Foundation pit safety risk intelligent early warning method based on spatial-temporal feature fusion

Abstract

The invention discloses a foundation pit safety risk intelligent early warning method based on spatial-temporal feature fusion, and the method comprises the steps: enabling a data layer to be connected with monitoring equipment disposed in a foundation pit, and enabling multi-dimensional monitoring data to be connected with a corresponding member in a BIM-GIS digital twin foundation pit model in a hanging manner; time and space alignment is carried out on various monitoring data, a sensor adopts a unified dual-mode time service synchronous clock, and a global standard time axis is established to serve as a data fusion reference; generating same-frequency data by adopting a dynamic sampling method according to the monitoring data with different sampling frequencies; mounting coordinates of the sensors are connected with the BI-GIS digital twin foundation pit model in a hanging mode, and local coordinates are converted into global coordinates through a rotation translation matrix; dividing the foundation pit into space grids, wherein each space grid has corresponding data information; the analysis layer adopts a risk quantification mathematical model and adopts sliding window statistics to dynamically generate an early warning threshold value; and the decision-making layer establishes a multi-level early warning mechanism, and performs early warning when the calculated risk index exceeds an early warning threshold value.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of foundation pit construction safety early warning, in particular to a foundation pit safety risk intelligent early warning method based on space-time feature fusion. BACKGROUND

[0002] With the acceleration of urbanization, land resources are in short supply, and the demand for underground space development is increasing. Deep foundation pit engineering is developing towards large depth, wide area and complex environment. Due to the great technical difficulty of deep foundation pit engineering, the changeable construction environment and the significant impact on the surrounding environment, it has become a high-risk project in the field of building construction, and safety accidents often occur.

[0003] The current foundation pit safety early warning method mainly faces the technical problems of multi-source data fragmentation, hysteresis and high false alarm rate, which are specifically as follows:

[0004] From the data level, the space-time reference of multi-source monitoring data such as displacement, stress and environment is not unified, and the time reference and sampling frequency are not the same, which leads to poor data correlation and delayed early warning.

[0005] From the model level, the existing early warning methods mostly rely on static thresholds (such as displacement > 30mm alarm) or offline finite element analysis, which cannot dynamically respond to changes in working conditions combined with changes in working conditions and environmental changes. If the threshold is not adjusted during heavy rain, the displacement of the supporting pile will exceed the standard but no alarm will be triggered. Machine learning models (such as single LSTM) are sensitive to construction noise and have a high false alarm rate.

[0006] From the decision-making level, the manual judgment process is tedious and takes more than 3 hours on average, and the accuracy of the judgment depends on the experience and responsibility of the technicians. SUMMARY

[0007] In view of the problems of multi-source data fragmentation, hysteresis and high false alarm rate in the existing foundation pit safety early warning method. The purpose of the present application is to provide a foundation pit safety risk intelligent early warning method based on space-time feature fusion.

[0008] The technical solution adopted by the present application to solve its technical problems is: a foundation pit safety risk intelligent early warning method based on space-time feature fusion, including data layer, analysis layer and decision-making layer three levels, the steps are as follows:

[0009] S1: the data layer is connected with the monitoring equipment arranged in the foundation pit, and the obtained multi-dimensional monitoring data is hung with the corresponding components in the BIM-GIS digital twin foundation pit model; the time and space of various types of monitoring data are aligned, including the following steps:

[0010] S101: the sensor adopts a unified dual-mode time synchronization clock to establish a global standard time axis as the data fusion reference;

[0011] S102: Dynamic sampling method is used to generate same frequency data for monitoring data of different sampling frequencies;

[0012] S103: Record the installation coordinates of the sensor on site, link the installation coordinates with the BIM-GIS digital twin foundation pit model, and convert the local coordinates to global coordinates through rotation and translation matrix;

[0013] S104: Divide the foundation pit into spatial grids, each spatial grid has at least corresponding data information of position displacement, steel stress, pore water pressure, and crack risk index;

[0014] S2: The analysis layer uses a risk quantification mathematical model to establish a risk index model as follows:

[0015] R = alpha * f(x) + beta * g(y) + gamma * h(z) + lambda * [1-e (-k-1) ]

[0016] Wherein, R is the risk index;

[0017] x is the displacement value;

[0018] y is the stress value;

[0019] z is the groundwater depth;

[0020] f(x) is the displacement mutation index;

[0021] g(y) is the imbalance degree;

[0022] h(z) is the environmental risk factor;

[0023] k is the correction coefficient;

[0024] l is the crack risk coefficient, which is the convolutional neural network identifying crack length x curvature;

[0025] alpha, beta, gamma, and lambda are weight coefficients, which are dynamically optimized by particle swarm algorithm;

[0026] S3: The analysis layer uses a sliding window to dynamically generate an early warning threshold A;

[0027] S4: The decision layer establishes a multi-level early warning mechanism, and when the risk index calculated in step S2 exceeds the range of the early warning threshold A calculated in step S3, a foundation pit safety risk warning is performed.

[0028] The present invention provides an intelligent early warning method for foundation pit safety risks based on spatiotemporal feature fusion. First, it constructs a multi-source data fusion mechanism for spatiotemporal features such as stress, displacement, environmental impact, and soil cracking, utilizing a dynamic sampling method to achieve synchronous data from multiple sources. Then, it establishes a risk quantification mathematical model considering the coupling of multi-source data, dynamically adjusting different parameter factors and thresholds in conjunction with environmental changes, thereby achieving dynamic analysis of multi-source monitoring data and early warning. This method has at least the following beneficial effects:

[0029] 1. Multi-source heterogeneous data can be synchronized and aligned in seconds, such as the fusion of μs-level stress and daily frequency settlement data;

[0030] 2. Real-time quantification of environmental coupling risks, such as the impact of rainfall on critical displacement;

[0031] 3. Reduce false alarm rate and shorten system response time. Test results show that the false alarm rate is reduced to no more than 5% and the response time is reduced to no more than 10 minutes.

[0032] 4. Dynamically adapts to changes in environment and geological conditions.

[0033] Furthermore, in step S2, the displacement mutation index f(x) is smoothed based on the Holt-Winters exponent to smooth prediction bias, and a series of displacement observations form a dataset {x}. t}(t=1、2,…T), using the Holt-Winters method to analyze x t Perform a one-step prediction to obtain the predicted value. To calculate the prediction bias;

[0034]

[0035] Where σ is the standard deviation of the prediction bias over a past period;

[0036] The imbalance degree g(y) is analyzed using the entropy of the internal force distribution of the support structure, taking the stress data over a certain period of time as [y]. i (i = 1, 2, 3, ..., n), after normalization, we get pi. Further calculation of stress entropy value When the stress distribution is uniform at all measuring points, the stress entropy value reaches its maximum value H. max =In n, stress loss is defined by the degree of entropy deviation;

[0037] The environmental risk factor h(z) = groundwater level change rate × soil permeability coefficient;

[0038] Groundwater level change rate V w =Δh w / Δt,Δh wΔh represents a change amount of a groundwater level, and Δt represents a time change amount.

[0039] Further, in the step S103, the sensor built-in inclination sensor monitors the sensor posture change, and calculates the displacement offset amount through an integral algorithm; the SLAM visual positioning technology and the IMU inertial data are combined, and the image coordinate error is compensated through a camera distortion correction model.

[0040] Further, in the step S3, a sliding window is used to statistically generate a warning threshold A:

[0041]

[0042] A is a warning threshold;

[0043] μ and σ are a historical 30-day risk index mean or standard deviation;

[0044] is a current rainfall intensity change rate;

[0045] η is a soil quality sensitivity coefficient;

[0046] t is time.

[0047] Further, in the step S4, the decision layer adopts a three-level warning mechanism, as follows:

[0048]

[0049]

[0050] Further, in the step S102, dynamic sampling is performed on different sampling frequency data as follows:

[0051] For high-frequency data, a pre-FIR filter is used, the filter cutoff frequency is set to half of the target frequency, and a piecewise cubic spline interpolation method is used to reconstruct a continuous curve;

[0052] For low-frequency data, a time window moving average is used to expand discrete points into period representative values.

[0053] Further, in the step S104, missing data is intelligently repaired, adjacent spatial grid monitoring data is referenced, and a smoothing constraint is set to avoid sudden changes in the completed values. DETAILED DESCRIPTION

[0054] The application will be further described in detail below in combination with specific embodiments.

[0055] The pit safety risk intelligent warning method based on space-time feature fusion of the application includes three levels of data layer, analysis layer and decision layer, and the specific steps are as follows:

[0056] S1: The data layer is connected with monitoring equipment such as inclinometers, reinforcement meters, water level meters, ground penetrating radars, high-definition cameras and the like arranged in the foundation pit, and the obtained multi-dimensional monitoring data is hung with corresponding components in the BIM-GIS digital twin foundation pit model.

[0057] The time and space alignment model is used to align various types of monitoring data in time and space, including the following steps:

[0058] S101: The sensor uses a unified GPS or Beidou dual-mode time synchronization clock to establish a global standard time axis as a data fusion reference;

[0059] S102: A dynamic sampling method is used to generate same-frequency data for monitoring data with different sampling frequencies;

[0060] S103: Record the installation coordinates of the sensor on site, hang the installation coordinates with the BIM-GIS digital twin foundation pit model, and convert the local coordinates into global coordinates through a rotation and translation matrix;

[0061] S104: A space-time coupling matrix model is constructed, and the foundation pit is divided into spatial grids according to 1m x 1m (length x width), and each spatial grid has corresponding position displacement (X, Y, Z in three directions), reinforcement stress, pore water pressure, crack risk index and other data information.

[0062] S2: The analysis layer uses a risk quantification mathematical model to establish a risk index model as follows:

[0063] R = a f(x) + b g(y) + g h(z) + l [1-e (-k·1) ]

[0064] Wherein, R is the risk index;

[0065] x is the displacement value;

[0066] y is the stress value;

[0067] z is the groundwater depth;

[0068] f(x) is the displacement mutation index;

[0069] g(y) is the imbalance degree;

[0070] h(z) is the environmental risk factor;

[0071] k is the correction coefficient, the specific value is determined according to the sensitivity of the crack risk coefficient to the risk contribution, and the specific value is determined according to the expert scoring method;

[0072] l is the crack risk coefficient, which is the product of the crack length and the curvature identified by the CNN (convolutional neural network).

[0073] a, b, g, l are weight coefficients, which are dynamically optimized by a particle swarm algorithm;

[0074] S3: The analysis layer uses a sliding window to statistically generate a warning threshold A;

[0075] S4: The decision layer establishes a multi-level warning mechanism, and when the risk index calculated in step S2 exceeds the range of the warning threshold A calculated in step S3, a foundation pit safety risk warning is performed.

[0076] The space-time feature fusion-based foundation pit safety risk intelligent warning method of the application first constructs a multi-source data fusion mechanism corresponding to space-time features such as stress, displacement, environmental influence, and soil cracking, and realizes the synchronization of multi-source data at the same frequency by using a dynamic sampling method; a risk quantification mathematical model considering the coupling of multi-source data is established, and different parameter factors and thresholds are dynamically adjusted in combination with environmental changes, so as to realize dynamic analysis of multi-source monitoring data and early warning. The method has at least the following beneficial effects:

[0077] 1. Multi-source heterogeneous data is synchronized and aligned at a second level, such as the fusion of stress at a level of μs and daily settlement data;

[0078] 2. Real-time quantification of environmental coupling risks, such as the influence of rainfall on critical displacement;

[0079] 3. Reducing the false alarm rate and shortening the system response time, with a test result of reducing the false alarm rate to not more than 5% and reducing the response time to not more than 10 minutes;

[0080] 4. Dynamic adaptation to changes in environmental and geological conditions.

[0081] In step S2, the displacement mutation index f(x) is based on the Holt-Winters exponential smoothing prediction bias, and the specific calculation method is as follows: a series of displacement observation values form a data set {x t}(t=1, 2, …T), and the Holt-Winters method is used to make one-step forward prediction on x t to obtain the predicted value to calculate the prediction bias (residual);

[0082] then

[0083] wherein, σ is the standard deviation of the prediction bias in the past period of time;

[0084] The imbalance degree g(y) is analyzed by using the internal force distribution entropy of the supporting structure, and the specific method is as follows: taking stress data in a certain period of time as [y i ](i=1, 2, 3, …n), and obtaining pi after normalization, Further calculate stress entropy value When the stress distribution of all measuring points is uniform, the stress entropy value takes the maximum value H max =In n, the stress loss is defined by the degree of deviation of entropy, the value is between 0-1, the greater the stress distribution is more uneven, that is, local stress concentration, high risk;

[0085] The environmental risk factor h(z) = groundwater level mutation rate x soil permeability coefficient;

[0086] Wherein, the groundwater level mutation rate V w =Δh w / Δt, Δh w represents the change of groundwater level, Δt represents the change of time, the water level changes fast and the soil permeability is strong, the seepage effect on the supporting structure is large, which is easy to cause water and soil pressure mutation or soil strength change.

[0087] In the step S103, the sensor built-in inclination sensor monitors the sensor posture change, and calculates the displacement offset through integral algorithm; combined with SLAM visual positioning technology and IMU inertial data, the image coordinate error is compensated through camera distortion correction model.

[0088] In the step S3, a sliding window is used to statistically generate a warning threshold A:

[0089]

[0090] Wherein, A is the warning threshold;

[0091] μ, σ are the mean or standard deviation of the historical 30-day risk index;

[0092] The current rainfall intensity change rate is

[0093] η is the soil quality sensitivity coefficient, the value is (clay = 0.8, sand = 1.2)

[0094] t is time.

[0095] The embodiment adopts a sliding window to statistically generate a warning threshold, which can adjust the warning threshold according to the dynamic change of data, thereby improving the accuracy and timeliness of the warning.

[0096] In the step S4, the decision layer of the embodiment adopts a three-level warning mechanism, which is as follows:

[0097] Risk level Risk index Response measure Class I R≥0.85 Stop work + evacuation Class II 0.7≤R<0.85 Limited time construction + reinforcement Class III 0.6≤R<0.7 Increased monitoring frequency

[0098] In the step S102, different sampling frequency data is dynamically sampled as follows:

[0099] For high-frequency data (such as 1Hz stress data), a pre-FIR filter is used, and the filter cutoff frequency is set to half of the target frequency, so as to suppress high-frequency noise interference; piecewise cubic spline interpolation is used to reconstruct continuous curves, so as to ensure the continuity of physical quantities such as displacement and acceleration, and ensure the smoothness of the curve without sudden changes.

[0100] For low-frequency data (such as daily settlement data), a time window moving average is used to expand discrete points into period representative values (such as evenly distributing single-day settlement to 24 hours).

[0101] In the step S104, the missing data is intelligently repaired, adjacent spatial grid monitoring data is referenced, and a smoothing constraint is set to avoid sudden changes in the completed values.

[0102] The above description is only a description of the preferred embodiments of the present application, and does not limit the scope of the present application in any way. Any modification or modification made by a person skilled in the art according to the above disclosure is within the scope of the claims.

Claims

1. A method for intelligent early warning of foundation pit safety risks based on spatiotemporal feature fusion, characterized in that, It includes three layers: data layer, analysis layer, and decision layer. The steps are as follows: S1: The data layer is connected to the monitoring equipment installed in the foundation pit, and the acquired multidimensional monitoring data is linked to the corresponding components in the BIM-GIS digital twin foundation pit model. Time and spatial alignment of various monitoring data is performed, including the following steps: S101: The sensor adopts a unified dual-mode timing synchronization clock to establish a global standard time axis as the data fusion benchmark; S102: Dynamic sampling method is used to generate same-frequency data for monitoring data with different sampling frequencies; S103: Record the installation coordinates of the sensor on site, connect the installation coordinates with the BIM-GIS digital twin foundation pit model, and convert the local coordinates into global coordinates through a rotation and translation matrix; S104: Divide the foundation pit into a spatial grid, and each spatial grid has at least corresponding data information on position displacement, steel stress, pore water pressure, and crack risk index; S2: The analysis layer adopts a risk quantification mathematical model to establish a risk index model as follows: R=α·f(x)+β·g(y)+γ·h(z)+λ·[1-e (-k·l) ] Where R is the risk index; x is the displacement value; y represents the stress value; z represents the depth of groundwater; f(x) is the displacement abrupt change exponent; g(y) represents the degree of imbalance; h(z) represents the environmental risk factor; k is the correction factor; l represents the crack hazard factor, which is the crack length × curvature identified by the convolutional neural network. α, β, γ, λ are weighting coefficients, which are dynamically optimized using the particle swarm optimization algorithm; S3: The analysis layer uses a sliding window statistical dynamic generation of the early warning threshold A; S4: The decision-making level establishes a multi-level early warning mechanism. When the risk index calculated in step S2 exceeds the range of the early warning threshold A calculated in step S3, an early warning of foundation pit safety risk is issued.

2. The intelligent early warning method for foundation pit safety risks based on spatiotemporal feature fusion according to claim 1, characterized in that: In step S2, the displacement mutation index f(x) is smoothed based on the Holt-Winters exponent to smooth prediction bias, and a series of displacement observations form a dataset {x}. t }(t=1、2,…T), using the Holt-Winters method to analyze x t Perform a one-step prediction to obtain the predicted value. To calculate the prediction bias; but Where σ is the standard deviation of the prediction bias over a past period; The imbalance degree g(y) is analyzed using the entropy of the internal force distribution of the support structure, taking the stress data over a certain period of time as [y]. i (i = 1, 2, 3, ..., n), after normalization, we get pi. Further calculation of stress entropy value When the stress distribution is uniform at all measuring points, the stress entropy value reaches its maximum value H. max =In n, stress loss is defined by the degree of entropy deviation; The environmental risk factor h(z) = groundwater level change rate × soil permeability coefficient Groundwater level change rate V w =Δh w / Δt,Δh w Δt represents the change in groundwater level, and Δt represents the change over time.

3. The intelligent early warning method for foundation pit safety risks based on spatiotemporal feature fusion according to claim 1, characterized in that: In step S103, the sensor has a built-in tilt sensor to monitor sensor attitude changes and calculates displacement offset through an integral algorithm; combined with SLAM visual positioning technology and IMU inertial data, the image coordinate error is compensated through a camera distortion correction model.

4. The intelligent early warning method for foundation pit safety risks based on spatiotemporal feature fusion according to claim 1, characterized in that, In step S3, a sliding window is used to dynamically generate the early warning threshold A: Where A is the warning threshold; μ and σ are the mean or standard deviation of the historical 30-day risk index; This represents the rate of change in current rainfall intensity. η is the soil sensitivity coefficient; t represents time.

5. The intelligent early warning method for foundation pit safety risks based on spatiotemporal feature fusion according to claim 1, characterized in that: In step S4, the decision-making layer adopts a three-level early warning mechanism, as follows:

6. The intelligent early warning method for foundation pit safety risks based on spatiotemporal feature fusion according to claim 1, characterized in that, In step S102, dynamic sampling is used for data at different sampling frequencies as follows: For high-frequency data, a pre-filter is used, with the filter cutoff frequency set to half of the target frequency, and the continuous curve is reconstructed using piecewise cubic spline interpolation. For low-frequency data, a time window moving average is used to expand the discrete points into representative values ​​for the time period.

7. The intelligent early warning method for foundation pit safety risks based on spatiotemporal feature fusion according to claim 1, characterized in that: In step S104, the missing data is intelligently repaired by referencing monitoring data from adjacent spatial grids and by setting smoothing constraints to avoid abrupt changes in the completed values.

Images