A method, apparatus, device, and medium for reconstructing the state of a distributed parameter system guided by sparse sensing information.
By constructing global and local reconstruction models and combining sparse sensing information and sensor layout, a multi-scale joint reconstruction field is generated, which solves the reconstruction accuracy problem of distributed parameter systems under sparse observation conditions, realizes the coordinated reconstruction of global trends and local details, and improves reconstruction accuracy and efficiency.
Patent Information
- Application Number
- CN202611303463.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-26
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2046-08-26
AI Technical Summary
Under sparse observation conditions, existing technologies lack an effective collaborative judgment mechanism for the global reconstruction model of distributed parameter systems, leading to inconsistencies in adjacent local corrections and affecting reconstruction accuracy.
By constructing a global reconstruction model and a local reconstruction model associated with multiple sensor subdomains, and using online observation data and sensor layout information, local triggering indicators and collaborative triggering indicators are determined. Local reconstruction models with activation intensity greater than a preset threshold are used to generate a correction field, and the global and local reconstruction fields are fused.
It significantly improves the accuracy of full-field state reconstruction under sparse observation conditions, ensures multi-scale collaborative reconstruction of global trends and local details, reduces unnecessary computational overhead, and enhances the reliability and computational efficiency of the reconstruction process.
Smart Images

Figure CN122818989B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of state reconstruction technology, and in particular to a method, apparatus, device and medium for state reconstruction of a distributed parameter system guided by sparse sensing information. Background Technology
[0002] Distributed parameter systems (DPS) are a class of infinite-dimensional systems in which state variables not only evolve over time but also depend on continuous spatial coordinates. They are widely found in physical fields such as temperature, concentration, and stress fields in process industries. Due to limitations such as high temperature and pressure, enclosed structures, and deployment costs, DPS systems can only acquire sparse sensor observation data from a very small number of discrete locations under actual operating conditions. Therefore, state reconstruction techniques are needed—that is, using limited sparse observation information to invert and reconstruct the complete continuous state distribution over the entire spatial domain.
[0003] Existing technologies typically acquire historical full-field state data offline through high-density measurement point acquisition or numerical simulation, and then construct a full-field state reconstruction model based on this data. Related reconstruction methods often employ reduced-order models or data-driven models to establish an overall mapping relationship from sparse observation data to the full-field state. Some technical solutions further introduce local correction strategies to improve the reconstruction accuracy of specific regions.
[0004] In existing methods that introduce local corrections, the activation and correction decisions of each local model are usually made independently, ignoring the spatial continuity of the distributed parameter system and the coupling relationship between adjacent regions. The lack of an effective collaborative judgment mechanism can easily lead to incoordination between adjacent local corrections, thereby affecting the reconstruction accuracy.
[0005] Therefore, how to achieve multi-scale collaborative reconstruction of global trends and local details under sparse observation conditions, while enabling the activation decisions of each local correction model to make collaborative judgments using spatial correlation information between adjacent regions, has become an urgent technical problem to be solved. Summary of the Invention
[0006] In view of this, the purpose of this invention is to overcome the shortcomings of the prior art and provide a state reconstruction method, apparatus, device, and medium for distributed parameter systems that can improve the accuracy of full-field state reconstruction. This invention provides the following technical solution: In a first aspect, this application provides a method for reconstructing the state of a distributed parameter system guided by sparse sensing information, the method comprising: Acquire the historical full-field state dataset, online observation dataset, and sensor layout information of the distributed parameter system. The historical full-field state dataset includes multiple historical local state datasets. A global reconstruction model is constructed based on the aforementioned historical full-field state dataset; Based on the online observation dataset and the global reconstruction model, the global reconstruction field is determined; After dividing the full-field space region where the distributed parameter system is located into multiple sensor-related subdomains based on the sensor layout information, a local reconstruction model corresponding to each sensor-related subdomain is constructed based on each historical local state dataset. Based on the online observation dataset and the global reconstruction field, the local triggering indicators for each sensor-related subdomain are determined; Based on the local triggering indicators and the sensor layout information, the collaborative triggering indicators of the associated subdomains of each sensor are determined; Based on the aforementioned collaborative triggering indicators, the activation intensity of each of the aforementioned local reconstruction models is determined. According to the activation intensity, a target local reconstruction model is called from multiple local reconstruction models to generate a local correction field. The local correction field and the global reconstruction field are fused to obtain a multi-scale joint reconstruction field. The target local reconstruction model is a local reconstruction model whose activation intensity is greater than a preset intensity threshold.
[0007] In one embodiment, the historical full-field state dataset includes: historical observation datasets at different historical moments, and the construction of a global reconstruction model based on the historical full-field state dataset includes: The historical full-field state dataset is subjected to eigenorthogonal decomposition to obtain the global space basis function matrix; Extract the global dominant space basis function matrix from the global space basis function matrix; Based on the global dominant space basis function matrix and the historical full-field state dataset, determine the global time coefficients corresponding to different historical moments; The historical observation datasets at different historical moments are used as input data, and the global time coefficients at the corresponding historical moments are used as output data to train the mapping model. The trained mapping model is then determined as the time coefficient prediction model. The global reconstruction model is determined based on the global dominant spatial basis function matrix and the time coefficient prediction model.
[0008] In one embodiment, each of the historical local state datasets corresponds to a different sensor association subdomain, and the step of constructing a local reconstruction model corresponding to each sensor association subdomain based on each of the historical local state datasets includes: The historical local state datasets corresponding to each of the sensor-associated subdomains are subjected to intrinsic orthogonal decomposition to obtain the local space basis function matrix corresponding to each of the sensor-associated subdomains. Extract the local dominant spatial basis function matrix corresponding to each sensor-related subdomain from each of the local spatial basis function matrices; Based on the local dominant space basis function matrices, the local reconstruction models corresponding to the sensor-related subdomains are determined respectively.
[0009] In one embodiment, the online observation dataset includes: online observation values corresponding to each of the sensor-associated subdomains; the step of determining the local triggering index for each of the sensor-associated subdomains based on the online observation dataset and the global reconstruction field includes: Extract the predicted values corresponding to each sensor-related subdomain from the global reconstruction field; For any of the sensor-associated subdomains, the difference between the corresponding online observation value and the predicted value is determined as the observation residual value of the corresponding sensor-associated subdomain. Based on the observation residual values at multiple consecutive time points, the local trigger index of each sensor-associated subdomain is determined.
[0010] In one embodiment, determining the collaborative triggering index for each sensor-associated subdomain based on each of the local triggering indices and the sensor layout information includes: A sensor adjacency structure graph is constructed based on the sensor layout information. The sensor adjacency graph includes the spatial adjacency relationships between the associated subdomains of each sensor. Based on the sensor adjacency structure graph, the adjacent associated subdomains of each sensor associated subdomain are determined; For any of the sensor-associated subdomains, the local triggering index of the sensor-associated subdomain is weighted and aggregated with the local triggering index of its adjacent associated subdomains to obtain the collaborative triggering index of the sensor-associated subdomain.
[0011] In one embodiment, determining the activation intensity of each local reconstruction model based on each of the collaborative triggering indicators includes: The time memory processing is performed on each of the aforementioned collaborative triggering indicators to obtain the memory triggering indicators of each of the aforementioned sensor-related subdomains; For each of the sensor-associated subdomains, the activation intensity of the local reconstruction model corresponding to each sensor-associated subdomain is determined based on the corresponding memory trigger index, the preset start activation threshold, and the saturation activation threshold.
[0012] In one embodiment, generating a local correction field by calling a target local reconstruction model from among the plurality of local reconstruction models according to each of the said activation intensities includes: Using the local correction coefficients of the target local reconstruction model as optimization variables, a target optimization function containing observation consistency constraints, spatial consistency constraints, and temporal continuity constraints is constructed. Solving the objective optimization function yields the objective local correction coefficients of the objective local reconstruction model; The local correction field is generated based on the target local reconstruction model and the target local correction coefficients.
[0013] Secondly, this application provides a sparse sensing information-guided distributed parameter system state reconstruction device, the device comprising: The data acquisition module is used to acquire the historical full-field state dataset, online observation dataset, and sensor layout information of the distributed parameter system. The historical full-field state dataset includes multiple historical local state datasets. The global model determination module is used to construct a global reconstruction model based on the historical full-field state dataset. The global reconstruction module is used to determine the global reconstruction field based on the online observation dataset and the global reconstruction model; The local model determination module is used to divide the full-field space region where the distributed parameter system is located into multiple sensor-related subdomains based on the sensor layout information, and then construct a local reconstruction model corresponding to each sensor-related subdomain based on each historical local state dataset. The local indicator determination module is used to determine the local trigger indicators of each sensor-related subdomain based on the online observation dataset and the global reconstruction field. The collaborative indicator determination module is used to determine the collaborative triggering indicator of each sensor-related subdomain based on each of the local triggering indicators and the sensor layout information; The fusion module is used to determine the activation intensity of each of the local reconstruction models based on the respective collaborative triggering indicators, and to call a target local reconstruction model from multiple local reconstruction models according to the activation intensity to generate a local correction field. The local correction field and the global reconstruction field are fused to obtain a multi-scale joint reconstruction field, wherein the target local reconstruction model is a local reconstruction model whose activation intensity is greater than a preset intensity threshold.
[0014] Thirdly, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the distributed parameter system state reconstruction method guided by sparse sensing information as described in the first aspect.
[0015] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the sparse sensing information-guided distributed parameter system state reconstruction method as described in the first aspect.
[0016] One or more technical solutions provided in this application have at least the following technical effects: By constructing a global reconstruction model and local reconstruction models corresponding to multiple sensor-related subdomains, the local reconstruction models are used to compensate for deviations and detailed changes in each subdomain, based on the global model ensuring the accuracy of the overall evolution trend of the physical field. This effectively solves the inherent defect that a single global model cannot take into account both the overall trend and local non-uniform features.
[0017] Based on online observation data and global reconstruction results, local triggering indicators for each sensor-related subdomain are determined. Furthermore, by combining sensor layout information, spatial correlation information of adjacent subdomains is introduced into the construction of collaborative triggering indicators, enabling activation decisions of each local reconstruction model to make collaborative judgments using the physical coupling relationship between adjacent regions, which significantly improves the accuracy and robustness of activation decisions. The activation intensity of each local reconstruction model is determined based on the collaborative triggering index, realizing on-demand differentiated activation of local corrections—the corresponding local reconstruction model is called with appropriate intensity only in the region that needs correction, avoiding unnecessary computational overhead and improving the computational efficiency of online reconstruction. By identifying local reconstruction models with activation strength greater than a preset activation strength as target local reconstruction models and generating local correction fields, unnecessary online calls to local reconstruction models can be reduced, enhancing the reliability of the reconstruction process. Finally, the local correction fields are fused with the global reconstruction fields, so that the reconstruction results maintain both the macroscopic evolution characteristics of the global scale and incorporate the refined information of the local scale. At the same time, the spatial coordination of corrections between adjacent subdomains is ensured by the cooperative triggering mechanism, significantly improving the accuracy of the full-field state reconstruction under sparse observation conditions. Attached Figure Description
[0018] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope of protection of the present invention. In the various drawings, similar components are numbered similarly.
[0019] Figure 1 A flowchart illustrating the state reconstruction method for a distributed parameter system provided in an embodiment of this application is shown. Figure 2 A schematic diagram of the state reconstruction device for a distributed parameter system provided in an embodiment of this application is shown. Figure 3 A schematic diagram of the structure of a computer device provided in an embodiment of this application is shown. Detailed Implementation
[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.
[0021] The components of the embodiments of the invention described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.
[0022] In the following, the terms “comprising,” “having,” and their cognates, which may be used in various embodiments of the invention, are intended only to indicate a particular feature, number, step, operation, element, component, or combination thereof, and should not be construed as excluding, firstly, the presence of one or more other features, numbers, steps, operations, elements, components, or combinations thereof, or adding the possibility of one or more features, numbers, steps, operations, elements, components, or combinations thereof.
[0023] Furthermore, the terms "first," "second," and "third" are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.
[0024] Unless otherwise specified, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which the various embodiments of the invention pertain. Terms (such as those defined in commonly used dictionaries) shall be interpreted as having the same meaning as in their contextual meaning in the relevant technical field and shall not be interpreted as having an idealized or overly formal meaning, unless clearly defined in the various embodiments of the invention.
[0025] Example 1 In existing technologies, methods that introduce local correction strategies based on a single global reconstruction model to improve the reconstruction accuracy of specific regions typically involve independent activation and correction strategies for each local model. This neglects the spatial continuity of the distributed parameter system and the coupling relationship between adjacent regions, lacking an effective collaborative judgment mechanism. This can easily lead to inconsistencies in adjacent local corrections, thereby affecting reconstruction accuracy. For further details, please refer to [link to relevant documentation]. Figure 1This application provides a method, apparatus, device and medium for reconstructing the state of a distributed parameter system guided by sparse sensing information, including steps S110 to S170.
[0026] Step S110: Obtain the historical full-field state dataset, online observation dataset, and sensor layout information of the distributed parameter system. The historical full-field state dataset includes multiple historical local state datasets.
[0027] Distributed parameter systems refer to a class of infinite-dimensional systems in which state variables not only evolve over time but also depend on continuous spatial coordinates. They are widely present in physical fields such as concentration fields and stress fields in process industries. Taking a volatilization kiln in a process industry as an example, the distributed parameter system to be reconstructed can be the temperature field inside the kiln, which is continuously distributed in space and dynamically evolves over time. The historical full-field state dataset consists of pre-collected snapshots of the full-field state at multiple different historical moments. Specifically, it includes the full-field spatial region where the distributed parameter system is located. The data is discretized into M spatial nodes, and the state values of the M spatial nodes are collected at N different historical moments to form a historical full-field state dataset. ,in, This represents the full-field state vector at the i-th historical moment. The online observation dataset is collected in real-time during the actual operation of the distributed parameter system, for example, within the full-field spatial region where the distributed parameter system is located. Internal fabric The sensor at a fixed position, the first The positions of the sensors are denoted as At any moment Each sensor reads its current state value, forming an online observation dataset. ,in, Indicates the first Each sensor at time The observed values. Sensor layout information includes the positions of each sensor across the entire field of space. Internal position coordinates The historical full-field state dataset includes multiple historical local state datasets. Each historical local state dataset corresponds to a sensor-associated subdomain, containing the state values of all spatial nodes within that subdomain at multiple historical moments. The following sections will describe in detail the method for dividing the sensor-associated subdomains and the specific methods for extracting the local state datasets of each subdomain from the historical full-field state dataset.
[0028] In an optional implementation, the historical full-field state dataset may also include historical data on system control variables, such as heating power, feed rate, and valve opening. Correspondingly, the online observation dataset can also synchronously collect the system control variables at the current moment. This control variable information can serve as auxiliary input to the global reconstruction model, improving the accuracy of model predictions.
[0029] Step S120: Construct a global reconstruction model based on the historical full-field state dataset.
[0030] In this embodiment, the global reconstruction model is used to establish a mapping relationship between sensor observation data and the global state distribution. Its construction process includes: global spatial basis function extraction, global time coefficient calculation, and mapping model training.
[0031] In one embodiment, the historical full-field state dataset includes: historical observation datasets at different historical moments. Specifically, for the i-th historical moment, from the full-field state vector... Extract The observations from each sensor constitute the historical observation dataset for that historical moment. .
[0032] The construction of the global reconstruction model based on the historical full-field state dataset includes the following steps: First, the historical full-field state dataset is subjected to Proper Orthogonal Decomposition (POD) to obtain the global space basis function matrix. Specifically, the historical full-field state dataset is processed through Singular Value Decomposition (SVD). It can be decomposed into the product of the following three matrices: ,in, This represents the number of nodes in the entire discrete space. This indicates the number of historical full-time snapshots. Represents the global space basis function matrix. Represents a singular value matrix. This represents a right singular matrix.
[0033] Next, the global dominant space basis function matrix is extracted from the global space basis function matrix. From the global space basis function matrix, the first... The global basis functions corresponding to the singular values constitute the global dominant basis function matrix. .
[0034] Next, based on the global dominant space basis function matrix and the historical full-field state dataset, the global time coefficients corresponding to different historical moments are determined. For the full-field state vector at the i-th historical moment... Its corresponding global time coefficient By projecting the entire field state vector onto the basis function matrix of the global dominant space, we obtain: Then, the historical observation datasets at different historical moments are used as input data, and the global time coefficients at the corresponding historical moments are used as output data to train a mapping model. The trained mapping model is then determined as the time coefficient prediction model. Specifically, the historical observation datasets at each historical moment are used as input data. As input data, the global time coefficient at the corresponding moment As output data, the weights and biases from the input layer to the hidden layer are randomly initialized, and the weights from the hidden layer to the output layer are calculated analytically. The mapping model is trained with the goal of minimizing the output error, resulting in a time coefficient prediction model. Preferably, the preset learning machine includes an extreme learning machine, and the input data of the mapping model may further include system control variables at corresponding historical moments. That is, the input data is The output data is This is to further improve the prediction accuracy of the time coefficient prediction model.
[0035] Finally, based on the global dominant spatial basis function matrix and the time coefficient prediction model, the global reconstruction model is determined. Specifically, the global reconstruction model comprises two components: the first component is the global dominant spatial basis function matrix determined in the offline phase. The first part is used to reconstruct the global state distribution from the global time coefficient; the second part is the time coefficient prediction model, used in the online phase to predict the global time coefficient at the current moment from the sensor observation data at the current moment. In the subsequent online reconstruction process, the workflow of the global reconstruction model is as follows: it uses the sensor observation dataset at the current moment... Input time coefficient prediction model Output the predicted global time coefficient value at the current moment. Then, based on the global time coefficient, the predicted value is... With the global dominant space basis function matrix By performing a linear combination, the global reconstructed field is obtained. .
[0036] Step S130: Determine the global reconstruction field based on the online observation dataset and the global reconstruction model.
[0037] In this embodiment, the global reconstruction field is generated based on the online observation dataset and the global reconstruction model constructed in step S120. The global reconstruction field is a preliminary estimate of the state of the entire physical field at the current moment, reflecting the overall evolution trend of the physical field and serving as the basis for subsequent local corrections.
[0038] Step S140: After dividing the full-field space region where the distributed parameter system is located into multiple sensor-related subdomains based on the sensor layout information, a local reconstruction model corresponding to each sensor-related subdomain is constructed based on each historical local state dataset.
[0039] Based on the sensor layout information, the entire field space area where the distributed parameter system is located is... It is divided into multiple sensor-associated subdomains. Each sensor-associated subdomain corresponds to a sensor, and the spatial nodes within the subdomain belong to that sensor in space.
[0040] In this embodiment, the nearest neighbor allocation method (i.e., Voronoi partitioning) is used for subdomain partitioning. Specifically, for the ... For each sensor, the corresponding sensor-associated subdomain is defined as: in, For the entire field space domain Any spatial node within, For the first The position coordinates of each sensor This represents the Euclidean distance. The physical meaning of this partitioning method is: to divide each spatial node in the entire field space region into the subdomain corresponding to the nearest sensor, thereby obtaining... Sensor-related subdomains And satisfy That is, the union of all sensor-associated subdomains covers the entire spatial region.
[0041] For each sensor-associated subdomain, from the historical full-field state dataset Extracting historical local state data within this subdomain yields historical local state datasets corresponding to each sensor-related subdomain. Specifically, for the first... Sensor-related subdomains Define the selection operator This selection operator is used to extract the sensor-associated subdomain from the overall state vector. The spatial node state values within, where, For sensor-associated subdomains The number of spatial nodes within the space satisfies Then, at the i-th historical moment, the sensor-associated subdomain The historical local state dataset is as follows: .
[0042] In one embodiment, each of the historical local state datasets corresponds to a different sensor association subdomain. The step of constructing a local reconstruction model corresponding to each sensor association subdomain based on each of the historical local state datasets includes the following steps: First, intrinsic orthogonal decomposition is performed on the historical local state dataset corresponding to each sensor-associated subdomain to obtain the local space basis function matrix corresponding to each sensor-associated subdomain. Taking the sensor-associated subdomain at the i-th historical moment... Corresponding historical local state dataset For example, perform singular value decomposition on it: ,in, For the first The local space basis function matrix of each sensor-associated subdomain, where each column is a local space basis function, describes a typical local variation pattern within the subdomain; This is the corresponding singular value matrix, whose diagonal elements are singular values arranged in descending order; This is the corresponding right singular matrix.
[0043] Next, the local dominant spatial basis function matrix corresponding to each sensor-related subdomain is extracted from each of the local spatial basis function matrices. Specifically, from each of the local spatial basis function matrices... Extracting from the middle separately The local dominant spatial basis functions corresponding to each singular value are used to obtain the local dominant spatial basis function matrix corresponding to each sensor-related subdomain.
[0044] Finally, based on the respective local dominant spatial basis function matrices, the local reconstruction models corresponding to each sensor-related subdomain are determined. Specifically, in addition to the local dominant spatial basis function matrices extracted in the offline stage, the local reconstruction models also include local correction coefficients solved in the online stage, used to determine the combined weights of each local basis function at the current time. The process of solving the local correction coefficients will be described below.
[0045] Step S150: Based on the online observation dataset and the global reconstruction field, determine the local triggering index for each of the sensor-associated subdomains.
[0046] The local trigger index is used to quantify the degree of deviation between the global reconstructed field and the actual sensor observations within each sensor-related subdomain, and is the basis for determining whether the local reconstruction model needs to be activated.
[0047] In one embodiment, the online observation dataset includes: online observation values corresponding to each of the sensor-associated subdomains; the step of determining the local triggering index for each of the sensor-associated subdomains based on the online observation dataset and the global reconstruction field includes the following steps: First, the predicted values corresponding to each sensor-related subdomain are extracted from the global reconstruction field. Because the global reconstruction field... This includes the reconstructed values of all M spatial nodes in the entire field, from which the values of the K sensor locations are extracted. The corresponding predicted values yield a predicted value vector: in, Indicates the global reconstruction field at the th Sensor locations The predicted value at that location.
[0048] Next, for any of the sensor-associated subdomains, the difference between the corresponding online observation value and the predicted value is determined as the observation residual value of the corresponding sensor-associated subdomain.
[0049] Specifically, for the first A sensor, whose observation residual at time t Defined as: in, For the first The actual observations of each sensor For the global reconstruction field in the first Predicted values at each sensor location. Observation residuals. It reflects the degree of deviation of the global reconstructed field at the sensor location—the larger the absolute value of the residual, the less accurate the global reconstructed field is at that location, and the more local correction is needed.
[0050] Finally, based on the observation residual values at multiple consecutive time points, the local triggering index for each sensor-associated subdomain is determined. Preferably, since the observation residual at a single time point may be affected by sensor noise or transient disturbances, a time window smoothing strategy can be used to aggregate the observation residuals at multiple consecutive time points to improve the robustness of the local triggering index.
[0051] It should be noted that in step S140, each sensor-associated subdomain is divided based on a fixed sensor (i.e., each subdomain contains one sensor). Therefore, the observation residual of each subdomain is directly determined by the observation residual of the unique sensor within that subdomain. However, in some embodiments, if other division methods result in a sensor-associated subdomain containing multiple sensors, the observation residuals of each sensor location within that subdomain can be aggregated, for example, by taking the root mean square or a weighted average, to obtain the observation residual of that subdomain.
[0052] Step S160: Based on each of the local triggering indicators and the sensor layout information, determine the collaborative triggering indicators of each sensor-associated subdomain.
[0053] In this embodiment, the collaborative triggering index incorporates spatial correlation information of adjacent subdomains in addition to the local triggering index. Since the physical field is spatially continuous, local anomalies typically affect adjacent regions. Therefore, only when both the local triggering index and the local triggering index of adjacent subdomains are high is it determined that a deviation requiring correction exists in that region. This collaborative mechanism effectively avoids false triggering caused by noise from a single sensor. The collaborative triggering index comprehensively reflects the degree of deviation in the sensor-associated subdomains and the deviation status of their adjacent subdomains, thereby reducing the sensitivity of triggering based solely on the instantaneous deviation of a single sensor.
[0054] In one embodiment, determining the collaborative triggering index for each sensor-associated subdomain based on each of the local triggering indices and the sensor adjacency graph includes the following steps: First, a sensor adjacency structure graph is constructed based on the sensor layout information. This graph includes the spatial adjacency relationships between the associated subdomains of each sensor. The sensor adjacency structure graph characterizes these relationships. Using each associated subdomain as a node, edges are established between spatially adjacent subdomains. If two associated subdomains share a boundary (i.e., there is a common spatial node), they are considered spatially adjacent, and an adjacency edge is established. All nodes and edges constitute the topology of the sensor adjacency structure graph. Edge weights are assigned to each edge to quantify the influence strength between adjacent subdomains. In this embodiment, the edge weights are determined based on the spatial distance between two adjacent sensors. Edge weights are inversely correlated with spatial distance—the closer the distance, the larger the edge weight, indicating a stronger spatial association between the two subdomains; the farther the distance, the smaller the edge weight, indicating a weaker spatial association between the two subdomains. In this embodiment, edge weights are constructed using a distance attenuation method. in, For adjacency indicators, and The first The sensor and the first The position coordinates of each sensor This is a spatial influence scale parameter used to control how quickly the influence intensity decays with distance. The sensor adjacency graph contains three elements: nodes, edges, and edge weights, denoted as... ,in For a set of nodes, Let be the set of edges. This is the weight matrix.
[0055] Next, based on the sensor adjacency structure graph, the adjacent associated subdomains of each sensor associated subdomain are determined. Specifically, based on the sensor adjacency structure graph... Determine the combination of adjacent subdomains of each sensor-associated subdomain. ,in Indicates the relationship with the first A sensor-associated subdomain is the set of all spatially adjacent subdomains.
[0056] Based on this, for any of the sensor-associated subdomains, the local triggering index of the sensor-associated subdomain is weighted and aggregated with the local triggering indices of its adjacent associated subdomains to obtain the collaborative triggering index of the sensor-associated subdomain. (The text then repeats the first instance, which is likely an error in the original.) Taking a sensor-associated subdomain as an example, its collaborative triggering indicator is: ,in For the first Local trigger indicators for each sensor-associated subdomain. For the first The set of adjacent subdomains of each sensor-associated subdomain. The adjacent edge weights defined in step S151, For the first Local triggering indicators for each adjacent subdomain, This is the neighborhood collaboration coefficient, used to adjust the contribution weight of neighborhood information to the collaboration triggering index.
[0057] Step S170: Based on each of the collaborative triggering indicators, determine the activation intensity of each of the local reconstruction models respectively. According to each activation intensity, call the target local reconstruction model from the multiple local reconstruction models to generate a local correction field. Fuse the local correction field and the global reconstruction field to obtain a multi-scale joint reconstruction field. The target local reconstruction model is the local reconstruction model whose activation intensity is greater than a preset intensity threshold.
[0058] In this embodiment, activation intensity is used to control the participation of the local reconstruction model of each sensor-associated subdomain in the final fusion.
[0059] In one embodiment, determining the activation intensity of each local reconstruction model based on each of the collaborative triggering indicators includes the following steps: First, time-memory processing is performed on each of the aforementioned collaborative triggering indices to obtain the memory triggering indices for each of the aforementioned sensor-associated subdomains. In this embodiment, time-memory smoothing processing is used to make the changes in the triggering indices have inertia, avoiding drastic jumps in activation intensity caused by instantaneous fluctuations in the collaborative triggering indices, thereby ensuring the continuity and smoothness of the reconstruction results in time. Specifically, for the first... Cooperative triggering indicators of sensor-related subdomains Perform time-memory smoothing to obtain the co-triggered index for this subdomain. : in, This is a memory trigger indicator from the previous moment. The time interval between two adjacent reconstruction moments. This is a preset memory coefficient used to control the degree to which historical memories affect the current moment.
[0060] Based on this, for each sensor-associated subdomain, the activation intensity of the local reconstruction model corresponding to each sensor-associated subdomain is determined according to the corresponding memory trigger index, a preset start activation threshold, and a saturation activation threshold. In this embodiment, the memory trigger index is mapped to a continuous activation intensity value between 0 and 1 to control the intensity of each local reconstruction model's participation in fusion. Specifically, the memory trigger index of each sensor-associated subdomain is... Each is compared with the preset start activation threshold. and saturation activation threshold By comparison, the activation intensity of the local reconstruction model corresponding to each sensor-associated subdomain is determined: in, Indicates the first The strength of each local reconstruction model participating in the joint reconstruction at the current moment. This is a truncation function.
[0061] In one embodiment, the step of generating a local correction field by calling a target local reconstruction model from among the plurality of local reconstruction models according to each of the activation intensities includes the following steps: First, using the local correction coefficients of the target local reconstruction model as optimization variables, a target optimization function is constructed that includes observation consistency constraints, spatial consistency constraints, and temporal continuity constraints. Specifically, this is based on the activation intensity of each sensor-related subdomain determined in step S155. The target local reconstruction model that needs to participate in the reconstruction is selected from multiple local reconstruction models.
[0062] Next, the objective optimization function is solved to obtain the objective local correction coefficients of the objective local reconstruction model. In this embodiment, the activation intensity... As determined in step S155, the activation intensity, as a known parameter, participates in the construction of the optimization function. The activation intensity determines the weighting of the local basis functions in the optimization objective—the greater the activation intensity of a subdomain, the greater the weight of its local basis functions in the optimization objective. Specifically, for the first... Let there be a sensor-associated subdomain, and define its local correction coefficient vector to be solved at the current time as follows: ,in The locally dominant space function is reserved for this subdomain. The local correction coefficient vectors of all subdomains are concatenated to form the global correction coefficient vector. In the activation intensity Given the established conditions, the global correction coefficient vector is used. To optimize the variables, the following objective function is constructed: ,in, Among them, among them, This represents the local correction coefficient vector to be solved at the current moment; These are the optimal local correction coefficients obtained at the current moment. Optimize the objective function for local correction coefficients; , and These are the weighting coefficients for the observation consistency term, spatial consistency term, and temporal continuity term, respectively. The observation operator is used to map the entire field state to the sensor position; This is the local space basis function matrix after being weighted by the activation intensity of the local model; This represents the observation residual between the current sparse sensor observations and the predicted values of the global reconstructed field at the sensor locations; It is a spatial boundary difference operator used to measure the difference in field values, gradients, or fluxes at the boundaries of associated subdomains of adjacent sensors; Reconstruct the field globally; The time interval between two adjacent reconstruction moments; The optimal local correction coefficients were obtained at the previous reconstruction time.
[0063] Finally, the local correction field is generated based on the target local reconstruction model and the target local correction coefficients. This is based on the local dominant space basis function matrix of each target local reconstruction model. and the corresponding optimal local correction coefficient Generate the local correction field corresponding to the local reconstruction model of each target: Based on the activation intensity, the local correction fields of each target are fused with the global reconstruction field to obtain a multi-scale joint reconstruction field.
[0064] The sparse sensing information-guided distributed parameter system state reconstruction method provided in this application constructs a global reconstruction model and local reconstruction models corresponding to multiple sensor-related subdomains. On the basis of ensuring the accuracy of the overall evolution trend of the physical field in the global model, the local reconstruction models are used to specifically compensate for the deviations and detailed changes in each subdomain, effectively solving the inherent defect that a single global model cannot take into account both the overall trend and local non-uniform features. Based on online observation data and global reconstruction results, local triggering indices for each sensor-related subdomain are determined. Furthermore, by incorporating sensor layout information, spatial correlation information of adjacent subdomains is introduced into the construction of collaborative triggering indices. This allows activation decisions for each local reconstruction model to be collaboratively judged using the physical coupling relationship between adjacent regions, significantly improving the accuracy and robustness of activation decisions. The activation intensity of each local reconstruction model is determined based on collaborative triggering indices, enabling on-demand differentiated activation of local corrections—calling the corresponding local reconstruction model with appropriate intensity only in the region requiring correction, avoiding unnecessary computational overhead and improving the computational efficiency of online reconstruction. By identifying local reconstruction models with activation intensities greater than a preset activation intensity as target local reconstruction models and generating a local correction field, unnecessary online calls to local reconstruction models are reduced, enhancing the reliability of the reconstruction process. Finally, the local correction field is fused with the global reconstruction field, ensuring that the reconstruction results maintain both the macroscopic evolution characteristics of the global scale and incorporate refined information at the local scale. Simultaneously, the collaborative triggering mechanism guarantees the spatial coordination of corrections between adjacent subdomains, significantly improving the accuracy of full-field state reconstruction under sparse observation conditions.
[0065] Example 2 In addition, please see Figure 2 This application provides a sparse sensing information-guided distributed parameter system state reconstruction device 200, the device comprising: Data acquisition module 201 is used to acquire historical full-field state dataset, online observation dataset and sensor layout information of the distributed parameter system, wherein the historical full-field state dataset includes multiple historical local state datasets; The global model determination module 202 is used to construct a global reconstruction model based on the historical full-field state dataset; The global reconstruction module 203 is used to determine the global reconstruction field based on the online observation dataset and the global reconstruction model; The local model determination module 204 is used to divide the full-field space region where the distributed parameter system is located into multiple sensor-related subdomains based on the sensor layout information, and then construct a local reconstruction model corresponding to each sensor-related subdomain based on each historical local state dataset. The local indicator determination module 205 is used to determine the local trigger indicators of each sensor-related subdomain based on the online observation dataset and the global reconstruction field. The collaborative indicator determination module 206 is used to determine the collaborative triggering indicator of each sensor associated subdomain based on each of the local triggering indicators and the sensor layout information. The fusion module 207 is used to determine the activation intensity of each of the local reconstruction models based on the collaborative triggering indicators, and to call the target local reconstruction model from the multiple local reconstruction models according to the activation intensity to generate a local correction field. The local correction field and the global reconstruction field are fused to obtain a multi-scale joint reconstruction field. The target local reconstruction model is the local reconstruction model whose activation intensity is greater than a preset intensity threshold.
[0066] The sparse sensing information-guided distributed parameter system state reconstruction device 200 provided in this embodiment can execute the steps of the sparse sensing information-guided distributed parameter system state reconstruction method provided in Embodiment 1. To avoid repetition, it will not be described again.
[0067] Example 3 In addition, see Figure 3 This embodiment provides a computer device 300, including a memory 303 and a processor 302. The memory 303 stores a computer program, and when the computer program is executed by the processor 302, it implements the state reconstruction method of the distributed parameter system described in Embodiment 1.
[0068] Specifically, the computer device 300 includes a transceiver 301, a bus interface, and a processor 302. The processor 302 is used to: acquire historical full-field state datasets, online observation datasets, and sensor layout information of the distributed parameter system; the historical full-field state datasets include multiple historical local state datasets; construct a global reconstruction model based on the historical full-field state datasets; determine a global reconstruction field based on the online observation datasets and the global reconstruction model; and divide the full-field spatial region where the distributed parameter system is located into multiple sensor-related subdomains based on the sensor layout information, and construct each sensor-related subdomain based on each of the historical local state datasets. The system defines local reconstruction models corresponding to each sensor-associated subdomain. Based on the online observation dataset and the global reconstruction field, it determines local triggering indicators for each sensor-associated subdomain. Based on the local triggering indicators and the sensor layout information, it determines collaborative triggering indicators for each sensor-associated subdomain. Based on the collaborative triggering indicators, it determines the activation intensity of each local reconstruction model. According to the activation intensity, it calls a target local reconstruction model from multiple local reconstruction models to generate a local correction field. It then fuses the local correction field and the global reconstruction field to obtain a multi-scale joint reconstruction field. The target local reconstruction model is a local reconstruction model whose activation intensity is greater than a preset intensity threshold.
[0069] In this embodiment of the application, the computer device 300 further includes a memory 303. Figure 3 In this context, the bus architecture can include any number of interconnected buses and bridges, specifically connecting various circuits of one or more processors represented by processor 302 and memory represented by memory 303. The bus architecture can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. Transceiver 301 can be multiple elements, including transmitters and receivers, providing a unit for communicating with various other devices over a transmission medium. Processor 302 is responsible for managing the bus architecture and general processing, and memory 303 can store data used by processor 302 during operation.
[0070] The computer device 300 provided in this embodiment can execute the steps of the state reconstruction method for the distributed parameter system provided in Embodiment 1. To avoid repetition, the steps will not be repeated.
[0071] Example 4 This disclosure provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the state reconstruction method for the distributed parameter system described in Embodiment 1.
[0072] In this embodiment, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.
[0073] The computer-readable storage medium provided in this embodiment can implement the state reconstruction method of the distributed parameter system provided in Embodiment 1. To avoid repetition, it will not be described again here.
[0074] In all examples shown and described herein, any specific values should be interpreted as merely exemplary and not as limitations; therefore, other examples of exemplary embodiments may have different values.
[0075] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.
[0076] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.
Claims
1. A sparse sensing information guided distributed parameter system state reconstruction method, characterized by, The method includes: Acquire the historical full-field state dataset, online observation dataset, and sensor layout information of the distributed parameter system. The historical full-field state dataset includes multiple historical local state datasets. A global reconstruction model is constructed based on the aforementioned historical full-field state dataset; Based on the online observation dataset and the global reconstruction model, the global reconstruction field is determined; After dividing the full-field space region where the distributed parameter system is located into multiple sensor-related subdomains based on the sensor layout information, a local reconstruction model corresponding to each sensor-related subdomain is constructed based on each historical local state dataset. Based on the online observation dataset and the global reconstruction field, the local triggering indicators for each sensor-related subdomain are determined; Based on the local triggering indicators and the sensor layout information, the collaborative triggering indicators of the associated subdomains of each sensor are determined; Based on each of the aforementioned collaborative triggering indicators, the activation intensity of each of the aforementioned local reconstruction models is determined. According to each of the aforementioned activation intensities, a target local reconstruction model is called from multiple of the aforementioned local reconstruction models to generate a local correction field. The local correction field and the global reconstruction field are fused to obtain a multi-scale joint reconstruction field. The target local reconstruction model is a local reconstruction model whose activation intensity is greater than a preset intensity threshold. The determination of the collaborative triggering indicators for each sensor-related subdomain based on each local triggering indicator and the sensor layout information includes: A sensor adjacency structure graph is constructed based on the sensor layout information. The sensor adjacency graph includes the spatial adjacency relationships between the associated subdomains of each sensor. Based on the sensor adjacency structure graph, the adjacent associated subdomains of each sensor associated subdomain are determined; For any of the sensor-associated subdomains, the local triggering index of the sensor-associated subdomain is weighted and aggregated with the local triggering index of its adjacent associated subdomains to obtain the collaborative triggering index of the sensor-associated subdomain. The determination of the activation intensity of each local reconstruction model based on each of the aforementioned collaborative triggering indicators includes: The time memory processing is performed on each of the aforementioned collaborative triggering indicators to obtain the memory triggering indicators of each of the aforementioned sensor-related subdomains; For each of the sensor-associated subdomains, the activation intensity of the local reconstruction model corresponding to each sensor-associated subdomain is determined based on the corresponding memory trigger index, the preset start activation threshold, and the saturation activation threshold.
2. The sparse sensing information guided distributed parameter system state reconstruction method of claim 1, wherein, The historical full-field state dataset includes: historical observation datasets at different historical moments. The construction of a global reconstruction model based on the historical full-field state dataset includes: The historical full-field state dataset is subjected to eigenorthogonal decomposition to obtain the global space basis function matrix; Extract the global dominant space basis function matrix from the global space basis function matrix; Based on the global dominant space basis function matrix and the historical full-field state dataset, determine the global time coefficients corresponding to different historical moments; The historical observation datasets at different historical moments are used as input data, and the global time coefficients at the corresponding historical moments are used as output data to train the mapping model. The trained mapping model is then determined as the time coefficient prediction model. The global reconstruction model is determined based on the global dominant spatial basis function matrix and the time coefficient prediction model.
3. The sparse sensing information guided distributed parameter system state reconstruction method of claim 1, wherein, Each of the aforementioned historical local state datasets corresponds to a different sensor association subdomain. The step of constructing a local reconstruction model corresponding to each of the aforementioned sensor association subdomains based on each of the aforementioned historical local state datasets includes: The historical local state datasets corresponding to each of the sensor-associated subdomains are subjected to intrinsic orthogonal decomposition to obtain the local space basis function matrix corresponding to each of the sensor-associated subdomains. Extract the local dominant spatial basis function matrix corresponding to each sensor-related subdomain from each of the local spatial basis function matrices; Based on the local dominant space basis function matrices, the local reconstruction models corresponding to the sensor-related subdomains are determined respectively.
4. The method for reconstructing the state of a distributed parameter system guided by sparse sensing information according to claim 1, characterized in that, The online observation dataset includes: online observation values corresponding to each of the sensor-related subdomains; the determination of local triggering indicators for each of the sensor-related subdomains based on the online observation dataset and the global reconstruction field includes: Extract the predicted values corresponding to each sensor-related subdomain from the global reconstruction field; For any of the sensor-associated subdomains, the difference between the corresponding online observation value and the predicted value is determined as the observation residual value of the corresponding sensor-associated subdomain. Based on the observation residual values at multiple consecutive time points, the local trigger index of each sensor-associated subdomain is determined.
5. The method for reconstructing the state of a distributed parameter system guided by sparse sensing information according to claim 1, characterized in that, The step of generating a local correction field by calling the target local reconstruction model from among the multiple local reconstruction models according to the activation intensity includes: Using the local correction coefficients of the target local reconstruction model as optimization variables, a target optimization function containing observation consistency constraints, spatial consistency constraints, and temporal continuity constraints is constructed. Solving the objective optimization function yields the objective local correction coefficients of the objective local reconstruction model; The local correction field is generated based on the target local reconstruction model and the target local correction coefficients.
6. A state reconstruction device for a distributed parameter system guided by sparse sensing information, characterized in that, The device includes: The data acquisition module is used to acquire the historical full-field state dataset, online observation dataset, and sensor layout information of the distributed parameter system. The historical full-field state dataset includes multiple historical local state datasets. The global model determination module is used to construct a global reconstruction model based on the historical full-field state dataset. The global reconstruction module is used to determine the global reconstruction field based on the online observation dataset and the global reconstruction model; The local model determination module is used to divide the full-field space region where the distributed parameter system is located into multiple sensor-related subdomains based on the sensor layout information, and then construct a local reconstruction model corresponding to each sensor-related subdomain based on each historical local state dataset. The local indicator determination module is used to determine the local trigger indicators of each sensor-related subdomain based on the online observation dataset and the global reconstruction field. The collaborative indicator determination module is used to determine the collaborative triggering indicator of each sensor-related subdomain based on each of the local triggering indicators and the sensor layout information; The fusion module is used to determine the activation intensity of each of the local reconstruction models based on the collaborative triggering indicators, call the target local reconstruction model from multiple local reconstruction models according to the activation intensity to generate a local correction field, and fuse the local correction field and the global reconstruction field to obtain a multi-scale joint reconstruction field. The target local reconstruction model is the local reconstruction model whose activation intensity is greater than a preset intensity threshold. The determination of the collaborative triggering indicators for each sensor-related subdomain based on each local triggering indicator and the sensor layout information includes: A sensor adjacency structure graph is constructed based on the sensor layout information. The sensor adjacency graph includes the spatial adjacency relationships between the associated subdomains of each sensor. Based on the sensor adjacency structure graph, the adjacent associated subdomains of each sensor associated subdomain are determined; For any of the sensor-associated subdomains, the local triggering index of the sensor-associated subdomain is weighted and aggregated with the local triggering index of its adjacent associated subdomains to obtain the collaborative triggering index of the sensor-associated subdomain. The determination of the activation intensity of each local reconstruction model based on each of the aforementioned collaborative triggering indicators includes: The time memory processing is performed on each of the aforementioned collaborative triggering indicators to obtain the memory triggering indicators of each of the aforementioned sensor-related subdomains; For each of the sensor-associated subdomains, the activation intensity of the local reconstruction model corresponding to each sensor-associated subdomain is determined based on the corresponding memory trigger index, the preset start activation threshold, and the saturation activation threshold.
7. A computer device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, which, when executed by the processor, implements the sparse sensing information-guided distributed parameter system state reconstruction method as described in any one of claims 1 to 5.
8. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the sparse sensing information-guided distributed parameter system state reconstruction method as described in any one of claims 1 to 5.
Citation Information
Patent Citations
Multi-source data fusion distribution processing analysis method and system
CN120316727A
Multi-source sensing storage environment cooperative monitoring and early warning method
CN120542162A
