Ground settlement sensor layout optimization method based on physical information neural network
By using a ground settlement sensor layout optimization method based on physical information neural networks, and generating sensor layout optimization schemes through predictive processing and physical constraints, the problem of insufficient sensor configuration under heavy rainfall conditions is solved, and more efficient monitoring results are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-10
- Publication Date
- 2026-07-10
AI Technical Summary
Under heavy rainfall conditions, the effectiveness of monitoring with limited sensor configuration is difficult to improve with existing technologies. The existing deployment methods cannot effectively quantify monitoring information in key areas or time periods, resulting in insufficient monitoring data.
A ground settlement sensor layout optimization method based on physical information neural network is adopted. By acquiring existing sensor data, performing predictive processing and quantitative evaluation, an optimized sensor layout scheme is generated. This introduces physical constraints on ground settlement evolution, reduces the randomness of sensor placement, and improves monitoring effectiveness.
This improved the monitoring effectiveness of sensor configuration under heavy rainfall conditions, reduced the randomness of sensor deployment, and enhanced the stability and accuracy of monitoring data.
Smart Images
Figure CN122366136A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of data processing technology, and in particular to a method for optimizing the layout of foundation settlement sensors based on physical information neural networks. Background Technology
[0002] Under conditions of heavy rainfall, rainfall infiltration, groundwater level fluctuations, and soil consolidation effects can cause foundation settlement to exhibit significant spatiotemporal heterogeneity and phased characteristics. Due to limitations in cost, construction conditions, power supply, communication, and maintenance, engineering sites often only allow for the deployment of a limited number of foundation settlement sensors to obtain discrete monitoring data.
[0003] Current deployment methods mostly employ empirical point selection, geometrically uniform point distribution, or static zonal densification. While simple to implement, these methods struggle to quantify the contribution of different points to monitoring information during critical phases of heavy rainfall, easily leading to insufficient information in key areas or time periods. Methods based on interpolation or statistical models can provide some assessment of errors or uncertainties, but they typically lack constraints from the physical laws governing settlement evolution, resulting in limited extrapolation stability under non-stationary and transient conditions caused by heavy rainfall. Therefore, existing technologies cannot improve the monitoring effectiveness of limited sensor configurations under heavy rainfall conditions. Summary of the Invention
[0004] The main objective of this application is to provide a method for optimizing the layout of foundation settlement sensors based on physical information neural networks. This method aims to solve the technical problem in related technologies where empirical point placement, geometrically uniform point placement, or static partitioning and densification methods fail to improve the monitoring effectiveness of limited sensor configurations under heavy rainfall conditions.
[0005] To achieve the above objectives, embodiments of this application provide a method for optimizing the layout of foundation settlement sensors based on a physical information neural network. The method includes:
[0006] Acquire settlement monitoring data from existing foundation settlement sensors within a preset monitoring time window. The settlement monitoring data includes the spatial location of the area to be monitored, the monitoring time point, and rainfall characteristic data.
[0007] Based on a pre-set neural network model, the settlement monitoring data is processed to predict the foundation settlement. The pre-set neural network model is obtained by joint training based on the monitoring data of existing foundation settlement sensors and the physical constraints of foundation settlement evolution.
[0008] By using a pre-set neural network model, the prediction results of foundation settlement are processed to determine the prediction confidence level and the distribution of physical residuals;
[0009] Based on the prediction confidence level and physical residual distribution, the benefits of adding each candidate point are quantitatively evaluated, and a sensor layout optimization scheme is generated.
[0010] In one possible implementation of this application, after acquiring settlement monitoring data from existing foundation settlement sensors within a preset monitoring time window, the method further includes:
[0011] The settlement monitoring data were time-aligned, outlier removed, and noise filtered to obtain the training dataset.
[0012] Based on the training dataset, the neural network model to be trained is iteratively trained to obtain the pre-set neural network model after training.
[0013] In one possible implementation of this application, based on a training dataset, the neural network model to be trained is iteratively trained to obtain a pre-defined neural network model that has been trained, including:
[0014] Based on the spatiotemporal evolution of foundation settlement, physical constraints are determined, and physical residuals are constructed based on these constraints.
[0015] A joint training objective is constructed based on the fitting error and physical residual of the monitoring data in the training dataset.
[0016] Based on the joint training objective, the neural network model to be trained is iteratively trained to obtain a trained physical information neural network model.
[0017] The physical information neural network model is used as the preset neural network model.
[0018] In one possible implementation of this application, the foundation settlement prediction results are processed using a pre-set neural network model to determine the prediction confidence level and the distribution of physical residuals, including:
[0019] By using a pre-set neural network model and model ensemble method, the prediction results of foundation settlement are processed to determine the prediction confidence level and physical residual distribution of the area to be monitored within the pre-set monitoring time window. The prediction confidence level is used to characterize the confidence level of the pre-set neural network model in predicting foundation settlement, and the physical residual distribution is used to characterize the degree of deviation between the model prediction results and physical laws.
[0020] In one possible implementation of this application, based on the prediction confidence level and physical residual distribution, the added benefits of each candidate point are quantitatively evaluated to generate a sensor layout optimization scheme, including:
[0021] Based on the prediction confidence level and physical residuals, a data acquisition function is constructed;
[0022] The additional revenue index for each candidate position is calculated using the data collection function.
[0023] Based on the added revenue indicators, the locations of new sensors are iteratively selected under the condition of meeting the deployment constraints, and a sensor layout optimization scheme is generated.
[0024] In one possible implementation of this application, the revenue indicator is added to include a prediction confidence term, a physical residual term, and a time weight, so that the revenue weight of the critical stage of heavy rainfall is greater than that of the non-critical stage.
[0025] In one possible implementation of this application, based on the added benefit index, the locations of newly added sensors are iteratively selected under the condition of satisfying the deployment constraints to generate a sensor layout optimization scheme, including:
[0026] Based on the added revenue indicator, in each round, the candidate point with the largest added revenue is selected from the candidate points that meet the deployment constraints as the new position;
[0027] After selecting the new location, the neighboring candidate points of the new location are removed from the candidate point set according to the minimum point spacing constraint, thus generating a sensor layout optimization scheme.
[0028] In one possible implementation of this application, after iteratively selecting the locations of new sensors based on the added revenue index and under the condition of satisfying the deployment constraints, and generating a sensor layout optimization scheme, the method further includes:
[0029] Obtain the newly added monitoring data of the sensor corresponding to the newly added sensor location;
[0030] By adding new monitoring data, the preset neural network model is updated and trained, and the prediction confidence level, physical residual, and added benefit indicators are recalculated in order to select new sensor layout locations in the area to be monitored.
[0031] This application provides a method for optimizing the layout of foundation settlement sensors based on a physical information neural network. In this application, settlement monitoring data from existing foundation settlement sensors within a preset monitoring time window is acquired. Then, based on a preset neural network model, the settlement monitoring data is processed to obtain a foundation settlement prediction result. This prediction result is then further processed using the preset neural network model to determine the prediction confidence level and physical residuals. Finally, based on the prediction confidence level and physical residuals, a sensor layout optimization scheme is generated. The physical constraints of foundation settlement evolution are introduced into the preset neural network model. Based on a limited amount of existing foundation settlement sensor data, the benefits of adding candidate points are quantitatively evaluated. This reduces the blindness of point placement caused by relying solely on experience or pure interpolation methods, improves the stability of layout optimization, and enhances the monitoring effectiveness of limited sensor configurations under heavy rainfall conditions. Attached Figure Description
[0032] Figure 1This is a flowchart illustrating the first embodiment of the foundation settlement sensor layout optimization method based on physical information neural network of this application;
[0033] Figure 2 This is a schematic diagram of the physical information neural network model structure and training involved in the foundation settlement sensor layout optimization method based on physical information neural network in this application;
[0034] Figure 3 This is a schematic diagram illustrating the calculation and selection of sensor addition benefits in the embodiments of this application. Detailed Implementation
[0035] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.
[0036] This application provides a method for optimizing the layout of foundation settlement sensors based on a physical information neural network. In the first embodiment of this method, referring to... Figure 1 The methods include:
[0037] Step S10: Obtain settlement monitoring data from existing foundation settlement sensors within a preset monitoring time window. The settlement monitoring data includes the spatial location of the area to be monitored, the monitoring time point, and rainfall characteristic data.
[0038] It should be noted that the method for optimizing the layout of foundation settlement sensors based on physical information neural networks can be applied to a device for optimizing the layout of foundation settlement sensors based on physical information neural networks. This device belongs to a system for optimizing the layout of foundation settlement sensors based on physical information neural networks, and this system belongs to a device for optimizing the layout of foundation settlement sensors based on physical information neural networks.
[0039] It should be noted that the area to be monitored can be a two-dimensional site plan, with the key areas affected by foundation settlement selected as the monitoring area. Its boundary is denoted as The monitoring plane coordinates are: The monitoring time is The settlement amount is (Unit can be mm), the existing set of foundation settlement sensors is represented as: The set of candidate new points is .
[0040] It should be understood that the preset monitoring time window can be a 1-hour, 4-hour, etc., and there is no specific limitation. The monitoring time window related to heavy rainfall is set as follows: .in, This is the baseline time before the heavy rainfall. This is to indicate that the subsidence has stabilized or reached a point of concern for engineering projects after heavy rainfall.
[0041] It is important to understand that rainfall characteristic data includes rainfall intensity and cumulative rainfall, which are used to determine the workable areas from the monitored area. Considering factors such as minimum point spacing constraints, non-deployable areas are generated. Let reachability indicator function This represents the reachability of power supply, communication, and maintenance. Combining the above constraints, the feasible candidate location set is as follows:
[0042]
[0043] The step S10, which involves acquiring settlement monitoring data from existing foundation settlement sensors within a preset monitoring time window, further includes:
[0044] The settlement monitoring data were time-aligned, outlier removed, and noise filtered to obtain the training dataset.
[0045] It is important to understand that existing sensors Within the preset monitoring time window The original settlement sequence or settlement monitoring data within is denoted as The set of sampling times is .
[0046] The time alignment process can be handled in the following ways:
[0047] Constructing a common time grid Alignment can be achieved using linear interpolation (or splines):
[0048]
[0049] Outlier removal can be achieved by calculating the median for each sensor sequence. MAD (Absolute Deviation from Median):
[0050]
[0051] like
[0052]
[0053] Then this point is considered an anomaly and removed / replaced. (where, 3 to 6 can be selected.
[0054] Noise filtering can be achieved by, for example, using a first-order low-pass filter:
[0055]
[0056] in, These are the filter coefficients.
[0057] The final training dataset is obtained as follows:
[0058]
[0059] Based on the training dataset, the neural network model to be trained is iteratively trained to obtain the pre-set neural network model after training.
[0060] It should be noted that, furthermore, this training dataset is used to iteratively train the neural network model or physical information neural network to be trained. First, the input-output and network structure of the physical information neural network are constructed, incorporating spatial location, time, and rainfall intensity. and cumulative rainfall As input to the neural network, the corresponding vector is The neural network outputs the predicted settlement value. ,in These are network parameters. The network structure can be a fully connected network. Let the network have a total of... Hidden layer, number The layer is mapped as follows:
[0061]
[0062] The output layer is:
[0063]
[0064] Activation function Optional The function facilitates automatic differentiation to obtain a smooth derivative.
[0065] The step of iteratively training the neural network model to be trained based on the training dataset to obtain the pre-trained neural network model includes:
[0066] Based on the spatiotemporal evolution of foundation settlement, physical constraints are determined, and physical residuals are constructed based on these constraints.
[0067] It should be noted that the dominant mechanism of subsidence under the influence of heavy rainfall can be abstracted as diffusion-type spatiotemporal evolution and rainfall source term driving, and its physical constraint equation is:
[0068]
[0069] in, It is the equivalent settlement diffusion coefficient (characterizing the strength of settlement transmission / coupling in a plane). This is the equivalent coupling coefficient between rainfall intensity and settlement rate; For the Laplace operator:
[0070]
[0071] Substitute the network output into the physical equations to construct the physical residuals:
[0072]
[0073] A joint training objective is constructed based on the fitting error and physical residual of the monitoring data in the training dataset.
[0074] It should be noted that the joint training objective includes data fitting loss and physical loss, whereby the data fitting loss can be expressed as:
[0075]
[0076] Physical loss can be expressed as:
[0077]
[0078] in, To include sampling within the monitoring domain and time window A set of physical coordinate points:
[0079]
[0080] Latin hypercube or stratified sampling can be used, with increased density for key stages of heavy rainfall. The final joint training objective can then be obtained.
[0081] =
[0082] in, These are the weighting coefficients. Minimize them using a gradient descent-like optimization algorithm. The trained model is obtained, and the training process is illustrated in the diagram below. Figure 2 As shown.
[0083] Based on the joint training objective, the neural network model to be trained is iteratively trained to obtain a trained physical information neural network model.
[0084] The physical information neural network model is used as the preset neural network model.
[0085] Step S20: Based on the preset neural network model, the settlement monitoring data is processed to predict the foundation settlement. The preset neural network model is obtained by joint training based on the monitoring data of existing foundation settlement sensors and the physical constraints of foundation settlement evolution.
[0086] Step S30: Process the foundation settlement prediction results using a preset neural network model to determine the prediction confidence level and the distribution of physical residuals.
[0087] Step S30 includes:
[0088] By using a pre-set neural network model and model ensemble method, the prediction results of foundation settlement are processed to determine the prediction confidence level and physical residual distribution of the area to be monitored within the pre-set monitoring time window. The prediction confidence level is used to characterize the confidence level of the pre-set neural network model in predicting foundation settlement, and the physical residual distribution is used to characterize the degree of deviation between the model prediction results and physical laws.
[0089] It should be noted that, in order to obtain the spatial-temporal uncertainty distribution, this embodiment adopts... An ensemble is formed by independently trained PINNs to obtain... .
[0090] It should be noted that the confidence level of the prediction can be a measure of the prediction uncertainty, calculated by integrating the mean and variance. Specifically:
[0091]
[0092]
[0093] in, That is, the quantity representing the uncertainty of prediction.
[0094] (2) Calculate the physical residuals for each model to obtain the physical residual distribution:
[0095]
[0096] And take the integrated residual strength:
[0097]
[0098] Step S40: Based on the prediction confidence level and physical residual distribution, the benefits of adding each candidate point are quantitatively evaluated to generate a sensor layout optimization scheme.
[0099] It should be noted that by integrating the added benefit index of prediction confidence level, physical residual distribution and key stage weight, the added benefit of each candidate point is quantitatively evaluated, and a sensor layout optimization scheme is generated.
[0100] Step S40 includes:
[0101] A data acquisition function is constructed based on the prediction confidence level and the physical residual.
[0102] It should be noted that, firstly, let the set of key stages be... The time weighting function is defined as follows:
[0103]
[0104] in To enhance the coefficient, This is an indicator function.
[0105] The acquisition function is used to calculate the additional revenue index of each candidate position corresponding to each candidate point.
[0106] It should be noted that for each candidate position Define its overall benefit within the time window and construct the acquisition function:
[0107]
[0108] in, Control the weights of "uncertainty exploration" and "physical consistency correction". Discretize to a common time grid. back:
[0109]
[0110] The added benefit indicators include a prediction confidence level term, a physical residual term, and a time weight, so that the benefit weight of the critical stage of heavy rainfall is greater than that of the non-critical stage.
[0111] Based on the added revenue indicators, the locations of new sensors are iteratively selected under the condition of meeting the deployment constraints, and a sensor layout optimization scheme is generated.
[0112] It should be noted that the constraint-iterative point selection is related to the closed-loop update of "training-evaluation-point selection-addition" (…). Figure 3 Let the number of new sensors to be added be... The currently selected set is The candidate set is .
[0113] The steps involved in iteratively selecting new sensor locations and generating an optimized sensor layout scheme based on the added revenue indicators and under the condition of satisfying the deployment constraints include:
[0114] Based on the added revenue indicator, in each round, the candidate point with the highest added revenue is selected from the candidate points that meet the deployment constraints as the new position.
[0115] It should be noted that setting constraints can be done by determining the workable area from the area to be monitored. Considering factors such as minimum point spacing constraints, non-deployable areas are generated. Let reachability indicator function This characterizes the accessibility of power supply, communication, and maintenance. Combining the above constraints generates deployment constraints, resulting in the following set of feasible candidate locations:
[0116]
[0117] After selecting the new location, the neighboring candidate points of the new location are removed from the candidate point set according to the minimum point spacing constraint, thus generating a sensor layout optimization scheme.
[0118] It should be noted that point selection occurs in a single iteration. In the... In the next iteration, the profit is calculated for all candidate points. ,choose:
[0119]
[0120] Then update the candidate point set:
[0121]
[0122] The process, which involves iteratively selecting new sensor locations based on added revenue indicators while meeting deployment constraints, and generating an optimized sensor layout scheme, also includes:
[0123] Obtain the newly added monitoring data of the sensor corresponding to the newly added sensor location.
[0124] By adding new monitoring data, the preset neural network model is updated and trained, and the prediction confidence level, physical residual, and added benefit indicators are recalculated in order to select new sensor layout locations in the area to be monitored.
[0125] It should be noted that new data and model updates are involved in the training process. Actual addition of sensors and acquisition of new monitoring data Then, it was merged into the dataset:
[0126]
[0127] and the joint losses mentioned above Continue training; then recalculate. and Then proceed to the next round of "evaluation-site selection".
[0128] Termination condition and output: When the number of iterations reaches... Or the candidate point return is lower than the threshold The process terminates at a certain time. The optimized layout is then output.
[0129]
[0130] This application provides a method for optimizing the layout of foundation settlement sensors based on a physical information neural network. In this application, settlement monitoring data from existing foundation settlement sensors within a preset monitoring time window is acquired. Then, based on a preset neural network model, the settlement monitoring data is processed to obtain a foundation settlement prediction result. This prediction result is then further processed using the preset neural network model to determine the prediction confidence level and physical residuals. Finally, based on the prediction confidence level and physical residuals, a sensor layout optimization scheme is generated. The physical constraints of foundation settlement evolution are introduced into the preset neural network model. Based on a limited amount of existing foundation settlement sensor data, the benefits of adding candidate points are quantitatively evaluated. This reduces the blindness of point placement caused by relying solely on experience or pure interpolation methods, improves the stability of layout optimization, and enhances the monitoring effectiveness of limited sensor configurations under heavy rainfall conditions.
[0131] The specific implementation method of the foundation settlement sensor layout optimization device based on physical information neural network in this application is basically the same as the various embodiments of the foundation settlement sensor layout optimization method based on physical information neural network described above, and will not be repeated here.
[0132] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.
[0133] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0134] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods of the various embodiments of this application.
[0135] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.
Claims
1. A method for optimizing the layout of foundation settlement sensors based on physical information neural networks, characterized in that, The method includes: Acquire settlement monitoring data from existing foundation settlement sensors within a preset monitoring time window. The settlement monitoring data includes the spatial location of the area to be monitored, the monitoring time point, and rainfall characteristic data. Based on a preset neural network model, the settlement monitoring data is processed to predict the foundation settlement, and the preset neural network model is obtained by joint training based on the monitoring data of existing foundation settlement sensors and the physical constraints of foundation settlement evolution. The foundation settlement prediction results are processed using the preset neural network model to determine the prediction confidence level and the distribution of physical residuals. Based on the predicted confidence level and physical residual distribution, the benefits of adding each candidate point are quantitatively evaluated, and a sensor layout optimization scheme is generated.
2. The method for optimizing the layout of foundation settlement sensors based on physical information neural networks as described in claim 1, characterized in that, After acquiring settlement monitoring data from existing foundation settlement sensors within a preset monitoring time window, the process further includes: The settlement monitoring data is time-aligned, outlier removed, and noise filtered to obtain a training dataset. Based on the training dataset, the neural network model to be trained is iteratively trained to obtain a pre-set neural network model that has been trained.
3. The method for optimizing the layout of foundation settlement sensors based on physical information neural networks as described in claim 2, characterized in that, The step of iteratively training the neural network model to be trained based on the training dataset to obtain a pre-set neural network model after training includes: Based on the spatiotemporal evolution of foundation settlement, physical constraints are determined, and physical residuals are constructed based on these physical constraints. Based on the fitting error of the monitoring data in the training dataset and the physical residual, a joint training objective is constructed; Based on the joint training objective, the neural network model to be trained is iteratively trained to obtain a trained physical information neural network model. The physical information neural network model is used as the preset neural network model.
4. The method for optimizing the layout of foundation settlement sensors based on physical information neural networks as described in claim 1, characterized in that, The process of processing the foundation settlement prediction results using the preset neural network model to determine the prediction confidence level and physical residual distribution includes: The foundation settlement prediction results are processed using the preset neural network model and model integration method to determine the prediction confidence level and physical residual distribution of the area to be monitored within the preset monitoring time window. The prediction confidence level is used to characterize the confidence level of the preset neural network model in predicting the foundation settlement, and the physical residual distribution is used to characterize the degree of deviation between the model prediction results and physical laws.
5. The method for optimizing the layout of foundation settlement sensors based on physical information neural networks as described in claim 1, characterized in that, The step of quantitatively evaluating the benefits of adding each candidate point based on the predicted confidence level and physical residual distribution, and generating a sensor layout optimization scheme, includes: Based on the predicted confidence level and physical residual, a data acquisition function is constructed; The additional revenue index for each candidate position is calculated using the data collection function. Based on the added revenue indicators, the locations of the new sensors are iteratively selected under the condition of satisfying the deployment constraints, and a sensor layout optimization scheme is generated.
6. The method for optimizing the layout of foundation settlement sensors based on physical information neural networks as described in claim 5, characterized in that, The added benefit indicators include a prediction confidence level term, a physical residual term, and a time weight, so that the benefit weight of the critical stage of heavy rainfall is greater than that of the non-critical stage.
7. The method for optimizing the layout of foundation settlement sensors based on physical information neural networks as described in claim 5, characterized in that, The step of iteratively selecting new sensor locations based on the added revenue indicators, under the condition of satisfying the deployment constraints, and generating a sensor layout optimization scheme includes: Based on the aforementioned additional revenue indicator, in each round, the candidate point with the highest additional revenue is selected from the candidate points that meet the deployment constraints as the new position; After selecting the new location, neighboring candidate points of the new location are removed from the candidate point set according to the minimum point spacing constraint, thus generating a sensor layout optimization scheme.
8. The method for optimizing the layout of foundation settlement sensors based on physical information neural networks as described in claim 5, characterized in that, After generating a sensor layout optimization scheme by iteratively selecting new sensor locations based on the added revenue index and under the condition of satisfying the layout constraints, the process further includes: Obtain the newly added monitoring data of the sensor corresponding to the newly added sensor location; By adding new monitoring data, the preset neural network model is updated and trained, and the prediction confidence level, physical residual, and added benefit index are recalculated in order to select new sensor layout locations in the area to be monitored.