Sensor dynamic information linkage analysis method based on feature matching and fusion

Abstract

The present invention proposes a sensor dynamic information linkage analysis method based on feature matching fusion, which includes the following steps: obtaining measurement data input by a hybrid distributed sensor, vectorizing the measurement data to obtain a first parameter vector group; performing unbiased estimation judgment on the first parameter vector group, and converging the first parameter vector group according to the estimated probability after judgment to obtain a second parameter vector group; using a time autoregressive model to refer to the second parameter vector group to form a time predictor, and the time predictor outputs a third parameter vector group; inputting the second parameter vector group and the third parameter vector group into a Hadoop platform, and analyzing the second parameter vector group and the third parameter vector group according to a preset neural network model; the method simulates a real environment by fusing multi-information sensor data and using a time autoregressive model to predict subsequent sensor data, so as to realize the prediction of geological disasters.

Description

Technical Field

[0001] The present invention relates to the technical field of geological disaster monitoring, and in particular to a sensor dynamic information linkage analysis method based on feature matching and fusion. Background Art

[0002] The traditional monitoring directions of mountain slopes are mainly divided into surface deformation monitoring, deep deformation monitoring, related physical quantity monitoring and landslide formation activity related monitoring. Through the landslide prediction analysis of the collected data, the prediction can be mainly divided into two aspects: on the one hand, the spatial displacement of the landslide is predicted by regression prediction, mainly using the logistic regression method; on the other hand, the landslide time is inferred by the change of the monitored displacement. The main methods are the creep experience prediction of the slope time by Saito Toshitaka or the time prediction of the Verhulst model. The shortcomings of monitoring can be mainly divided into three aspects: the single and limited variables of slope analysis, the complex relationship between each internal variable and the difficulty of calculation, and the inaccurate time prediction. There are shortcomings such as being restricted by meteorological conditions, unstable monitoring, and poor integration and coordination of monitoring technology;

[0003] Multi-sensor collaboration includes two levels: collaborative data inversion and collaborative monitoring process. The former focuses on data application, which is an indirect post-evaluation mode of sensor performance and is also the current mainstream direction; the latter focuses on the performance of the sensor itself, and selects the combination scheme that best matches the target requirements from the inherent physical parameters according to the specific task requirements, which is also the frontier direction of research. In addition, the ground monitoring environment is complex and changeable, and the monitoring performance of multiple sensors needs to comprehensively consider their inherent technical indicators and task requirements. The existing performance evaluation models mostly rely on expert decisions or experience values, which are prone to subjective interference;

[0004] The Chinese invention patent with publication number CN115440008B discloses a landslide radar wave monitoring method and system. The invention generates a landslide monitoring cycle based on the meteorological analysis results by performing meteorological analysis of rainfall and earthquakes; performs periodic slope deformation monitoring on the landslide monitoring position; generates a landslide warning signal when there is a landslide risk; determines multiple landslide warning positions, performs video monitoring and shooting, and generates monitoring shooting data; identifies multiple landslide positions when there is a landslide, and issues a landslide position alarm; and still has the following defects:

[0005] Only through radar monitoring, the slope analysis variables are single and limited, it is difficult to accurately predict the landslide time, it is impossible to fully integrate the data of multiple sensors, and it is impossible to provide more accurate early warning information;

[0006] Sensor signals are easily interfered with, resulting in false alarms;

[0007] To this end, the present invention proposes a sensor dynamic information linkage analysis method based on feature matching and fusion. Summary of the invention

[0008] In order to solve the above problems, the present invention proposes a sensor dynamic information linkage analysis method based on feature matching and fusion to more accurately solve the above problems.

[0009] The present invention is achieved through the following technical solutions:

[0010] The present invention proposes a sensor dynamic information linkage analysis method based on feature matching and fusion, comprising the following steps:

[0011] S1: Acquire measurement data input by a hybrid distributed sensor, and vectorize the measurement data to obtain a first parameter vector group;

[0012] S2: performing unbiased estimation judgment on the first parameter vector group, and converging the first parameter vector group according to the estimated probability after judgment to obtain a second parameter vector group;

[0013] S3: using a time autoregressive model to refer to the second parameter vector group to form a time predictor, and the time predictor outputs a third parameter vector group;

[0014] S4: Input the second parameter vector group and the third parameter vector group into the Hadoop platform, and analyze the second parameter vector group and the third parameter vector group according to a preset neural network model.

[0015] Furthermore, the sensor dynamic information linkage analysis method based on feature matching and fusion, the step S4 includes:

[0016] S41: Identify the input time of the second parameter vector group and the timestamp of forming the third parameter vector group;

[0017] S42: performing feature projection transformation on the third parameter vector group to obtain a feature vector group corresponding to the hybrid distributed sensor;

[0018] S43: Perform spatial differentiation on the eigenvector group to generate corresponding eigenvariance;

[0019] S44: Substitute the feature variance into the log-likelihood function simulated by the hybrid distributed sensor to calculate the real environment value of the hybrid distributed sensor;

[0020] S45: An environmental score is obtained by analyzing and calculating the actual environmental value, and compared with a preset threshold value. If the environmental score is less than the preset threshold value, there is a risk of a geological disaster.

[0021] Furthermore, in the sensor dynamic information linkage analysis method based on feature matching fusion, the step of obtaining measurement data input by a hybrid distributed sensor and vectorizing the measurement data to obtain a first parameter vector group includes:

[0022] The measurement data includes hybrid distributed sensor information Y, and the hybrid distributed sensor information includes sensing information, functional information and location information;

[0023] Based on the hybrid distributed sensor information Perform vector transformation to obtain the first parameter vector group matching the measured data .

[0024] Furthermore, in the sensor dynamic information linkage analysis method based on feature matching fusion, the step of performing unbiased estimation and judgment on the first parameter vector group, and converging the first parameter vector group according to the estimated probability after judgment to obtain the second parameter vector group includes:

[0025] Perform unbiased estimation on the first parameter vector group, and Node estimation , each of the hybrid distributed sensor information meets the preset sensor threshold , then , determining that the error of the hybrid distributed sensor information is within a preset value;

[0026] The first parameter vector group after unbiased estimation is converged to the true value , get the second parameter vector group , where p is the coefficient of convergence to the true value.

[0027] Furthermore, in the sensor dynamic information linkage analysis method based on feature matching and fusion, the step of using a time autoregressive model to refer to the second parameter vector group to form a time predictor, and the step of the time predictor outputting a third parameter vector group includes:

[0028] Assume that the time series t in the temporal predictor is mixed with only one of the distributed sensor information There is a linear relationship, then:

[0029] ;

[0030] in, They are hybrid distributed sensor information by The weight of the impact, is independent of the hybrid distributed sensor information White noise sequence.

[0031] Furthermore, in the sensor dynamic information linkage analysis method based on feature matching and fusion, the step of performing feature projection transformation on the third parameter vector group to obtain a feature vector group corresponding to the hybrid distributed sensor includes:

[0032] The feature projection algorithm is:

[0033] ;

[0034] In the formula, the neural network convolution layer h=1, 2, ..., m preset in the neural network model; L is the preset value of each hybrid distributed sensor in the neural network convolution layer h, and X is a feature vector group.

[0035] Furthermore, in the sensor dynamic information linkage analysis method based on feature matching and fusion, the step of performing spatial differentiation on the feature vector group and generating corresponding feature variance includes:

[0036] ;

[0037] In the formula, x is one of the vectors in the feature vector group X. Each vector in the feature vector group X is spatially differentiated, and the convolution coefficient corresponding to the convolution layer h is Calculate the characteristic variance .

[0038] Furthermore, in the sensor dynamic information linkage analysis method based on feature matching fusion, the step of substituting the feature variance into the log-likelihood function simulated by the hybrid distributed sensor to calculate the real environment value of the hybrid distributed sensor includes:

[0039] The log-likelihood function is:

[0040] ;

[0041] Real environment value The characteristic variance , convolutional layer coefficients , each vector x in the eigenvector group X is calculated.

[0042] Furthermore, in the sensor dynamic information linkage analysis method based on feature matching and fusion, the step of analyzing and calculating the environmental score based on the real environmental value includes:

[0043] The real environment value Substituting into the scoring algorithm, the scoring algorithm is:

[0044] ;

[0045] Among them, the environmental score is S, is the corresponding variance produced by the simulation.

[0046] Beneficial effects of the present invention:

[0047] The sensor dynamic information linkage analysis method based on feature matching fusion proposed in the present invention proposes unbiased estimation judgment and true value convergence of the first parameter vector group, thereby improving the accuracy of sensor data and reducing misjudgment caused by sensor information errors;

[0048] The feature projection extraction algorithm proposed in the present invention: by using feature vector projection transformation, the calculation of feature vector groups can well describe the characteristics of different environmental subjects, which is conducive to the model's evaluation of the real environment and provides accurate data support for subsequent calculations;

[0049] The temporal autoregressive model proposed in the present invention uses timestamps as prediction nodes and arranges the values ​​of each sensor input into a sequence, which is conducive to prediction based on timestamps;

[0050] The neural network model proposed in the present invention is combined with spatial differentiation technology to further optimize the characteristic variance calculation and improve the overall prediction accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a flow chart of the sensor dynamic information linkage analysis method based on feature matching and fusion of the present invention;

[0052] Figure 2 It is a flow chart of step S4 in the sensor dynamic information linkage analysis method based on feature matching and fusion of the present invention; Implementation

[0053] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below;

[0054] It should be noted that the terms "first", "second", "symmetrical", "array", etc. are only used to distinguish between description and position description purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the definition of "first", "symmetrical", etc. can explicitly or implicitly include one or more of these features; similarly, when the quantity of certain features is not limited in the form of words such as "two" or "three", it should be noted that this feature also explicitly or implicitly includes one or more feature quantities;

[0055] In the present invention, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may mean that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, a first feature being "above", "above" or "above" a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. A first feature being "below", "below" or "below" a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature; at the same time, all axial descriptions, such as the X-axis, Y-axis, Z-axis, one end of the X-axis, the other end of the Y-axis or the other end of the Z-axis, etc., are based on a Cartesian coordinate system.

[0056] In the present invention, unless otherwise clearly specified and limited, the terms such as "installation", "connection", "fixation" and the like should be understood in a broad sense; for example, it can be a fixed connection, a detachable connection, or an integral molding; it can be a mechanical connection, a direct connection, welding, or an indirect connection through an intermediate medium, or it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood based on the description and drawings combined with specific circumstances.

[0057] In the existing technology, there is a lack of methods for monitoring geological hazards by fusing multiple sensors; for this purpose, please refer to Figure 1-Figure 2 , the present invention provides a sensor dynamic information linkage analysis method based on feature matching and fusion to solve the above technical problems;

[0058] Please refer to Figure 1-Figure 2 ,The present invention proposes a sensor dynamic information linkage analysis method based on feature matching and fusion;

[0059] In this embodiment, a sensor dynamic information linkage analysis method based on feature matching fusion is proposed, which includes the following steps:

[0060] S1: Acquire measurement data input by a hybrid distributed sensor, and vectorize the measurement data to obtain a first parameter vector group;

[0061] S2: performing unbiased estimation judgment on the first parameter vector group, and converging the first parameter vector group according to the estimated probability after judgment to obtain a second parameter vector group;

[0062] S3: using a time autoregressive model to refer to the second parameter vector group to form a time predictor, and the time predictor outputs a third parameter vector group;

[0063] S4: Input the second parameter vector group and the third parameter vector group into the Hadoop platform, and analyze the second parameter vector group and the third parameter vector group according to a preset neural network model.

[0064] In this embodiment,

[0065] S1: Data collection and vectorization

[0066] In this step, we first obtain the measurement data of the mixed distributed sensor input. These sensors may be located in different places and may monitor various types of information, such as temperature, humidity, pressure, etc. After obtaining these data, these measurement data are vectorized for mathematical processing and analysis. Vectorization means converting these data into numerical form so that they can be effectively processed in the algorithm.

[0067] S2: Unbiased Estimation and Convergence

[0068] The purpose of this step is to evaluate the accuracy of the data collected from the sensor. Unbiased estimation aims to determine whether the data deviates from the true value and determine the possibility or probability of these deviations. Once these deviations are determined, these estimated probabilities can be used to correct or "converge" the data to make it closer to the true value. In this way, a more accurate, corrected data vector can be obtained, that is, the second parameter vector group;

[0069] S3: Time prediction

[0070] This step uses a temporal autoregressive model. A temporal autoregressive model is a statistical method used to predict time series data. With reference to the second parameter vector group, a temporal predictor is constructed; this means that the previous data (such as the observations at the previous few time points) are used to predict future data. The output of this predictor is a third parameter vector group, which may be the predicted value at a certain time point in the future.

[0071] S4: Hadoop platform and neural network analysis

[0072] The second parameter vector group and the third parameter vector group are input into the Hadoop platform. Hadoop is an open source, distributed storage and computing framework that is often used to process big data. On the Hadoop platform, data can be analyzed and processed, especially in a distributed environment. Further, the data is analyzed according to a preset neural network model. The neural network is a powerful machine learning algorithm that is often used for recognition, classification, and prediction tasks. Here, it may be used to further analyze and understand the data obtained from the sensor, as well as the predicted data.

[0073] Specifically, measurement data of hybrid distributed sensor input are obtained, which cover input data of multiple sensors. These measurement data are vectorized to obtain a first parameter vector group. The information that can be expressed in the first parameter vector group includes: sensing information, functional information and position information, etc. The sensing information and position information can be used in this scheme, and the functional information can be used to judge whether the sensor functions are intact. After obtaining the first parameter vector group, it is necessary to verify the accuracy of its information transmission, make an unbiased estimation and judgment on the first parameter vector group, converge the first vector parameter group according to the estimated probability, and obtain a second vector parameter group. The second vector parameter group is more accurate than the first vector parameter group. Subsequently, a time autoregressive model is used to predict the hybrid distributed sensor information in the second parameter vector group with timestamps as nodes to obtain a third parameter vector group. The third parameter vector group is input into a neural network model for analysis to obtain analysis results of the occurrence of geological disasters.

[0074] In one embodiment, the step S4 includes:

[0075] S41: Identify the input time of the second parameter vector group and the timestamp of forming the third parameter vector group;

[0076] S42: performing feature projection transformation on the third parameter vector group to obtain a feature vector group corresponding to the hybrid distributed sensor;

[0077] S43: Perform spatial differentiation on the eigenvector group to generate corresponding eigenvariance;

[0078] S44: Substitute the feature variance into the log-likelihood function simulated by the hybrid distributed sensor to calculate the real environment value of the hybrid distributed sensor;

[0079] S45: An environmental score is obtained by analyzing and calculating the actual environmental value, and compared with a preset threshold value. If the environmental score is less than the preset threshold value, there is a risk of a geological disaster.

[0080] Specific: S41: Timestamp identification

[0081] In this step, the input time is identified from the second parameter vector group, that is, the specific moment when the data is collected is determined. At the same time, a corresponding timestamp is generated for the third parameter vector group (output by the time predictor) to ensure that the subsequent analysis can be performed based on time;

[0082] S42: Feature Projection Transformation

[0083] The data in the third parameter vector group will undergo feature projection transformation, which is a dimensionality reduction method that aims to extract the main features from a large amount of data. After this transformation, a feature vector group corresponding to the hybrid distributed sensor is obtained. These feature vectors may be more compact, meaningful, and better represent the core characteristics of the original data;

[0084] S43: Spatial heterogeneity and characteristic variance

[0085] In this step, the feature vector groups are spatially differentiated, and the feature vectors are sorted or organized to identify the spatial pattern or structure in the data. The result of this step is the generation of corresponding feature variances, which can provide information about the variability of the data for subsequent analysis;

[0086] S44: Log-likelihood function and true environmental value

[0087] The feature variance will be substituted into a log-likelihood function simulated by the hybrid distributed sensor. The log-likelihood function is usually used to evaluate the fit of the model. Through this step, the real environment value of the hybrid distributed sensor can be calculated, which reflects the actual situation of the current environment.

[0088] S45: Environmental scores and risk assessment

[0089] Based on the actual environmental values, further analysis is carried out to calculate the environmental score, which is a quantitative representation of the actual environmental values. Then, the environmental score is compared with the preset threshold. If the environmental score is less than the preset threshold, it indicates the risk of geological disasters.

[0090] In one embodiment, the step of acquiring measurement data input by the hybrid distributed sensor and vectorizing the measurement data to obtain a first parameter vector group includes:

[0091] The measurement data includes hybrid distributed sensor information Y, and the hybrid distributed sensor information includes sensing information, functional information and location information;

[0092] Based on the hybrid distributed sensor information Perform vector transformation to obtain the first parameter vector group matching the measured data .

[0093] Further, the step of performing unbiased estimation judgment on the first parameter vector group and converging the first parameter vector group according to the estimated probability after judgment to obtain the second parameter vector group includes:

[0094] Perform unbiased estimation on the first parameter vector group, and Node estimation , each of the hybrid distributed sensor information meets the preset sensor threshold , then , determining that the error of the hybrid distributed sensor information is within a preset value;

[0095] The first parameter vector group after unbiased estimation is converged to the true value , get the second parameter vector group , where p is the coefficient of convergence to the true value.

[0096] In one embodiment, the step of using the time autoregressive model to refer to the second parameter vector group to form a time predictor, and the step of outputting the third parameter vector group by the time predictor includes:

[0097] Assume that the time series t in the temporal predictor is mixed with only one of the distributed sensor information There is a linear relationship, then:

[0098] ;

[0099] in, They are hybrid distributed sensor information by The weight of the impact, is independent of the hybrid distributed sensor information White noise sequence.

[0100] Specifically, for example, in landslide monitoring, rain sensor information is used as The third parameter vector group has a linear relationship under certain conditions. Other sensor information, such as soil moisture sensor information, soil pressure sensor information, etc., are related to the rain sensor information. With the change of the rain sensor information, the soil moisture sensor information, soil pressure sensor information, etc. also change, which ultimately leads to the change of the third parameter vector group.

[0101] In one embodiment, landslides are monitored using distributed sensors by collecting data from a point in the landslide soil:

[0102] ; P is the sliding force, R is the anti-sliding force balanced with the sliding force; the calculation formulas for the sliding force and the anti-sliding force balanced with the sliding force are , G is the gravity of the soil, is the slope angle of the mountain, ; is the vertical depth of the soil, is the soil density, , F is the cohesion in the soil, e is the moisture index in the soil, r is the water-soil density ratio in the soil, is the soil internal friction angle; in equilibrium, , this state is the normal state. On rainy days, due to the influence of soil friction, R increases with the increase of P. After the soil friction reaches the maximum value, if P still increases, the soil moves. After a layer of soil slides, the stress of other soils changes, and then a landslide occurs. In this embodiment, the data received by the distributed sensor can be used as the first parameter vector group to finally simulate the real environment of a point in the landslide soil. Under the time series t, these data do not introduce external change factors, and their real environment will basically not change; and the main factor causing the landslide is rainfall. The change of the moisture index e in the soil takes into account the rainfall:

[0103] ; Where C is the effective rainfall, k is the hydraulic conductivity, and b is the unit water storage area; then:

[0104] ;

[0105] The influence of the data of the rain sensor on the third parameter vector group is linear. As the rainfall increases, the data in the third parameter vector group is further reduced or increased according to the size of the weight.

[0106] Furthermore, the step of performing feature projection transformation on the third parameter vector group to obtain a feature vector group corresponding to the hybrid distributed sensor includes:

[0107] The feature projection algorithm is:

[0108] ;

[0109] In the formula, the neural network convolution layer h=1, 2, ..., m preset in the neural network model; L is the preset value of each hybrid distributed sensor in the neural network convolution layer h, and X is a feature vector group.

[0110] Furthermore, the step of performing spatial differentiation on the feature vector group to generate corresponding feature variances includes:

[0111] ;

[0112] In the formula, x is one of the vectors in the feature vector group X. Each vector in the feature vector group X is spatially differentiated, and the convolution coefficient corresponding to the convolution layer h is Calculate the characteristic variance .

[0113] Furthermore, the step of substituting the characteristic variance into the log-likelihood function simulated by the hybrid distributed sensor to calculate the real environment value of the hybrid distributed sensor includes:

[0114] The log-likelihood function is:

[0115] ;

[0116] Real environment value The characteristic variance , convolutional layer coefficients , each vector x in the eigenvector group X is calculated.

[0117] Furthermore, the step of analyzing and calculating the environmental score based on the real environmental value includes:

[0118] The real environment value Substituting into the scoring algorithm, the scoring algorithm is:

[0119] ;

[0120] Among them, the environmental score is S, is the corresponding variance produced by the simulation.

[0121] In one embodiment, the specific implementation details of the sensor dynamic information linkage analysis method based on feature matching fusion applied to landslide monitoring include the following steps:

[0122] Obtaining measurement data from hybrid distributed sensors: Obtaining the required measurement data from each sensor; for example, this may include multiple sensors such as inclinometers, meteorological sensors, and soil sensors;

[0123] Vectorize measurement data: vectorize the acquired measurement data and convert it into vector form; this converts the measurement value of each sensor into a corresponding vector;

[0124] Unbiased estimation judgment and parameter vector group convergence: Perform unbiased estimation judgment on the first parameter vector group to judge the parameter value and obtain the estimated probability. According to the estimated probability, perform convergence operation on the first parameter vector group to obtain the second parameter vector group. At this time, the second parameter vector group is more accurate than the first parameter vector group.

[0125] Constructing a time predictor and outputting a third parameter vector group: constructing a time autoregressive model predictor using the second parameter vector group; outputting a third parameter vector group through the time predictor, the vector group including parameter values ​​obtained through time prediction, and the information included in the third parameter vector group includes: prediction data of multiple sensors such as inclinometer sensors, meteorological sensors, and soil sensors after the timestamp;

[0126] Identify input time and form timestamp: According to the input time of the second parameter vector group, identify and record its timestamp; the time information becomes the prediction starting point, and the timestamp is the prediction end point when the disaster is predicted to occur.

[0127] Feature projection transformation: Perform feature projection transformation on the third parameter vector group to obtain a feature vector group corresponding to the hybrid distributed sensor, and map the parameter vector group to a new feature space to express the actual landslide data;

[0128] Spatial differentiation and feature variance generation: Perform spatial analysis on the feature vector group and calculate the feature variation measure in space, i.e., feature variance;

[0129] Predict the real environment value of the hybrid distributed sensor: Substitute the feature variance into the log-likelihood function simulated by the hybrid distributed sensor to calculate the real environment value of the hybrid distributed sensor. The specific form of the log-likelihood function will depend on the model selection and data characteristics.

[0130] Analysis and calculation based on real environmental values: Use the calculated real environmental values ​​for further environmental analysis and calculation, and use geological space models and geological force analysis techniques to determine the size of the real environmental values;

[0131] Determine the risk of landslide: Determine the risk of landslide by comparing it with the preset threshold. If the actual environment score is lower than the preset threshold, it is considered that there is a possibility of landslide.

[0132] Of course, the present invention may have many other implementations. Based on this implementation, other implementations obtained by ordinary technicians in this field without any creative work are all within the scope of protection of the present invention.

Claims

1. Sensor dynamic information linkage analysis method based on feature matching and fusion, It is characterized in that The following steps are involved: S1: Acquire measurement data input by a hybrid distributed sensor, and vectorize the measurement data to obtain a first parameter vector group; S2: performing unbiased estimation judgment on the first parameter vector group, and converging the first parameter vector group according to the estimated probability after judgment to obtain a second parameter vector group; S3: using a time autoregressive model to refer to the second parameter vector group to form a time predictor, and the time predictor outputs a third parameter vector group; S4: inputting the second parameter vector group and the third parameter vector group into the Hadoop platform, and analyzing the second parameter vector group and the third parameter vector group according to a preset neural network model; The S4 step includes: S41: Identify the input time of the second parameter vector group and the timestamp of forming the third parameter vector group; S42: performing feature projection transformation on the third parameter vector group to obtain a feature vector group corresponding to the hybrid distributed sensor; S43: Perform spatial differentiation on the eigenvector group to generate corresponding eigenvariance; S44: Substitute the feature variance into the log-likelihood function simulated by the hybrid distributed sensor to calculate the real environment value of the hybrid distributed sensor; S45: An environmental score is obtained by analyzing and calculating the actual environmental value, and compared with a preset threshold value. If the environmental score is less than the preset threshold value, there is a risk of a geological disaster.

2. According to claim 1, the sensor dynamic information linkage analysis method based on feature matching fusion, It is characterized in that The step of obtaining measurement data input by the hybrid distributed sensor and vectorizing the measurement data to obtain a first parameter vector group includes: The measurement data includes hybrid distributed sensor information Y, and the hybrid distributed sensor information includes sensing information, functional information and location information; Based on the hybrid distributed sensor information Perform vector transformation to obtain the first parameter vector group matching the measured data .

3. According to claim 2, the sensor dynamic information linkage analysis method based on feature matching fusion, It is characterized in that The step of performing an unbiased estimation judgment on the first parameter vector group, and converging the first parameter vector group according to the estimated probability after the judgment to obtain the second parameter vector group includes: performing an unbiased estimation judgment on the first parameter vector group, and converging the first parameter vector group according to the estimated probability after the judgment to obtain the second parameter vector group. Node estimation , each of the hybrid distributed sensor information meets the preset sensor threshold , then , determining that the error of the hybrid distributed sensor information is within a preset value; The first parameter vector group after unbiased estimation is converged to the true value , get the second parameter vector group , where p is the coefficient of convergence to the true value.

4. According to claim 3, the sensor dynamic information linkage analysis method based on feature matching fusion, It is characterized in that The step of using the time autoregressive model to refer to the second parameter vector group to form a time predictor, and having the time predictor output a third parameter vector group includes: Assume that the time series t in the temporal predictor is mixed with only one of the distributed sensor information There is a linear relationship, then: ; in, They are hybrid distributed sensor information by The weight of the impact, is independent of the hybrid distributed sensor information White noise sequence.

5. According to claim 4, the sensor dynamic information linkage analysis method based on feature matching fusion, It is characterized in that The step of performing feature projection transformation on the third parameter vector group to obtain a feature vector group corresponding to the hybrid distributed sensor includes: The feature projection algorithm is: ； In the formula, the neural network convolution layer h=1, 2, ..., m preset in the neural network model; L is the preset value of each hybrid distributed sensor in the neural network convolution layer h, and X is a feature vector group.

6. According to claim 5, the sensor dynamic information linkage analysis method based on feature matching and fusion, It is characterized in that The step of performing spatial differentiation on the eigenvector group to generate the corresponding eigenvariance comprises: ； In the formula, x is one of the vectors in the feature vector group X. Each vector in the feature vector group X is spatially differentiated, and the convolution coefficient corresponding to the convolution layer h is Calculate the characteristic variance .

7. The sensor dynamic information linkage analysis method based on feature matching and fusion according to claim 6, It is characterized in that The step of substituting the characteristic variance into the log-likelihood function simulated by the hybrid distributed sensor to calculate the real environment value of the hybrid distributed sensor includes: The log-likelihood function is: ; Real environment value The characteristic variance , convolutional layer coefficients , each vector x in the eigenvector group X is calculated.

8. According to claim 7, the sensor dynamic information linkage analysis method based on feature matching and fusion, It is characterized in that The step of analyzing and calculating the environmental score based on the real environmental value includes: The real environment value Substituting into the scoring algorithm, the scoring algorithm is: ; Among them, the environmental score is S, is the corresponding variance produced by the simulation.

Images