Geothermal development risk monitoring method and device combined with multidimensional data analysis
By adopting multidimensional data analysis in geothermal development combined with IoT perceived topological network, the problems of poor accuracy and insufficient predictiveness of risk monitoring in the existing technology are solved, and more efficient risk monitoring and prediction are achieved.
Patent Information
- Application Number
- CN202411718404.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-28
- Publication Date
- 2025-02-18
- Estimated Expiration
- 2044-11-28
Smart Images
Figure CN119203011B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of geothermal development technology, and in particular to a geothermal development risk monitoring method and device combined with multi-dimensional data analysis. Background Art
[0002] As a clean and renewable energy source, the development and utilization of geothermal energy can not only effectively reduce dependence on fossil fuels, but also help mitigate climate change. However, there are many risks in the geothermal development process, including geological uncertainty, environmental impact and equipment failure, which in turn lead to increased project costs, development delays and even environmental disasters.
[0003] Existing risk monitoring technologies for geothermal development mainly rely on geological exploration technology and traditional monitoring equipment, which have limitations in processing and risk assessment of multi-point monitoring data, and are difficult to effectively process and analyze large-scale, multi-dimensional monitoring data. There are technical problems such as poor accuracy and insufficient predictability of risk monitoring. Summary of the invention
[0004] The present invention provides a geothermal development risk monitoring method and device combined with multi-dimensional data analysis to solve the technical problems of poor risk monitoring accuracy and insufficient predictability in the prior art, and achieve the technical effect of improving the accuracy and predictability of risk monitoring.
[0005] In a first aspect, the present invention provides a geothermal development risk monitoring method combined with multidimensional data analysis, wherein the method comprises:
[0006] An Internet of Things perception topology network is deployed in the target geothermal development area, wherein the Internet of Things perception topology network includes K topological nodes, each topological node is deployed with at least one type of monitoring sensor, wherein K is an integer greater than or equal to 1.
[0007] The historical abnormality monitoring data of the K topological nodes are collected to obtain K historical node abnormality monitoring data sets, wherein each historical node abnormality monitoring data corresponds to a monitoring time node.
[0008] In combination with the monitoring time node, the K historical node abnormal monitoring data sets are traversed to perform centralized identification of neighboring data intervals, and the abnormal monitoring scales of the K nodes are determined.
[0009] K adaptive convolutional network layers are constructed based on the K node anomaly monitoring scales, and the K adaptive convolutional network layers are serialized in ascending order of the node anomaly monitoring scales to obtain an adaptive convolutional network layer sequence, and each adaptive convolutional network layer is used to extract the monitoring data of the K topological nodes within a preset monitoring period to obtain a global perception monitoring topology data set sequence.
[0010] The global perception monitoring topology data set sequence is traversed to perform feature recognition within the sequence to obtain a global perception monitoring topology feature set sequence.
[0011] A neighbor matrix interaction analysis is performed on the global perception monitoring topology feature set sequence to obtain a neighbor interaction perception monitoring feature set sequence.
[0012] The risk monitoring network is called to perform cross-scale feature fusion analysis on the neighbor interaction perception monitoring feature set sequence to obtain the risk monitoring result.
[0013] In a second aspect, the present invention further provides a geothermal development risk monitoring device combined with multi-dimensional data analysis, wherein the device comprises:
[0014] A perception topology deployment module, which is used to deploy an Internet of Things perception topology network in a target geothermal development area, wherein the Internet of Things perception topology network includes K topological nodes, each topological node is deployed with at least one type of monitoring sensor, wherein K is an integer greater than or equal to 1.
[0015] The historical data collection module is used to collect historical abnormality monitoring data of the K topological nodes to obtain K historical node abnormality monitoring data sets, wherein each historical node abnormality monitoring data corresponds to a monitoring time node.
[0016] A neighbor data identification module is used to combine the monitoring time nodes, traverse the K historical node abnormal monitoring data sets to perform centralized identification of neighbor data intervals, and determine the abnormal monitoring scales of the K nodes.
[0017] An adaptive convolution extraction module is used to construct K adaptive convolution network layers based on the K node anomaly monitoring scales, and serialize the K adaptive convolution network layers in order from small to large according to the node anomaly monitoring scales to obtain an adaptive convolution network layer sequence, and use each adaptive convolution network layer to extract the monitoring data of the K topological nodes within a preset monitoring period to obtain a global perception monitoring topology data set sequence.
[0018] A feature recognition module is used to traverse the global perception monitoring topology data set sequence to perform feature recognition within the sequence and obtain a global perception monitoring topology feature set sequence.
[0019] A neighbor analysis module, the neighbor analysis module is used to perform neighbor matrix interaction analysis on the global perception monitoring topology feature set sequence to obtain a neighbor interaction perception monitoring feature set sequence.
[0020] The risk monitoring and analysis module is used to call the risk monitoring network to perform cross-scale feature fusion analysis on the neighbor interaction perception monitoring feature set sequence to obtain risk monitoring results.
[0021] The present invention discloses a geothermal development risk monitoring method and device combined with multidimensional data analysis, including: laying out an Internet of Things perception topological network in a target geothermal development area, wherein the topological network includes K topological nodes, each node is equipped with at least one type of monitoring sensor, wherein K is an integer greater than or equal to 1. Performing historical abnormal monitoring data collection of K topological nodes to obtain K historical node abnormal monitoring data sets, each data set corresponding to a monitoring time node. In combination with the monitoring time node, the K historical node abnormal monitoring data sets are subjected to centralized identification of neighbor data intervals, thereby determining the abnormal monitoring scale of each node. Based on the abnormal monitoring scales of these nodes, K adaptive convolutional network layers are constructed, and these network layers are serialized according to the size order of the node abnormal monitoring scales to obtain an adaptive convolutional network layer sequence. Using each adaptive convolutional network layer, data is extracted from the monitoring data of the K topological nodes within a preset monitoring period, thereby obtaining a global perception monitoring topological data set sequence. Traversing the global perception monitoring topological data set sequence, performing feature recognition within the sequence, and obtaining a global perception monitoring topological feature set sequence. The feature set sequence is subjected to neighbor matrix interaction analysis to obtain a neighbor interaction perception monitoring feature set sequence. Finally, the risk monitoring network is called to perform cross-scale feature fusion analysis on the neighbor interaction perception monitoring feature set sequence to obtain the risk monitoring result. The geothermal development risk monitoring method and device combined with multidimensional data analysis disclosed in the present invention solves the technical problems of poor accuracy and insufficient predictability of risk monitoring, and achieves the technical effect of improving the accuracy and predictability of risk monitoring. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 It is a schematic flow chart of the geothermal development risk monitoring method combined with multi-dimensional data analysis of the present invention;
[0023] Figure 2 It is a structural schematic diagram of the geothermal development risk monitoring device combined with multi-dimensional data analysis of the present invention.
[0024] Explanation of the accompanying drawings: perception topology layout module 11, historical data collection module 12, neighbor data identification module 13, adaptive convolution extraction module 14, feature recognition module 15, neighbor analysis module 16, risk monitoring and analysis module 17. DETAILED DESCRIPTION
[0025] The technical solution provided in the embodiments of the present invention is to solve the technical problems of poor risk monitoring accuracy and insufficient predictability in the prior art. The overall idea adopted is as follows:
[0026] First, an IoT sensing topological network is deployed in the target geothermal development area, wherein the network includes K topological nodes, each node is equipped with at least one type of monitoring sensor, wherein K is an integer greater than or equal to 1; then, historical abnormal monitoring data collection of the K topological nodes is performed to obtain K historical node abnormal monitoring data sets, each data corresponding to a monitoring time node; then, in combination with the monitoring time node, the K historical node abnormal monitoring data sets are traversed to perform centralized identification of neighboring data intervals, and the abnormal monitoring scales of the K nodes are determined; then, based on the K node abnormal monitoring scales, K adaptive convolutional network layers are constructed, and according to The node anomaly monitoring scales are serialized in order from small to large to obtain an adaptive convolutional network layer sequence. Each adaptive convolutional network layer is used to extract the monitoring data of K topological nodes within the preset monitoring period to obtain a global perception monitoring topology data set sequence. Then, the global perception monitoring topology data set sequence is traversed to perform feature recognition within the sequence to obtain a global perception monitoring topology feature set sequence. Finally, the neighbor matrix interaction analysis is performed on the global perception monitoring topology feature set sequence to obtain a neighbor interaction perception monitoring feature set sequence, and the risk monitoring network is called to perform cross-scale feature fusion analysis to obtain the risk monitoring results.
[0027] The above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods of the specification to better understand the above technical solution. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments of the present invention. It should be understood that the present invention is not limited to the example embodiments used only to explain the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In addition, it should be noted that, for the convenience of description, only the parts related to the present invention are shown in the drawings, rather than all of them.
[0028] Embodiment 1
[0029] Figure 1 The figure is a flow chart of a geothermal development risk monitoring method combined with multi-dimensional data analysis according to the present invention, wherein the method comprises:
[0030] An Internet of Things perception topology network is deployed in the target geothermal development area, wherein the Internet of Things perception topology network includes K topological nodes, each topological node is deployed with at least one type of monitoring sensor, wherein K is an integer greater than or equal to 1.
[0031] Specifically, the perception topology network of the target area is composed of K topological nodes, where K ≥ 1, indicating that the area can have one or more nodes, and the topological nodes are distributed according to factors such as the size, geographical distribution, and monitoring needs of the geothermal development area. Each topological node can be distributed at different locations of geothermal resources, such as drilling wellheads, underground, around equipment, and environmental monitoring points.
[0032] Specifically, each topological node is equipped with at least one type of monitoring sensor, and the type and type of sensor are selected according to the needs of the monitoring target. Common sensor types include temperature sensors, pressure sensors, flow sensors, seismic sensors, gas sensors, humidity sensors, conductivity sensors, etc.
[0033] Specifically, each topological node is responsible for collecting relevant data about the surrounding environment or equipment, that is, collecting data through sensors and converting them into transmittable signals. All topological nodes are directly or indirectly connected to a central node or server to facilitate centralized management and data aggregation.
[0034] Optionally, each node has a certain edge computing capability to perform preliminary processing on the collected data, such as denoising, data format conversion, simple anomaly detection, etc., thereby reducing transmission pressure and real-time response delay.
[0035] By deploying an IoT sensing topology network in the target geothermal development area, multi-dimensional and real-time monitoring of the geothermal development area can be achieved, which will help improve the accuracy of monitoring data and the stability of the network, and provide data support for the intelligent management of the geothermal development area.
[0036] The historical abnormality monitoring data of the K topological nodes are collected to obtain K historical node abnormality monitoring data sets, wherein each historical node abnormality monitoring data corresponds to a monitoring time node.
[0037] Specifically, each topological node performs synchronous monitoring within a preset monitoring window, and associates the obtained data with a specific time node to generate K historical node abnormal monitoring data sets. The K historical node abnormal monitoring data sets are data collected at each topological node in the target geothermal development area that reflect the abnormal status of the equipment or system. The abnormal data obtained by each topological node at the historical monitoring time node is organized into a data set.
[0038] In combination with the monitoring time node, the K historical node abnormal monitoring data sets are traversed to perform centralized identification of neighboring data intervals, and the abnormal monitoring scales of the K nodes are determined.
[0039] Specifically, the abnormal monitoring scale refers to the time difference between two consecutive abnormal monitorings, that is, the time interval from one monitoring time node to the next monitoring time node. In the monitoring data of the target geothermal development area, the time difference between two adjacent monitoring time nodes will reflect the system's monitoring frequency and data update status during that time period.
[0040] Optionally, among the K topological nodes, different topological nodes and different types of sensors correspond to different anomaly monitoring scales. For each node, all its historical anomaly monitoring records are traversed, and the time difference between each two adjacent time nodes is analyzed to determine the corresponding node anomaly monitoring scale. The above anomaly monitoring scale determination strategy based on time node intervals helps to improve the analysis accuracy of monitoring data.
[0041] In some embodiments, in combination with the monitoring time node, the K historical node abnormal monitoring data sets are traversed to perform centralized identification of neighboring data intervals, and the K node abnormal monitoring scales are determined, including:
[0042] Traverse the K historical node abnormality monitoring data sets to perform neighbor data interval identification and obtain K abnormal neighbor data interval sets; calculate the means of the K abnormal neighbor data interval sets respectively to obtain K abnormal neighbor data interval means; calculate the fluctuation variances of the K abnormal neighbor data interval sets respectively, and determine K concentrated identification step sizes according to the fluctuation variances; use the K abnormal neighbor data interval means as K identification starting points, perform neighbor data interval concentrated identification on the K abnormal neighbor data interval sets according to the K concentrated identification steps, and obtain the K node abnormality monitoring scales.
[0043] Specifically, first, for each historical node, the time interval between adjacent time nodes is calculated as the neighbor data interval, and the historical abnormal data sets of K nodes are traversed to obtain K abnormal neighbor data interval sets. Exemplarily, the abnormal monitoring time nodes of node 1 are T1, T2, T3, ..., Tk, and the time intervals between adjacent time points are calculated: time interval 1 = T2-T1, time interval 2 = T3-T2, ..., time interval (k-1) = Tk-T(k-1).
[0044] Furthermore, for each node, the mean of the corresponding abnormal neighbor data interval set is calculated as the mean of the K abnormal neighbor data intervals, and the fluctuation variance of the K abnormal neighbor data interval set is calculated to measure the degree of fluctuation of the time interval. The larger the variance, the greater the change in the abnormal monitoring interval of the corresponding node and the greater the instability. Conversely, nodes with smaller variances show more regular monitoring intervals.
[0045] Specifically, according to the size of the fluctuation variance, a "centralized identification step" is determined for each node. If the variance is small, it means that the monitoring interval of the node is relatively stable, and a smaller step can be set for centralized identification. If the variance is large, it means that the monitoring interval of the node fluctuates greatly, and a larger step is set.
[0046] Specifically, starting from the mean value of each node, the abnormal monitoring interval set of each node is centrally identified according to its centralized identification step, and the monitoring interval value including the preset proportion of abnormal monitoring intervals is determined, and the output is the abnormal monitoring scale corresponding to the node. Exemplarily, if the mean value of the abnormal neighbor data interval of node 1 is 13 minutes and the centralized identification step is 5 minutes, then the centralized identification interval of node 1 will start from 13 minutes, with a step of 5 minutes, and the adjacent monitoring intervals will be divided into centralized areas. When the preset proportion (such as 90%) of the abnormal monitoring intervals are included in the divided centralized area, the centralized area is used as the abnormal monitoring scale of node 1.
[0047] Through the above steps, the abnormal monitoring data sets of K historical nodes are traversed, and the mean, variance and concentrated identification step of the monitoring interval of each node are calculated. The monitoring rules and potential instability between nodes can be identified, and effective analysis of the node abnormal monitoring data intervals can be achieved. Finally, the abnormal monitoring scale of each node is obtained, which is conducive to the subsequent efficient monitoring and analysis of each node.
[0048] In some implementations, obtaining the K node anomaly monitoring scales includes:
[0049] Iterate the K identification starting points in any direction in the K abnormal neighbor data interval set according to the K concentrated identification step lengths to obtain K first iteration abnormal neighbor data intervals, and use the directions from the K identification starting points to the K first iteration abnormal neighbor data intervals as K first iteration directions respectively; determine whether the concentrated identification density of the K first iteration abnormal neighbor data intervals is greater than or equal to the concentrated identification density of the K identification starting points, and if so, update the K first iteration abnormal neighbor data intervals to the K identification starting points and continue to iterate; if not, add the K first iteration directions to the K taboo iteration direction space, and then iterate the K identification starting points in the K abnormal neighbor data interval set again according to the K concentrated identification step lengths in any direction except the iteration directions stored in the K taboo iteration direction space until the preset number of iterations is met, and obtain K target abnormal neighbor data intervals; use the K target abnormal neighbor data intervals as K node abnormality monitoring scales.
[0050] Specifically, firstly, a step direction is randomly selected, and iteration is performed according to the centralized identification step of the node, including forward (increasing the time interval) or reverse (decreasing the time interval) iteration. Then, the abnormal neighbor data interval of the first iteration is calculated, that is, the data interval obtained by iteration starting from the identification starting point.
[0051] Specifically, for each node, the concentrated identification density of the first iteration abnormal neighbor data interval is calculated and compared with the concentrated identification density of the identification starting point. A larger concentrated identification density means that the data points in the area are more concentrated, and the abnormal neighbor data interval of this iteration may be more representative.
[0052] Specifically, if the concentrated recognition density of the first iteration is greater than or equal to the density of the recognition starting point, the iteration result is considered valid, the recognition starting point can be updated, and the iteration continues. Otherwise (i.e., the concentrated recognition density of the abnormal neighbor data interval of the first iteration is less than the concentrated recognition density of the recognition starting point), the iteration in this direction is considered inappropriate, and the direction is added to the taboo iteration direction space, and it is forbidden to use this direction in subsequent iterations.
[0053] Specifically, after excluding the taboo direction, iteration is still performed in any direction in the remaining optional directions according to the concentrated identification step size, and the first iterative abnormal neighbor data interval and the corresponding concentrated identification density after iterative update are continuously calculated.
[0054] Specifically, when the number of iterations reaches the preset maximum number, if the appropriate target abnormal neighbor data interval is still not obtained, the iteration process is terminated, and the iterative abnormal neighbor data interval with the highest identification density in the corresponding set is used as the target abnormal neighbor data interval. Then, K nodes are traversed, and K target abnormal neighbor data intervals are used as K node abnormality monitoring scales.
[0055] Through the above-mentioned iterative optimization method, the abnormal monitoring scale of each node is gradually adjusted to obtain a more accurate monitoring interval, which helps to track and analyze abnormal events more effectively, thereby improving the performance and sensitivity of the monitoring system.
[0056] Further, the K taboo iteration direction spaces include K preset taboo iteration numbers, wherein the iteration directions added into the K taboo iteration direction spaces are forbidden to have the K preset taboo iteration numbers in the iteration.
[0057] Specifically, the taboo iteration direction space is a space containing multiple "taboo directions". In each iteration, if an iteration direction does not meet the requirements (for example, the concentrated recognition density of its recognition results is low), the direction will be added to the taboo iteration direction space. After being added, the direction will be disabled in subsequent iterations and cannot be selected again.
[0058] Specifically, each direction added to the taboo iteration direction space corresponds to a preset "taboo iteration count". For example, if an iteration direction is judged to be inappropriate and added to the taboo space, the taboo count of the direction is set to 1, indicating that the direction is prohibited from being used once in subsequent iterations; if the direction is judged to be inappropriate again, the taboo count is updated, and so on. When the taboo count of a direction reaches the preset maximum number, the direction can be re-added to the iteration direction to ensure that the best node anomaly monitoring scale is obtained as much as possible.
[0059] K adaptive convolutional network layers are constructed based on the K node anomaly monitoring scales, and the K adaptive convolutional network layers are serialized in ascending order of the node anomaly monitoring scales to obtain an adaptive convolutional network layer sequence, and each adaptive convolutional network layer is used to extract the monitoring data of the K topological nodes within a preset monitoring period to obtain a global perception monitoring topology data set sequence.
[0060] Specifically, by constructing an adaptive convolutional network layer based on K node anomaly monitoring scales and performing serial processing in the order of the size of the node anomaly monitoring scales, the node monitoring data in the target IoT perception topology network is gradually analyzed. Among them, the anomaly monitoring scale provides an adaptive scale basis for the convolutional field of view of the convolutional network. In other words, the parameters of the K adaptive convolutional network layers are set according to the K node anomaly monitoring scales.
[0061] Optionally, a convolutional layer is configured according to the abnormality monitoring scale of each node. Each convolutional network layer adjusts its receptive field, convolution kernel size, convolution step size and other parameters according to the abnormality monitoring scale of its corresponding node to ensure the best response capability to fluctuations of different scales, and generates K adaptive convolutional network layers. For example, if the abnormality monitoring scale of the node is small (that is, when the node fluctuates faster), the corresponding convolutional layer will be refined into a higher resolution convolution kernel to capture more fine-grained time information.
[0062] Specifically, in order to handle the fluctuation response of nodes at different scales, the convolutional network layer performs serial processing in the order of the abnormal monitoring scale of the nodes from small to large, so as to first focus on nodes with faster response and stronger fluctuation, and gradually turn to nodes with slower response and weaker fluctuation, to ensure that the information within each scale range can be fully extracted. By processing the adaptive convolutional network layer in the above order from small to large, it can be ensured that fine-grained abnormal changes (i.e., rapid fluctuations) to macro trends (i.e., slower fluctuations) can be effectively monitored and extracted. In other words, the serialization process is similar to the "progressive" analysis method, ensuring that abnormal changes in any time period are not missed.
[0063] Specifically, multiple adaptive convolutional network layers perform feature extraction on their corresponding node data. Through filtering by the adaptive convolutional network layers, it is possible to focus on abnormal fluctuations within a specific time interval based on the data characteristics of multiple nodes, and obtain a global perception monitoring topology data set sequence. This sequence contains the abnormal monitoring data of all topological nodes within the preset monitoring period, forming a multi-level network topology data stream with rich time information, providing accurate multi-time scale information for subsequent decision-making.
[0064] By constructing an adaptive convolutional network layer based on K node anomaly monitoring scales, we can achieve step-by-step and hierarchical monitoring data extraction, and finally obtain a global perception monitoring topology data set sequence. This process combines detailed and macro analysis and data extraction, ensuring that the time response differences between nodes will not affect the overall monitoring results.
[0065] The global perception monitoring topology data set sequence is traversed to perform feature recognition within the sequence to obtain a global perception monitoring topology feature set sequence.
[0066] In some embodiments, traversing the global perception monitoring topology data set sequence to perform intra-sequence feature recognition to obtain a global perception monitoring topology feature set sequence includes:
[0067] Acquire multiple sample global perception monitoring topology data and multiple sample global perception monitoring topology features as training sample data; divide the training sample data into a training set and a validation set according to a preset ratio; use the training set to train a framework built based on a feedforward neural network, and use the validation set to supervise the training process, learn the mapping relationship between the global perception monitoring topology data and the global perception monitoring topology features, until the training converges, and obtain a trained feature recognition network layer; use the feature recognition network layer to perform intra-sequence feature recognition on the global perception monitoring topology data set sequence, and obtain the global perception monitoring topology feature set sequence.
[0068] Specifically, first, obtain multiple sample global perception monitoring topology data and multiple sample global perception monitoring topology features, representing the monitoring data at different time nodes and between different nodes, and the relationship between the corresponding topological features, where the topological features are used to reflect the key information of the monitoring data, such as the correlation between nodes, abnormal fluctuations, etc.; then, divide the training sample data into a training set and a validation set according to a preset ratio, the training set is used to train the neural network, and the validation set is used to evaluate the generalization ability of the network to avoid overfitting.
[0069] Specifically, the mapping relationship between sample global perception monitoring topology data and sample global perception monitoring topology features is learned through a feedforward neural network, so as to obtain the corresponding relationship between the global perception monitoring topology data and the global perception monitoring topology features as a feature recognition network layer.
[0070] Specifically, the feature recognition network layer selects the corresponding feature extraction strategy (such as convolution kernel) to extract effective features according to the data characteristics of multiple data sets in the input global perception monitoring topology data set sequence, and maps the acquired features to the corresponding topological feature set to obtain the global perception monitoring topology feature set sequence.
[0071] The above-mentioned feature extraction method combining convolutional neural network and attention mechanism can more flexibly respond to different types of monitoring data and dynamically select appropriate convolution kernels for feature extraction, thereby improving the accuracy of feature recognition.
[0072] A neighbor matrix interaction analysis is performed on the global perception monitoring topology feature set sequence to obtain a neighbor interaction perception monitoring feature set sequence.
[0073] Specifically, the global perception monitoring topology feature set sequence is a fine-grained analysis of the monitoring data, which easily ignores fluctuations with slower responses. Therefore, through the neighbor matrix interaction analysis of the global perception monitoring topology feature set sequence, the feature interaction of the global perception monitoring topology feature set at different time scales is realized, and information is extracted and fused from feature sets of different scales to obtain a more expressive neighbor interaction perception monitoring feature set sequence, and the hidden relationship of features at different time scales is obtained, thereby improving the feature expression and reasoning capabilities of the monitoring data.
[0074] In some embodiments, performing neighbor matrix interaction analysis on the global perception monitoring topology feature set sequence to obtain the neighbor interaction perception monitoring feature set sequence includes:
[0075] Extract the first global perception monitoring topology feature set that is at the first position and the second global perception monitoring topology feature set that is at the second position in the global perception monitoring topology feature set sequence; perform neighbor matrix interaction analysis on the first global perception monitoring topology feature set and the second global perception monitoring topology feature set to obtain a first neighbor interaction perception monitoring feature set; extract the third global perception monitoring topology feature set that is at the third position in the global perception monitoring topology feature set sequence again; perform neighbor matrix interaction analysis based on the second global perception monitoring topology feature set and the third global perception monitoring topology feature set to obtain a second neighbor interaction perception monitoring feature set; perform multiple neighbor matrix interaction analyses in sequence to obtain the Kth neighbor interaction perception monitoring feature set; obtain the neighbor interaction perception monitoring feature set sequence based on the first neighbor interaction perception monitoring feature set, the second neighbor interaction perception monitoring feature set and the Kth neighbor interaction perception monitoring feature set.
[0076] Specifically, for the first global perception monitoring topology feature set and the second global perception monitoring topology feature set, a neighbor matrix interaction analysis is performed, that is, the features of two different scales are mapped through the adjacency matrix, so that the information in the first feature set can interact with the information in the second feature set, and the enhanced features are extracted to generate the first neighbor interaction perception monitoring feature set.
[0077] The adjacency matrix can capture the relationship between the two scales and avoid the redundancy caused by information fusion. At the same time, the graph convolution operation can further process the interactive information and output richer features.
[0078] Similarly, the second global perception monitoring topology feature set and the third global perception monitoring topology feature set, ..., the k-1th global perception monitoring topology feature set and the kth global perception monitoring topology feature set are further subjected to neighbor matrix interaction analysis to obtain the Kth neighbor interaction perception monitoring feature set.
[0079] Specifically, the neighbor interaction perception monitoring feature set sequence is composed of interactive information of multiple different scales. It is a collection of feature information after effectively integrating multi-level and multi-time scale features, forming a more comprehensive and accurate system monitoring feature, which in turn helps to better analyze the monitoring data and facilitates subsequent risk monitoring and analysis.
[0080] In some implementations, obtaining a first neighbor interaction awareness monitoring feature set includes:
[0081] The first global perception monitoring topology feature set and the second global perception monitoring topology feature set are subjected to an inner product calculation and a softmax calculation to obtain a first neighbor interaction matrix; the first neighbor interaction matrix and the first global perception monitoring topology feature set are subjected to a convolution operation to obtain a first neighbor interaction perception monitoring feature set.
[0082] Optionally, firstly, based on the attention mechanism, local spatial enhancement is performed on the first feature set and the second feature set respectively, so as to focus on important parts of the feature sets and suppress redundant information.
[0083] Specifically, the inner product calculation is used to interact the first feature set and the second feature set to obtain the correlation information between them; then, the inner product result is subjected to a Softmax operation to obtain a first neighbor interaction matrix, which represents the correspondence and interaction information between the features of the first feature set and the second feature set at different scales. In other words, the first neighbor interaction matrix captures the correlation between the first feature set and the second feature set in the local space; finally, the graph convolution operation is used to convolve the first neighbor interaction matrix with the first global perception monitoring topology feature set to obtain a feature set after multi-scale information fusion, which is output as the first neighbor interaction perception monitoring feature set.
[0084] Through the above process, the first neighbor interactive perception monitoring feature set obtained contains the monitoring features fused from the first feature set and the second feature set at two feature scales, thereby effectively integrating the complementary information of different feature sets, improving the perception ability of the complex system state, and providing more accurate and comprehensive support for subsequent risk monitoring.
[0085] The risk monitoring network is called to perform cross-scale feature fusion analysis on the neighbor interaction perception monitoring feature set sequence to obtain the risk monitoring result.
[0086] Specifically, the neighbor interaction perception monitoring feature set sequence is used as input data, and a cross-scale feature fusion analysis is performed through a preset risk monitoring network to obtain the risk monitoring results of the target geothermal development area. Among them, the risk monitoring network is an existing, pre-set risk analysis model, which is used to analyze and judge multiple monitoring data sources to assess potential risks. Exemplarily, the risk monitoring network is composed of a multi-level neural network, a decision tree or other types of machine learning models, which can process feature information from different monitoring data.
[0087] By calling on the existing risk monitoring network, while improving the accuracy and reliability of monitoring results, the existing risk prediction and analysis tools can be effectively reused to avoid the waste of resources in repeatedly building risk monitoring networks, thereby helping to reduce the cost of improving the monitoring capabilities of existing geothermal development risk monitoring scenarios.
[0088] In summary, the geothermal development risk monitoring method combined with multidimensional data analysis provided by the present invention has the following technical effects:
[0089] By laying out the Internet of Things perception topological network in the target geothermal development area, the topological network includes K topological nodes, each node is equipped with at least one type of monitoring sensor, where K is an integer greater than or equal to 1. Perform historical abnormal monitoring data collection of K topological nodes to obtain K historical node abnormal monitoring data sets, each data set corresponding to a monitoring time node. Combined with the monitoring time node, the K historical node abnormal monitoring data sets are centrally identified by neighbor data intervals, so as to determine the abnormal monitoring scale of each node. Based on the abnormal monitoring scales of these nodes, K adaptive convolutional network layers are constructed, and these network layers are serialized according to the size order of the node abnormal monitoring scales to obtain an adaptive convolutional network layer sequence. Each adaptive convolutional network layer is used to extract the monitoring data of K topological nodes within a preset monitoring period, thereby obtaining a global perception monitoring topological data set sequence. The global perception monitoring topological data set sequence is traversed, and feature recognition is performed within the sequence to obtain a global perception monitoring topological feature set sequence. The feature set sequence is subjected to neighbor matrix interaction analysis to obtain a neighbor interaction perception monitoring feature set sequence. Finally, the risk monitoring network is called to perform cross-scale feature fusion analysis on the neighbor interaction perception monitoring feature set sequence to obtain the risk monitoring results, thereby achieving the technical effect of improving the accuracy and predictability of risk monitoring.
[0090] Embodiment 2
[0091] Figure 2 Schematic diagram of the structure of the geothermal development risk monitoring device combined with multi-dimensional data analysis of the present invention. For example, Figure 1 The flow chart of the geothermal development risk monitoring method combined with multidimensional data analysis in the present invention can be shown as follows: Figure 2 The structure shown is implemented.
[0092] Based on the same concept as the geothermal development risk monitoring method combined with multidimensional data analysis in the above embodiment, the present invention also provides a geothermal development risk monitoring device combined with multidimensional data analysis, including:
[0093] The perception topology deployment module 11 is used to deploy the Internet of Things perception topology network in the target geothermal development area, wherein the Internet of Things perception topology network includes K topological nodes, each topological node is deployed with at least one type of monitoring sensor, wherein K is an integer greater than or equal to 1.
[0094] The historical data collection module 12 is used to collect the historical abnormality monitoring data of the K topological nodes to obtain K historical node abnormality monitoring data sets, wherein each historical node abnormality monitoring data corresponds to a monitoring time node.
[0095] The neighbor data identification module 13 is used to combine the monitoring time nodes, traverse the K historical node abnormal monitoring data sets to perform centralized identification of neighbor data intervals, and determine the abnormal monitoring scales of the K nodes.
[0096] The adaptive convolution extraction module 14 is used to construct K adaptive convolution network layers based on the K node anomaly monitoring scales, and serialize the K adaptive convolution network layers in order from small to large according to the node anomaly monitoring scales to obtain an adaptive convolution network layer sequence, and use each adaptive convolution network layer to extract the monitoring data of the K topological nodes within a preset monitoring period to obtain a global perception monitoring topology data set sequence.
[0097] The feature recognition module 15 is used to traverse the global perception monitoring topology data set sequence to perform feature recognition within the sequence and obtain a global perception monitoring topology feature set sequence.
[0098] The neighbor analysis module 16 is used to perform neighbor matrix interaction analysis on the global perception monitoring topology feature set sequence to obtain a neighbor interaction perception monitoring feature set sequence.
[0099] The risk monitoring and analysis module 17 is used to call the risk monitoring network to perform cross-scale feature fusion analysis on the neighbor interaction perception monitoring feature set sequence to obtain the risk monitoring result.
[0100] Among them, the neighbor data identification module 13 includes:
[0101] The neighbor data interval identification unit is used to traverse the K historical node abnormal monitoring data sets to perform neighbor data interval identification and obtain K abnormal neighbor data interval sets.
[0102] The data interval mean calculation unit is used to calculate the means of the K abnormal neighbor data interval sets respectively to obtain the means of the K abnormal neighbor data intervals.
[0103] The fluctuation variance calculation unit is used to calculate the fluctuation variance of the K abnormal neighbor data interval sets respectively, and determine the K concentrated identification step sizes according to the fluctuation variance sizes.
[0104] The centralized identification unit is used to use the mean values of the K abnormal neighbor data intervals as K identification starting points, perform centralized neighbor data interval identification on the K abnormal neighbor data interval sets according to the K centralized identification step sizes, and obtain the K node abnormality monitoring scales.
[0105] In some implementations, the centralized identification unit in the neighbor data identification module 13 includes:
[0106] An iterative identification unit is used to iterate the K identification starting points in any direction in the K abnormal neighbor data interval sets according to the K concentrated identification steps to obtain K first iterative abnormal neighbor data intervals, and respectively use the directions from the K identification starting points to the K first iterative abnormal neighbor data intervals as K first iterative directions.
[0107] The concentrated identification density judgment unit is used to judge whether the concentrated identification density of the K first iteration abnormal neighbor data intervals is greater than or equal to the concentrated identification density of the K identification starting points. If so, the K first iteration abnormal neighbor data intervals are updated to K identification starting points and the iteration is continued.
[0108] The taboo update iteration unit is used to add the K first iteration directions to the K taboo iteration direction spaces if not, and then iterate the K identification starting points in any direction except the iteration directions stored in the K taboo iteration direction spaces in the K abnormal neighbor data interval sets according to the K concentrated identification steps again until the preset number of iterations is met to obtain K target abnormal neighbor data intervals.
[0109] The taboo iteration management unit is used to iterate the K identification starting points in the K abnormal neighbor data interval sets in any direction except the iteration direction stored in the K taboo iteration direction space according to the K concentrated identification step sizes again until a preset number of iterations is met to obtain K target abnormal neighbor data intervals.
[0110] The abnormality monitoring scale determination unit is used to use the K target abnormal neighbor data intervals as K node abnormality monitoring scales.
[0111] Further, the K taboo iteration direction spaces include K preset taboo iteration numbers, wherein the iteration directions added into the K taboo iteration direction spaces are forbidden to have the K preset taboo iteration numbers in the iteration.
[0112] In some embodiments, the feature recognition module 15 includes:
[0113] The training data preparation unit is used to obtain a plurality of sample global perception monitoring topology data and a plurality of sample global perception monitoring topology features as training sample data.
[0114] The data division unit is used to divide the training sample data into a training set and a validation set according to a preset ratio.
[0115] The training process unit is used to train the framework built based on the feedforward neural network using the training set, and supervise the training process using the verification set, to learn the mapping relationship between the global perception monitoring topology data and the global perception monitoring topology features, until the training converges, and obtain the trained feature recognition network layer.
[0116] The feature recognition unit is used to perform intra-sequence feature recognition on the global perception monitoring topology data set sequence using the feature recognition network layer to obtain the global perception monitoring topology feature set sequence.
[0117] In some embodiments, the neighbor analysis module 16 includes:
[0118] The feature extraction unit is used to extract a first global perception monitoring topology feature set located at the first position and a second global perception monitoring topology feature set located at the second position in the global perception monitoring topology feature set sequence.
[0119] The interactive analysis unit is used to perform neighbor matrix interactive analysis on the first global perception monitoring topology feature set and the second global perception monitoring topology feature set to obtain a first neighbor interactive perception monitoring feature set.
[0120] The feature extraction unit is used to extract again a third global perception monitoring topology feature set located at a third position in the global perception monitoring topology feature set sequence.
[0121] The interaction analysis unit is used to perform neighbor matrix interaction analysis based on the second global perception monitoring topology feature set and the third global perception monitoring topology feature set to obtain a second neighbor interaction perception monitoring feature set.
[0122] The interaction analysis unit is used to perform multiple neighbor matrix interaction analyses in sequence to obtain the Kth neighbor interaction perception monitoring feature set.
[0123] The feature set generation unit is used to obtain the neighbor interaction perception monitoring feature set sequence according to the first neighbor interaction perception monitoring feature set, the second neighbor interaction perception monitoring feature set and the Kth neighbor interaction perception monitoring feature set.
[0124] In some implementations, the interactive analysis unit in the neighbor analysis module 16 includes:
[0125] An inner product calculation unit is used to perform an inner product calculation and a softmax calculation on the first global perception monitoring topology feature set and the second global perception monitoring topology feature set to obtain a first neighbor interaction matrix.
[0126] A convolution operation unit is used to perform a convolution operation on the first neighbor interaction matrix and the first global perception monitoring topology feature set to obtain a first neighbor interaction perception monitoring feature set.
[0127] It should be understood that the embodiments mentioned in this specification focus on their differences from other embodiments. The specific embodiments in the aforementioned embodiment one are also applicable to the geothermal development risk monitoring device combined with multi-dimensional data analysis described in embodiment two. For the sake of brevity of the specification, they will not be further elaborated here.
[0128] It should be understood that the embodiments disclosed in the present invention and the above description can enable those skilled in the art to use the present invention to implement the present invention. At the same time, the present invention is not limited to the above-mentioned embodiments. It should be understood that those skilled in the art can still modify the technical solutions recorded in the above-mentioned embodiments, or replace some of the technical features therein by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.
Claims
1. A geothermal development risk monitoring method combined with multidimensional data analysis, characterized in that: The method comprises: Deploy an Internet of Things sensing topology network in the target geothermal development area, wherein the Internet of Things sensing topology network includes K topological nodes, each topological node is deployed with at least one type of monitoring sensor, wherein K is an integer greater than or equal to 1; Execute the historical abnormality monitoring data collection of the K topological nodes to obtain K historical node abnormality monitoring data sets, wherein each historical node abnormality monitoring data corresponds to a monitoring time node; Combined with the monitoring time node, traverse the K historical node abnormal monitoring data sets to perform centralized identification of neighboring data intervals, and determine the K node abnormal monitoring scales; Based on the K node anomaly monitoring scales, K adaptive convolutional network layers are constructed, and the K adaptive convolutional network layers are serialized in order from small to large according to the node anomaly monitoring scales to obtain an adaptive convolutional network layer sequence, and each adaptive convolutional network layer is used to extract the monitoring data of the K topological nodes within a preset monitoring period to obtain a global perception monitoring topology data set sequence; Traversing the global perception monitoring topology data set sequence to perform feature recognition within the sequence to obtain a global perception monitoring topology feature set sequence; Performing neighbor matrix interaction analysis on the global perception monitoring topology feature set sequence to obtain a neighbor interaction perception monitoring feature set sequence; Calling the risk monitoring network to perform cross-scale feature fusion analysis on the neighbor interaction perception monitoring feature set sequence to obtain a risk monitoring result; Combined with the monitoring time node, the K historical node abnormal monitoring data sets are traversed to perform centralized identification of neighboring data intervals, and the abnormal monitoring scales of the K nodes are determined, including: Traversing the K historical node abnormal monitoring data sets to identify neighbor data intervals, and obtaining K abnormal neighbor data interval sets; Calculate the mean of the K abnormal neighbor data interval sets respectively to obtain the mean of the K abnormal neighbor data intervals; Calculate the fluctuation variances of the K abnormal neighbor data interval sets respectively, and determine K concentrated identification step sizes according to the fluctuation variances; Taking the mean values of the K abnormal neighbor data intervals as K identification starting points, performing neighbor data interval centralized identification on the K abnormal neighbor data interval sets according to the K centralized identification step sizes, and obtaining the K node abnormality monitoring scales; include: Iterate the K identification starting points in the K abnormal neighbor data interval sets in any direction according to the K concentrated identification step lengths to obtain K first iteration abnormal neighbor data intervals, and respectively use the directions from the K identification starting points to the K first iteration abnormal neighbor data intervals as K first iteration directions; Determine whether the concentrated identification density of the K first iteration abnormal neighbor data intervals is greater than or equal to the concentrated identification density of the K identification starting points, and if so, update the K first iteration abnormal neighbor data intervals to K identification starting points, and continue iterating; If not, the K first iteration directions are added to the K taboo iteration direction spaces, and then the K identification starting points are iterated in any direction except the iteration directions stored in the K taboo iteration direction spaces in the K abnormal neighbor data interval sets according to the K concentrated identification step sizes again until the preset number of iterations is met, and K target abnormal neighbor data intervals are obtained; The K target abnormal neighbor data intervals are used as K node abnormality monitoring scales; The K taboo iteration direction spaces include K preset taboo iteration numbers, wherein the iteration directions added into the K taboo iteration direction spaces are forbidden to be within the K preset taboo iteration numbers in the iteration.
2. The geothermal development risk monitoring method combined with multidimensional data analysis according to claim 1, characterized in that: Traversing the global perception monitoring topology data set sequence to perform feature recognition within the sequence to obtain a global perception monitoring topology feature set sequence, including: Acquire multiple sample global perception monitoring topology data and multiple sample global perception monitoring topology features as training sample data; Dividing the training sample data into a training set and a validation set according to a preset ratio; The framework based on the feedforward neural network is trained using the training set, and the training process is supervised using the validation set to learn the mapping relationship between the global perception monitoring topology data and the global perception monitoring topology features until the training converges to obtain a trained feature recognition network layer; The feature recognition network layer is used to perform intra-sequence feature recognition on the global perception monitoring topology data set sequence to obtain the global perception monitoring topology feature set sequence.
3. The geothermal development risk monitoring method combined with multidimensional data analysis according to claim 1, characterized in that: Performing neighbor matrix interaction analysis on the global perception monitoring topology feature set sequence to obtain a neighbor interaction perception monitoring feature set sequence, including: Extracting a first global perception monitoring topology feature set at the first position and a second global perception monitoring topology feature set at the second position in the global perception monitoring topology feature set sequence; Performing neighbor matrix interaction analysis on the first global perception monitoring topology feature set and the second global perception monitoring topology feature set to obtain a first neighbor interaction perception monitoring feature set; extracting again a third global perception monitoring topology feature set located at the third position in the global perception monitoring topology feature set sequence; Performing neighbor matrix interaction analysis according to the second global perception monitoring topology feature set and the third global perception monitoring topology feature set to obtain a second neighbor interaction perception monitoring feature set; Perform multiple neighbor matrix interaction analyses in sequence to obtain the Kth neighbor interaction perception monitoring feature set; The neighbor interaction perception monitoring feature set sequence is obtained according to the first neighbor interaction perception monitoring feature set, the second neighbor interaction perception monitoring feature set and the Kth neighbor interaction perception monitoring feature set.
4. The geothermal development risk monitoring method combined with multidimensional data analysis according to claim 3, characterized in that: include: Perform an inner product calculation and a softmax calculation on the first global perception monitoring topology feature set and the second global perception monitoring topology feature set to obtain a first neighbor interaction matrix; A convolution operation is performed on the first neighbor interaction matrix and the first global perception monitoring topology feature set to obtain a first neighbor interaction perception monitoring feature set.
5. A geothermal development risk monitoring device combined with multi-dimensional data analysis, characterized in that: The device is used to execute the geothermal development risk monitoring method combined with multidimensional data analysis according to any one of claims 1 to 4, and the device comprises: A sensing topology layout module, wherein the sensing topology layout module is used to layout an Internet of Things sensing topology network in a target geothermal development area, wherein the Internet of Things sensing topology network includes K topological nodes, each topological node is equipped with at least one type of monitoring sensor, wherein K is an integer greater than or equal to 1; A historical data collection module, wherein the historical data collection module is used to collect historical abnormality monitoring data of the K topological nodes to obtain K historical node abnormality monitoring data sets, wherein each historical node abnormality monitoring data corresponds to a monitoring time node; A neighbor data identification module, wherein the neighbor data identification module is used to combine the monitoring time node, traverse the K historical node abnormal monitoring data sets to perform centralized identification of neighbor data intervals, and determine the abnormal monitoring scales of the K nodes; An adaptive convolution extraction module, wherein the adaptive convolution extraction module is used to construct K adaptive convolution network layers based on the K node anomaly monitoring scales, and to serialize the K adaptive convolution network layers in order from small to large according to the node anomaly monitoring scales, to obtain an adaptive convolution network layer sequence, and to use each adaptive convolution network layer to extract the monitoring data of the K topological nodes within a preset monitoring period, to obtain a global perception monitoring topology data set sequence; A feature recognition module, the feature recognition module is used to traverse the global perception monitoring topology data set sequence to perform feature recognition within the sequence, and obtain a global perception monitoring topology feature set sequence; A neighbor analysis module, the neighbor analysis module is used to perform neighbor matrix interaction analysis on the global perception monitoring topology feature set sequence to obtain a neighbor interaction perception monitoring feature set sequence; The risk monitoring and analysis module is used to call the risk monitoring network to perform cross-scale feature fusion analysis on the neighbor interaction perception monitoring feature set sequence to obtain risk monitoring results.
Citation Information
Patent Citations
Reliability analysis method for improving monitoring of internet of things on coal mine major hazard
CN103326909A
Power distribution network topology identification method and system based on hierarchical clustering method
CN110718908A