement the intermediate states of third data with a third frequency range to obtain a third multi-source heterogeneous data component; wherein the first frequency range is higher than the second frequency range, and the second frequency range is higher than the third frequency range, and the first to third multi-source heterogeneous data components together constitute the multi-source heterogeneous data.

[0006] In one embodiment of this application, the second data includes temperature data, and physical constraint verification of the second data includes aligning the temperature data using a coolant heat transfer delay model.

[0007] In one embodiment of this application, after performing cross-modal tensor fusion on the multi-source heterogeneous data to obtain the fused tensor, the method further includes: adaptive noise reduction on the fused tensor.

[0008] In one embodiment of this application, adaptive noise reduction of the fused tensor includes filtering the fused tensor in both the time domain and the frequency domain.

[0009] In one embodiment of this application, the method further includes determining the feature confidence level of the interpretable feature and dynamically allocating computing resources based on the feature confidence level.

[0010] In one embodiment of this application, the interpretable basis functions include basis functions based on thermal-hydraulic process design, basis functions based on neutron dynamics design, and basis functions based on mechanical vibration design.

[0011] This application also proposes an electronic device comprising: a memory for storing instructions executable by a processor; and a processor for executing the instructions to implement the method described above.

[0012] This application also proposes a computer storage medium storing computer program code that, when executed by a processor, implements the method described above.

[0013] This application also proposes a computer program product including computer program code, which, when executed by one or more processors, implements the steps described above.

[0014] Compared with the prior art, this application has the following advantages: (1) By using interpretable basis functions designed based on physical processes and activating causal edges based on physical constraints, the extracted features are associated with physical causal logic, so as to integrate physical prior knowledge into the multi-source heterogeneous data processing process, thereby improving the interpretability of feature representation; (2) By projecting onto the interpretable subspace, interpretable features are obtained, thereby realizing the reversible mapping from high-dimensional data to physical meaning, so that the features obtained from processing multi-source heterogeneous data reflect the true state of the object. Attached Figure Description

[0015] The accompanying drawings are included to provide a further understanding of this application; they are incorporated into and constitute a part of this application. The drawings illustrate embodiments of this application and, together with this specification, serve to explain the principles of this application. In the drawings: Figure 1 This is a flowchart illustrating a multi-source heterogeneous data processing method that integrates prior physical knowledge in one embodiment of this application. Figure 2 This is a schematic diagram of the spatiotemporal synchronization and alignment process in one embodiment of this application; Figure 3 This is a schematic flowchart of adaptive noise reduction in one embodiment of this application; Figure 4 This is a flowchart illustrating a multi-source heterogeneous data processing method that integrates prior physical knowledge in another embodiment of this application. Figure 5 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some examples or embodiments of this application. For those skilled in the art, these drawings can be applied to other similar scenarios without creative effort. Unless obvious from the context or otherwise specified, the same reference numerals in the drawings represent the same structures or operations.

[0017] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0018] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this application. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following drawings denote similar items; therefore, once an item is defined in one drawing, it need not be further discussed in subsequent drawings.

[0019] Furthermore, it should be noted that the use of terms such as "first" and "second" to define components is merely for the purpose of distinguishing the corresponding components. Unless otherwise stated, these terms have no special meaning and therefore should not be construed as limiting the scope of protection of this application. In addition, although the terminology used in this application is selected from commonly known and used terms, some terms mentioned in this application's specification may have been chosen by the applicant according to his or her judgment, and their detailed meanings are explained in the relevant sections of this description. Moreover, this application should be understood not only through the actual terms used, but also through the meaning implied by each term.

[0020] It should be understood that when a component is referred to as "on another component," "connected to another component," "coupled to another component," or "in contact with another component," it can be directly on, connected to, coupled to, or in contact with that other component, or there may be an intervening component. In contrast, when a component is referred to as "directly on another component," "directly connected to," "directly coupled to," or "directly in contact with" another component, there is no intervening component. Similarly, when a first component is referred to as "electrically contacting" or "electrically coupled to" a second component, there is an electrical path between the first and second components that allows current to flow. This electrical path may include capacitors, coupled inductors, and / or other components that allow current to flow, even if there is no direct contact between the conductive components.

[0021] This application uses flowcharts to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously. Furthermore, other operations may be added to these processes, or one or more steps may be removed from them. Next, the technical solutions of this application will be described through specific embodiments.

[0022] refer to Figure 1 The flowchart shown in one embodiment illustrates a multi-source heterogeneous data processing method that integrates prior physical knowledge. The multi-source heterogeneous data processing method includes the following steps.

[0023] Step S101: Obtain multi-source heterogeneous data that has been spatiotemporally synchronized and aligned. The multi-source heterogeneous data has time dimension, spatial dimension and channel dimension. The spatial dimension represents the monitoring position of the data in the object space, and the channel dimension represents the type of physical quantity corresponding to the data.

[0024] Specifically, spatiotemporal synchronization and alignment processing achieves simultaneity and co-location of multiple data streams in time and space. In other words, the physical quantities in each channel of the time-synchronized and aligned multi-source heterogeneous data correspond to the same time and space. Multi-source heterogeneous data comprises three dimensions: time dimension, spatial dimension, and channel dimension. It should be noted that this application does not limit the order of the three dimensions of multi-source heterogeneous data; the order can be time dimension, spatial dimension, and channel dimension, or time dimension, channel dimension, and spatial dimension. It can be understood that the monitoring location represented by the spatial dimension can be either a relative or absolute location. In some embodiments, the monitoring object is a nuclear power plant, and the types of physical quantities include temperature, pressure, flow rate, neutron flux, radiation dose, and equipment status. In this embodiment, the channel dimension represents temperature, pressure, flow rate, neutron flux, radiation dose, and equipment status.

[0025] Different physical quantities in multi-source heterogeneous data have vastly different sampling frequencies, ranging from milliseconds to minutes. To ensure uniform processing of these physical quantities in subsequent steps, the original sampled data for these physical quantities undergoes spatiotemporal synchronization and alignment. (Reference) Figure 2 The diagram shows a flow chart of spatiotemporal synchronization and alignment in one embodiment. Spatiotemporal synchronization and alignment includes the following steps.

[0026] Step S201: Sample first data with a first frequency range using a variable time window to obtain a first multi-source heterogeneous data component, wherein the width of the variable time window is related to the instantaneous frequency of the first data.

[0027] Specifically, the first data is time-series data obtained by sampling the original signal within a first frequency range. The original signal is a physical quantity signal monitored at a monitoring location in the object space. The first data includes timestamps arranged in chronological order and corresponding signal values ​​for those timestamps. The first multi-source heterogeneous data component is time-series data arranged in chronological order and has a corresponding object space monitoring location and a corresponding physical quantity type.

[0028] In some embodiments, a variable time window is used to sample the first data. Within each time window, the first data is aggregated or downsampled to obtain a first multi-source heterogeneous data component. The width of the variable time window is dynamically adjusted according to the instantaneous frequency of the first data. The relationship between the width of the variable time window and the instantaneous frequency of the first data is expressed by the following formula: In the formula, for The width of the time window can be adjusted at any time. for The instantaneous dominant frequency of the first data point at any given time. In this way, the first multi-source heterogeneous data component obtained through variable time window sampling can retain transient change characteristics.

[0029] Step S202: Perform timestamp dynamic correction and physical constraint verification on the second data with the second frequency range to obtain the second multi-source heterogeneous data component.

[0030] It is understandable that the second data differs from the first data in frequency range; the second data is time-series data obtained by sampling the original signal within a second frequency range. The second multi-source heterogeneous data component is time-series data arranged in chronological order, and it has corresponding object space monitoring locations and corresponding physical quantity types. In this step, the second data is dynamically corrected using timestamps, which can eliminate the deviation of timestamps at different monitoring locations in the object space caused by transmission and acquisition delays, ensuring clock consistency of the second data. For example, a Precision Time Protocol (PTP) is used to achieve μs-level time synchronization between the second data. Simultaneously, physical constraint verification is used to process the second data, ensuring that the processed second multi-source heterogeneous data component still conforms to the physical model (such as a thermal-hydraulic model).

[0031] In some embodiments, the second data includes temperature data, and physical constraint verification of the second data includes aligning the temperature data using a coolant heat transfer delay model. For example, in a nuclear power plant object space, it takes a certain amount of time for coolant to flow from the reactor core to the steam generator, which depends on factors such as coolant flow rate, pipe length, and heat capacity. The theoretical delay time for coolant to flow from the reactor core to the steam generator is calculated using a coolant heat transfer delay model (such as a one-dimensional heat transfer equation), and dynamic time warping is used to align the temperature data in the second data corresponding to the monitoring location of the reactor core and the temperature data corresponding to the monitoring location of the steam generator. In other embodiments, after calculating the theoretical delay time, cross-correlation analysis can also be used to align the temperature data.

[0032] Step S203: Use a probabilistic graphical model to complete the intermediate states of the third data with the third frequency range to obtain the third multi-source heterogeneous data component.

[0033] It is understood that the third data differs in frequency range from the first and second data. The third data is time-series data obtained by sampling the original signal within the third frequency range. The third multi-source heterogeneous data component is time-series data arranged in chronological order, and has a corresponding object space monitoring location and a corresponding physical quantity type. In this step, a probabilistic graphical model (PGM) is used for upsampling, that is, to infer and complete the intermediate states between adjacent timestamps of the third data, thereby completing the evolution path of the third data within the sampling interval of the third frequency range to obtain the completed third multi-source heterogeneous data component. In some embodiments, in the nuclear power plant object space, the reactor system is modeled as a dynamic Bayesian network on the time axis using a probabilistic graphical model. Each timestamp contains latent variables (i.e., physical states that cannot be directly monitored) and the third data at the corresponding time. The most probable distribution of the latent variables is calculated based on the prior probability and the posterior probability using a Bayesian algorithm. The specific implementation of the probabilistic graphical model is not the focus of this application; please refer to relevant technologies, and it will not be elaborated here.

[0034] In embodiments of this application, the first frequency range is higher than the second frequency range, and the second frequency range is higher than the third frequency range. This creates a three-level cascaded spatiotemporal synchronization and alignment process, enabling adaptive processing of the original signals sampled from various frequency ranges. In some embodiments, the first frequency range is a millisecond-level sampling frequency, the second frequency range is a second-level sampling frequency, and the third frequency range is a minute-level sampling frequency.

[0035] The first to third multi-source heterogeneous data components are all time-series data, and each has a corresponding object space monitoring location and a corresponding physical quantity type. The first to third multi-source heterogeneous data components are combined to form multi-source heterogeneous data based on the time dimension, the object space monitoring location, and the physical quantity type in the channel dimension.

[0036] Step S102: Perform cross-modal tensor fusion on multi-source heterogeneous data to obtain a fused tensor, which has time dimension, spatial dimension and feature dimension.

[0037] In this step, the input to the cross-modal tensor fusion layer is multi-source heterogeneous data, and the output is a fused tensor. The cross-modal tensor fusion layer fuses information from several dimensions (one, two, or three dimensions) of the multi-source heterogeneous data. For example, it might fuse information from the channel dimension while preserving the time and spatial dimensions independently; or it might fuse information from the spatial and channel dimensions while preserving the time dimension independently; or it might fuse information from the time, space, and channel dimensions. It can be understood that the number of feature dimensions of the fused tensor can be the same as or different from the number of channel dimensions of the multi-source heterogeneous data.

[0038] In some embodiments, after step S102, the method further includes: adaptive noise reduction of the fused tensor.

[0039] In some embodiments, adaptive denoising of the fused tensor includes filtering the fused tensor in both the time and frequency domains. (See reference) Figure 3 The flowchart of adaptive noise reduction in one embodiment shown includes the following steps.

[0040] Step S301 involves separating noise components from valid components in the time domain using nonlinear variational mode decomposition. Specifically, for each vector along the time dimension in the fused tensor, nonlinear variational mode decomposition is applied to obtain each modal component. The kurtosis of each modal component is calculated, with the modal component having higher kurtosis being the noise component. Among all modal components, those with energy entropy below a preset threshold are retained as valid components, and the valid components are reconstructed into vectors along the time dimension.

[0041] Step S302 involves using causal coherence analysis in the frequency domain to suppress unrelated noise components. Specifically, each vector along the time dimension in the fusion tensor is decomposed to obtain the frequency band components. The Granger causality test is used to confirm the coherence of each frequency band component, suppressing frequency band components that cannot truly reflect physical correlations, and reconstructing each suppressed frequency band component into a vector along the time dimension.

[0042] Step S303 employs an adaptive mechanism to dynamically adjust the filtering parameters based on the filtered signal-to-noise ratio (SNR). For example, the SNR of the vector reconstructed along the time dimension after steps S301 and S302 is calculated. Based on the SNR, the fundamental and nonlinear filtering parameters in the nonlinear variational mode decomposition are adjusted, such as the number of modal components, bandwidth control parameters, noise tolerance, and nonlinear constraint weights, thereby balancing denoising intensity and fidelity.

[0043] It is important to note that Figure 3In the illustrated embodiment, steps S301 and S302 (which can be interchanged) are executed serially. In other embodiments, time-domain filtering in step S301 and frequency filtering in step S302 can be executed in parallel, and the outputs of steps S301 and S302 are synchronously integrated. Specific integration methods include, but are not limited to, average integration and weighted integration.

[0044] Step S103: Based on the interpretable basis function and fused tensor designed based on the physical process, sparse coding is used to iteratively solve the problem to obtain the sparse coefficient matrix, which has time dimension, spatial dimension and sparse coefficient dimension.

[0045] In this step, interpretable basis functions are designed based on the physical process. Then, based on these interpretable basis functions and the fused tensor obtained in step S102, sparse coding is used for iterative solving. This involves converting the representation of the original fused tensor in the vector space composed of time, space, and channel dimensions into a representation in the vector space composed of time, space, and sparse coefficient dimensions. The iterative objective is to minimize the error between the reconstructed fused tensor based on the interpretable basis functions and the sparse coefficient matrix and the original fused tensor, outputting the iteratively obtained sparse coefficient matrix. The formula for the iterative objective is as follows: In the formula, For the original fusion tensor, It is a library of interpretable basis functions. It is a sparse coefficient matrix. This is the regularization coefficient. In some embodiments, in nuclear engineering... The value ranges from 0.05 to 0.2.

[0046] In some embodiments, interpretable basis functions include basis functions based on thermal-hydraulic process design, basis functions based on neutron dynamics design, and basis functions based on mechanical vibration design. For example, basis functions based on thermal-hydraulic process design include basis functions used to describe the oscillating decay process of coolant temperature; basis functions based on neutron dynamics design include basis functions used to simulate transient changes in neutron flux; and basis functions based on mechanical vibration design include basis functions used to characterize the vibration modes of the main pump or steam generator.

[0047] In some embodiments, after step S103, an online feature quality evaluator is deployed, and the aforementioned parameters are updated using new data while retaining historical knowledge. For example, the new data is used to collaboratively optimize the parameters and interpretable basis function selection of the adaptive denoising step. The feature quality evaluator uses multi-dimensional metrics to quantify the representation quality of the sparse coefficient matrix to monitor the effectiveness of the sparse coefficient matrix obtained in step 103 in real time. The multi-dimensional metrics include signal-to-noise ratio, physical consistency, and modal coherence.

[0048] Step S104: Input the sparse coefficient matrix into the dynamic causal network, activate the causal edges based on physical constraints, and output the node feature matrix and the causal adjacency matrix.

[0049] In this step, the input to the dynamic causal network is a sparse coefficient matrix, and the output is a node feature matrix and a causal adjacency matrix. The dynamic causal network activates causal edges in the causal adjacency matrix based on physical constraints. In some embodiments, during nuclear power plant monitoring, the dynamic causal network determines necessary causal relationships according to the thermo-hydraulic equations. Specifically, a method combining time-varying Granger causality analysis and convergent cross mapping (CCM) is used to monitor the activation coefficients of key causal edges in the dynamic causal network in real time. The dynamic causal network is not the focus of this application; please refer to related technologies for further details.

[0050] Step S105: Perform weighted aggregation based on the node feature matrix and the causal adjacency matrix to output the aggregated node feature matrix.

[0051] In this step, the topology of the dynamic causal network from step S104 is used to guide the weighted fusion of node features along causal edges, thereby achieving feature correlation constraints. The topology of the dynamic causal network consists of a node feature matrix and a causal adjacency matrix. In some embodiments, a graph attention network (GAT) is constructed to achieve weighted feature fusion, with the node feature matrix and causal adjacency matrix as inputs and an aggregated node feature matrix as output.

[0052] Step S106: Project the aggregated node feature matrix to the interpretable subspace to obtain interpretable features, so as to realize the mapping from high-dimensional data to physical meaning.

[0053] In some embodiments, physical kernel principal component analysis is used to define a dedicated interpretable feature space for nuclear accident diagnosis. The aggregated node feature matrix is ​​projected onto this interpretable space to obtain the final interpretable features, thereby achieving the mapping from high-dimensional data to physical meaning. For example, the interpretable subspace has a time dimension and an interpretable feature dimension, each of which has actual physical meaning. The interpretable features are located in the interpretable subspace, and their elements correspond to physically meaningful feature values ​​at each time step.

[0054] refer to Figure 4 As shown, in some embodiments, the multi-source heterogeneous data processing method that integrates physical prior knowledge includes steps S401 to S407, wherein steps S401 to S406 correspond one-to-one with the above steps S101 to S106, and will not be repeated here.

[0055] In this embodiment, step S407 determines the feature confidence level of interpretable features and dynamically allocates computing resources based on the feature confidence level. Specifically, the feature confidence level includes three confidence dimensions: physical reliability, statistical stability, and model consistency. In some embodiments, in nuclear power plant monitoring, the feature confidence level is calculated comprehensively from these three confidence dimensions (e.g., average, weighted average, etc.). Specifically, the physical reliability of the feature is obtained through constraint verification based on the nuclear power plant's physical equations; statistical stability is obtained by calculating the statistical properties of the feature values ​​through a sliding window; and model consistency is obtained by comparing the prediction results of multiple independent models. High-confidence interpretable features are transmitted in a simplified mode (e.g., 0 / 1 identifier), while low-confidence interpretable features are allocated computing resources for in-depth analysis (e.g., calling step S403 for enhanced processing). It can be understood that the object of allocating computing resources in this step is the newly added data.

[0056] In some embodiments, the multi-source heterogeneous data processing method incorporating prior physical knowledge further includes: establishing a security early warning and resource overrun feedback mechanism. For example, when the method proposed in this application has a delay exceeding 50ms, the processing accuracy of non-critical channels is automatically reduced. Specifically, this may involve using a simplified mode (such as 0 / 1 identification) to transmit data from non-critical channels.

[0057] An embodiment of this application also proposes a method such as Figure 5 The electronic device 500 is shown. According to... Figure 5 The electronic device 500 may include an internal communication bus 501, a processor 502, a read-only memory (ROM) 503, a random access memory (RAM) 504, and a communication port 505. When used in a personal computer, the electronic device may also include a hard disk 506.

[0058] The internal communication bus 501 enables data communication between components of the electronic device 500. The processor 502 can perform judgments and issue prompts. In some embodiments, the processor 502 may consist of one or more processors. The communication port 505 enables data communication between the electronic device 500 and external devices. In some embodiments, the electronic device 500 can send and receive information and data from a network through the communication port 505.

[0059] Electronic device 500 may also include different forms of program storage units and data storage units, such as hard disk 506, read-only memory (ROM) 503, and random access memory (RAM) 504, capable of storing various data files used for computer processing and / or communication, as well as possible program instructions executed by processor 502. The processor executes these instructions to implement the main parts of the methods described above. The results of processor processing are transmitted to user equipment via a communication port and displayed on a user interface.

[0060] This application also proposes a computer-readable medium storing computer program code that, when executed by a processor, implements the method described above.

[0061] In addition, this application also proposes a computer program product, including computer program code, which, when executed by one or more processors, enables the implementation of the steps described above.

[0062] The basic concepts have been described above. Obviously, for those skilled in the art, the above disclosure is merely illustrative and does not constitute a limitation of this application. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and corrections to this application. Such modifications, improvements, and corrections are suggested in this application, and therefore remain within the spirit and scope of the exemplary embodiments of this application.

[0063] Furthermore, this application uses specific terms to describe embodiments of the application. For example, "an embodiment," "one embodiment," and / or "some embodiments" refer to a particular feature, structure, or characteristic related to at least one embodiment of the application. Therefore, it should be emphasized and noted that "an embodiment," "one embodiment," or "an alternative embodiment" mentioned twice or more in different locations in this specification do not necessarily refer to the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of the application can be appropriately combined.

[0064] Similarly, it should be noted that, in order to simplify the description of the present application and thus aid in the understanding of one or more embodiments, the foregoing description of the embodiments of the present application sometimes combines multiple features into a single embodiment, drawing, or description thereof. However, this disclosure method does not imply that the subject matter of the present application requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of the single embodiments disclosed above.

[0065] In some embodiments, numbers describing the quantity of components and attributes are used. It should be understood that such numbers used in the description of embodiments are modified in some examples with the terms "approximately," "approximately," or "generally." Unless otherwise stated, "approximately," "approximately," or "generally" indicates that the numbers are allowed to vary by ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, which may be changed depending on the characteristics required by individual embodiments. In some embodiments, numerical parameters should take into account specified significant digits and employ a general method of digit reservation. Although the numerical ranges and parameters used to confirm their breadth of scope in some embodiments of this application are approximate values, in specific embodiments, such values ​​are set as precisely as feasible.

[0066] Although this application has been described with reference to specific embodiments, those skilled in the art should recognize that the above embodiments are only used to illustrate this application, and various equivalent changes or substitutions can be made without departing from the spirit of this application. Therefore, any changes or modifications to the above embodiments within the essential spirit of this application will fall within the scope of the claims of this application.

Claims

1. A method for processing multi-source heterogeneous data that integrates prior physical knowledge, characterized in that, Includes the following steps: Acquire multi-source heterogeneous data that has been spatiotemporally synchronized and aligned. The multi-source heterogeneous data has a time dimension, a spatial dimension, and a channel dimension. The spatial dimension represents the monitoring position of the data in the object space, and the channel dimension represents the type of physical quantity corresponding to the data. The multi-source heterogeneous data is fused across modalities to obtain a fused tensor, which has the time dimension, the spatial dimension, and the feature dimension. Based on the interpretable basis function designed based on the physical process and the fusion tensor, a sparse coefficient matrix is ​​obtained by iterative solution using sparse coding, and the sparse coefficient matrix has the time dimension, the spatial dimension and the sparse coefficient dimension. The sparse coefficient matrix is ​​input into the dynamic causal network, and causal edges are activated based on physical constraints, outputting the node feature matrix and the causal adjacency matrix; The node feature matrix and the causal adjacency matrix are weighted and aggregated to output the aggregated node feature matrix; as well as The aggregated node feature matrix is ​​projected onto the interpretable subspace to obtain interpretable features.

2. The method as described in claim 1, characterized in that, The acquisition of spatiotemporally synchronized and aligned multi-source heterogeneous data includes: A first multi-source heterogeneous data component is obtained by sampling first data with a first frequency range using a variable time window, wherein the width of the variable time window is related to the instantaneous frequency of the first data. The second data with a second frequency range is subjected to dynamic timestamp correction and physical constraint verification to obtain the second multi-source heterogeneous data component. The intermediate states of the third data with the third frequency range are supplemented by using a probabilistic graphical model to obtain the third multi-source heterogeneous data component. The first frequency range is higher than the second frequency range, and the second frequency range is higher than the third frequency range. The first to third multi-source heterogeneous data components together constitute the multi-source heterogeneous data.

3. The method as described in claim 2, characterized in that, The second data includes temperature data, and physical constraint verification of the second data includes aligning the temperature data using a coolant heat transfer delay model.

4. The method as described in claim 1, characterized in that, After fusing the multi-source heterogeneous data across modalities to obtain the fused tensor, the method further includes: adaptive noise reduction of the fused tensor.

5. The method as described in claim 4, characterized in that, Adaptive denoising of the fused tensor includes filtering the fused tensor in both the time and frequency domains.

6. The method as described in claim 1, characterized in that, Also includes: Determine the feature confidence level of the interpretable feature, and dynamically allocate computing resources based on the feature confidence level.

7. The method as described in claim 1, characterized in that, The interpretable basis functions include basis functions based on thermal-hydraulic process design, basis functions based on neutron dynamics design, and basis functions based on mechanical vibration design.

8. An electronic device, comprising: Memory is used to store instructions that can be executed by the processor; as well as A processor for executing the instructions to implement the method as described in any one of claims 1-7.

9. A computer storage medium storing computer program code, said computer program code implementing the method as claimed in any one of claims 1-7 when executed by a processor.

10. A computer program product comprising computer program code, wherein when the computer program code is executed by one or more processors, the one or more processors implement the steps of the method as described in any one of claims 1-7.

Images