Automatic monitoring method of radioactive tailings pond based on multi-source heterogeneous data fusion

Abstract

The invention discloses an automatic monitoring method for a radioactive tailings pond based on multi-source heterogeneous data fusion, which belongs to the field of data monitoring technology, and includes: setting monitoring equipment for a radioactive tailings pond, collecting monitoring data of the radioactive tailings pond, and transmitting all monitoring data to a monitoring platform based on a transmission mode; the monitoring platform constructs a fusion model, and the fusion model fuses all monitoring data based on data type and time characteristics to generate a heterogeneous data combination; constructs a learning model, inputs the heterogeneous data combination into the learning model for learning and training, generates a prediction and early warning model, and extracts state characteristics corresponding to the radioactive tailings pond; outputs a state prediction result of the radioactive tailings pond within a prediction time based on the state characteristics, obtains the accuracy of the prediction and early warning model, and completes the early warning monitoring of the radioactive tailings pond based on the state prediction result, the accuracy and the preset safety threshold. The present invention can improve the timeliness and accuracy of monitoring of radioactive tailings ponds.

Description

Technical Field

[0001] The invention belongs to the technical field of data monitoring, and in particular relates to an automated monitoring method for a radioactive tailings pond based on multi-source heterogeneous data fusion. Background Art

[0002] As an important part of mining activities, tailings ponds undertake the important task of storing residual materials after mineral processing, especially in the treatment of radioactive materials, and their safety supervision is particularly important. Existing tailings pond monitoring methods mostly rely on a single data source and cannot fully reflect the safety status of tailings ponds. Due to their particularity, radioactive tailings ponds require more accurate and comprehensive monitoring methods to achieve automated safety supervision.

[0003] Similar prior art includes a Chinese patent application with publication number CN112883098A, which discloses a multi-source heterogeneous system data fusion method, including: obtaining the monitoring data of the enterprise tailings pond, and transmitting the monitoring data to the IoT host through the TCP protocol / HTTP protocol; publishing API services; calling API services, and determining the enterprise reported data in the monitoring data; using the operator's dedicated network to report the enterprise reported data through the HTTP protocol; receiving the enterprise reported data in real time to obtain the enterprise-side data; verifying the enterprise-side data, and storing the verified enterprise-side data in the database. This invention adopts the method of uniformly publishing API interfaces, and the enterprise terminal reports data in accordance with the data specification format by calling the interface, which solves the problem of the existing technical solutions being unsuitable for enterprises with backward conditions and high investment costs. There is also a Chinese patent application with publication number CN116579607A, which provides a regional tailings pond risk monitoring method with multi-source data collaboration, including: obtaining basic data of tailings ponds, building a tailings pond basic data system based on the basic data; establishing a risk warning system, using the risk warning system to conduct risk warning for tailings ponds; establishing an online monitoring system, and conducting online monitoring for tailings ponds through the online monitoring system; coordinating the tailings pond basic data system, risk warning system and online monitoring system to conduct regional risk analysis of tailings ponds based on the tailings pond basic data system, tailings pond warning information and tailings pond monitoring information, determine the risk status of regional tailings ponds, and obtain a comprehensive analysis report of regional tailings ponds. This invention not only realizes the monitoring of tailings pond risks from a regional perspective, improves the accuracy of tailings pond risk monitoring, but also reduces the time to obtain information when conducting comprehensive analysis of tailings pond risk monitoring, thereby improving the efficiency of comprehensive analysis of tailings pond risk monitoring.

[0004] However, in the above-mentioned prior art, the data from multiple sources of heterogeneous data are reported in a standardized format, which reduces the integrity and effectiveness of the data, and only conducts comprehensive risk monitoring analysis on regional tailings ponds, which reduces the timeliness and accuracy of tailings pond monitoring. In actual situations, it is necessary to accurately integrate multiple sources of heterogeneous data and conduct automated and accurate monitoring of tailings ponds through data prediction and analysis. Summary of the invention

[0005] In order to solve the above problems, the present invention provides an automated monitoring method for radioactive tailings ponds based on multi-source heterogeneous data fusion to solve the problems in the prior art.

[0006] In order to achieve the above-mentioned invention object, the present invention proposes an automatic monitoring method for radioactive tailings pond based on multi-source heterogeneous data fusion, comprising:

[0007] S1: Setting a monitoring device for a radioactive tailings pond, wherein the monitoring device collects monitoring data of the radioactive tailings pond based on a preset monitoring mode, sets a transmission mode, and transmits all the monitoring data to a monitoring platform based on the transmission mode;

[0008] S2: The monitoring platform obtains the data type and time characteristics of the monitoring data, and constructs a fusion model, wherein the fusion model fuses all the monitoring data based on the data type and the time characteristics to generate a heterogeneous data combination;

[0009] S3: constructing a learning model, inputting the heterogeneous data combination into the learning model for learning and training, and generating a prediction and early warning model, wherein the prediction and early warning model extracts state features corresponding to the radioactive tailings pond from the heterogeneous data combination based on the time features;

[0010] S4: Outputting the state prediction result of the radioactive tailings pond within the prediction time based on the state characteristics, obtaining the accuracy of the prediction and early warning model, and completing the early warning monitoring of the radioactive tailings pond based on the state prediction result, the accuracy and the preset safety threshold.

[0011] Further, transmitting all the monitoring data to the monitoring platform based on the transmission mode comprises the following steps:

[0012] Based on the monitoring mode, the collection intervals corresponding to all the monitoring devices are obtained respectively, the least common multiple values ​​of all the collection intervals are obtained, the multiples of any collection interval are calculated based on the least common multiple values, the multiples are set to the number of transmission nodes of the monitoring device in the transmission mode corresponding to any of the collection intervals, the corresponding transmission nodes are selected based on the number of transmission nodes to connect to generate a transmission path for the monitoring device, and the monitoring data corresponding to the monitoring device is transmitted to the monitoring platform based on the transmission path.

[0013] Furthermore, the generating of heterogeneous data combination comprises the following steps:

[0014] The monitoring platform stores the monitoring data transmitted using the same transmission path in the same database and sets it as a sample database;

[0015] Setting a data extraction function, wherein the fusion model extracts the corresponding monitoring data from all sample databases based on the data extraction function and sets the data as retrieval data;

[0016] A standard format is set, a table file is constructed based on the standard format, all the search data are adjusted based on the standard format, and are sequentially written into the table file based on the time feature, and the data information contained in the table file is set as the heterogeneous data combination.

[0017] Furthermore, adjusting all the retrieved data based on the standard format comprises the following steps:

[0018] Based on the data type, all the sample databases are divided into an image sample database and a non-image sample database. If the sample database is the image sample database, the search data is converted into the corresponding standard format based on an image processing method.

[0019] If the sample database is the non-image sample library, the information type of the search data is obtained, a data tag is constructed based on the information type, a style control parameter is set by the data tag, and the search data is converted into the corresponding standard format based on the style control parameter.

[0020] Furthermore, the generation of the prediction and early warning model comprises the following steps:

[0021] Classifying the heterogeneous data combination into first type data and second type data based on the corresponding data types, grouping and arranging the first type data and the second type data based on a preset arrangement method to generate a plurality of corresponding first data blocks and second data blocks, and combining all the first data blocks and all the second data blocks to generate a first data combination;

[0022] The learning model is a restricted Boltzmann machine network model, which includes a visible layer and a hidden layer. The first data combination is stored in the visible layer, and the output value corresponding to the hidden layer is obtained. All the output values ​​are learned based on the contrastive divergence algorithm and a second data combination is generated. The sum of square errors between the first data combination and the second data combination are calculated and set as the first error of the first data combination. The parameters of the learning model are adjusted based on the first error until the first error corresponding to the first data combination is minimized, and the learning model after the parameter adjustment is set as the prediction and warning model.

[0023] Further, adjusting the parameters of the learning model based on the first error comprises the following steps:

[0024] If the first error is greater than a first threshold, obtaining a feature correlation between the first data block and any second data block in the first data combination, adjusting the second data block with the largest feature correlation to a position adjacent to the first data block, repeating this step until all the first data blocks and all the second data blocks are rearranged and combined, generating a new first data combination, and setting it as the updated data combination;

[0025] Calculate the feature change amount between the updated data combination and the first data combination, set the numerical value corresponding to the feature change amount as the change ratio corresponding to the parameter of the learning model, and adjust the parameter of the learning model based on the change ratio.

[0026] Furthermore, the prediction and early warning model outputs the state prediction result of the radioactive tailings pond within the prediction time based on the state characteristics, including the following steps:

[0027] The state characteristics are set with feature labels based on the type of the monitoring equipment, all the state characteristics are clustered to generate multiple clustering results, the feature label corresponding to the cluster center in any clustering result is obtained, and set as the prediction label, the prediction and early warning model outputs the prediction value corresponding to the prediction label within the prediction time based on all the state characteristics contained in the clustering result, all the prediction values ​​are combined in sequence, a prediction value combination corresponding to all the prediction labels is generated, and set as the state prediction result.

[0028] Furthermore, the early warning monitoring of the radioactive tailings pond based on the state prediction result, the accuracy rate and the preset safety threshold comprises the following steps:

[0029] Extracting a validation set from the heterogeneous data combination based on the time feature, acquiring the accuracy of the prediction and early warning model based on the validation set, and generating a state fluctuation range of the state prediction result based on the accuracy;

[0030] Obtain the safety threshold of the monitoring equipment corresponding to any prediction tag, combine all the safety thresholds to generate a warning status data group for the radioactive tailings pond, if the status fluctuation range contains any value in the warning status data group, obtain the corresponding prediction time point, generate safety warning information, send the safety warning information and the prediction time point to the management terminal, and complete the warning monitoring of the radioactive tailings pond.

[0031] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0032] The present invention first collects various monitoring data of the radioactive tailings pond through a transmission mode, which can improve the timeliness and accuracy of the monitoring data, wherein the monitoring data corresponding to different monitoring devices use different transmission paths, which can reduce the interference of the data transmission process. Then, by constructing a fusion model, the monitoring data corresponding to different data types are fused to generate a heterogeneous data combination, which can improve the integrity and analyzability of the monitoring data. Finally, by constructing a learning model, the heterogeneous data combination is trained and learned to generate a prediction and early warning model, which can improve the learning rate of the learning model and quickly generate a prediction and early warning model.

[0033] The present invention also predicts the radioactive tailings pond through the prediction and early warning model to output the state prediction result, and completes the early warning monitoring of the radioactive tailings pond in combination with the accuracy of the model and the preset safety threshold, which can improve the timeliness and accuracy of the automated monitoring of the radioactive tailings pond. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 The present invention is a flowchart of the steps of an automated monitoring method for radioactive tailings ponds based on multi-source heterogeneous data fusion. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0036] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, without departing from the scope of this application, a first xx script may be referred to as a second xx script, and similarly, a second xx script may be referred to as a first xx script.

[0037] like Figure 1 As shown, an automated monitoring method for radioactive tailings ponds based on multi-source heterogeneous data fusion includes:

[0038] S1: Setting up monitoring equipment for the radioactive tailings pond, the monitoring equipment collects monitoring data of the radioactive tailings pond based on a preset monitoring mode, setting a transmission mode, and transmitting all monitoring data to the monitoring platform based on the transmission mode.

[0039] Specifically, multi-source heterogeneous data refers to a data set from different data sources with different characteristics, formats or structures. The present invention performs automated safety monitoring and early warning of radioactive tailings ponds by fusing multi-source heterogeneous data. In this embodiment, the radioactive tailings pond refers to a facility library for storing and managing tailings generated from the mining and beneficiation of radioactive minerals. The monitoring equipment refers to sensor equipment for all-weather, automated monitoring of the radioactive tailings pond, including but not limited to GNSS monitoring, deep displacement monitoring, groundwater level monitoring, infiltration line monitoring, radon exhalation rate monitoring, ground gamma radiation monitoring and other corresponding equipment types. The monitoring mode refers to the method of using the monitoring equipment, including information such as the installation location, sampling monitoring frequency and monitoring type. The monitoring data is the data information generated by the monitoring equipment through the monitoring mode, for example, the displacement data collected by the GNSS monitoring equipment, and the gamma ray radiation image data measured by the ground gamma radiation monitoring equipment. The transmission mode refers to the transmission mode of each monitoring data, and the monitoring platform refers to a terminal for collecting various monitoring data and performing multi-source heterogeneous data fusion processing. The transmission mode can quickly transmit and obtain monitoring data.

[0040] S2: The monitoring platform obtains the data type and time characteristics of the monitoring data and builds a fusion model. The fusion model fuses all monitoring data based on the data type and time characteristics to generate a heterogeneous data combination.

[0041] Specifically, in this embodiment, the data type refers to the type of monitoring data, including image type and non-image type, and the time feature refers to the timestamp information corresponding to the collected monitoring data. The fusion model refers to a model for fusing monitoring data corresponding to multiple different data types, that is, fusing all monitoring data to generate a heterogeneous data combination. The heterogeneous data combination generated by the fusion model can improve the closeness between various monitoring data, which is convenient for subsequent comprehensive analysis of the monitoring data of the radioactive tailings pond.

[0042] S3: Build a learning model, input the heterogeneous data combination into the learning model for learning and training, and generate a prediction and early warning model. The prediction and early warning model extracts the status characteristics corresponding to the radioactive tailings pond from the heterogeneous data combination based on time characteristics.

[0043] Specifically, in this embodiment, the learning model refers to a neural network model, and the learning model can be trained and learned using a heterogeneous data combination, and a prediction and early warning model can be generated by adjusting the parameters of the learning model, wherein the prediction and early warning model can use the change trend of the heterogeneous data combination composed of a variety of collected monitoring data to monitor the environment of the radioactive tailings pond in real time, evaluate its safety and stability, and output possible risk values ​​through prediction results. State features refer to a combination of data features that describe the state changes of the radioactive tailings pond through the data features contained in the heterogeneous data combination, wherein the data features contained in the heterogeneous data combination are feature representations generated by the prediction and early warning model through feature learning.

[0044] S4: Output the status prediction results of the radioactive tailings pond within the prediction time based on the status characteristics, obtain the accuracy of the prediction and early warning model, and complete the early warning monitoring of the radioactive tailings pond based on the status prediction results, accuracy and preset safety thresholds.

[0045] Specifically, in this embodiment, the neural network in the prediction and early warning model outputs the corresponding state prediction result after predicting and analyzing the changing trend of the state characteristics, wherein the state prediction result is composed of the prediction values ​​corresponding to multiple groups of monitoring equipment, the accuracy rate refers to the accuracy assessment of the current prediction and early warning model, and the preset safety threshold refers to the safety warning value corresponding to various monitoring equipment. The combination of state prediction results, accuracy rate and safety threshold can improve the accuracy and timeliness of early warning monitoring of radioactive tailings ponds.

[0046] As a preferred technical solution of the present invention, transmitting all monitoring data to the monitoring platform based on the transmission mode includes the following steps:

[0047] Based on the monitoring mode, the collection intervals corresponding to all monitoring devices are obtained respectively, the lowest common multiple value of all collection intervals is obtained, the multiples of any collection interval are calculated based on the lowest common multiple value, the multiples are set to the number of transmission nodes of the monitoring device corresponding to any collection interval in the transmission mode, the corresponding transmission node is selected based on the number of transmission nodes to connect to generate a transmission path for the monitoring device, and the monitoring data corresponding to the monitoring device is transmitted to the monitoring platform based on the transmission path.

[0048] Specifically, in this embodiment, the collection interval refers to the sampling monitoring frequency included in the monitoring mode, for example, 10 times / minute, 1 time / hour. The least common multiple value refers to the smallest positive integer that can be divided by several given integers. The transmission mode is a transmission path formed by various transmission nodes through different connection methods. For example, the collection interval of monitoring device A1 is 10 (times / minute), and the collection interval of monitoring device A2 is 1 (times / hour). The least common multiple value corresponding to the collection interval 10 (times / minute) and 1 (times / hour) is 1 hour. Then, the multiple corresponding to the collection interval 10 (times / minute) is 6, and the multiple corresponding to the collection interval 1 (times / hour) is 1. The number of transmission nodes of monitoring device A1 is 6, and the number of transmission nodes of monitoring device A2 is 1. Then, the number of transmission nodes is connected in sequence to form a transmission path, wherein the monitoring device with a shorter collection interval is transmitted through the transmission path of more transmission nodes, which can avoid the phenomenon of channel congestion and collision when multiple monitoring devices perform data transmission. Therefore, the transmission mode can reduce the phenomenon of reduced data timeliness caused by transmission delays between monitoring devices with different collection intervals. At the same time, it can reduce transmission interference of the monitoring data corresponding to all monitoring devices and realize fast and stable transmission of data to the monitoring platform.

[0049] As a preferred technical solution of the present invention, generating a heterogeneous data combination includes the following steps:

[0050] The monitoring platform stores the monitoring data transmitted using the same transmission path in the same database and sets it as a sample database.

[0051] Set the data extraction function. The fusion model extracts the corresponding monitoring data from all sample databases based on the data extraction function and sets it as the retrieval data.

[0052] Set a standard format, build a table file based on the standard format, adjust all search data based on the standard format, write them into the table file in sequence based on time characteristics, and set the data information contained in the table file as a heterogeneous data combination.

[0053] Specifically, in this embodiment, monitoring data transmitted through the same transmission path are aggregated into the same database, which can avoid data loss, so that the monitoring data corresponding to the same monitoring device is completely stored in the database, and each database is set as a sample database. The data extraction function refers to a database command function for accessing a sample database, such as SQL syntax, and the abstract syntax tree can be used to perform a syntax check on the data extraction function to see if there are any semantic errors. Retrieval data refers to the monitoring data retrieved from the sample database through the data extraction function, which not only retains the original monitoring data, but also can quickly extract the monitoring data.

[0054] The standard format refers to a preset data format, through which all monitoring data can be adjusted to the same preset data format. The table file is a network system used to create, edit, calculate, analyze and display table data. It can generate corresponding table files after integrating and summarizing multi-source heterogeneous data, which is convenient for subsequent comprehensive analysis of multi-source heterogeneous data. Therefore, the data information contained in the table file is set as a heterogeneous data combination.

[0055] As a preferred technical solution of the present invention, adjusting all search data based on the standard format includes the following steps:

[0056] All sample databases are divided into image sample databases and non-image sample databases based on data types. If the sample database is an image sample database, the retrieval data is converted into a corresponding standard format based on the image processing method.

[0057] If the sample database is a non-image sample database, the information type of the retrieved data is obtained, a data label is constructed based on the information type, a style control parameter is set by the data label, and the retrieved data is converted into a corresponding standard format based on the style control parameter.

[0058] Specifically, in this embodiment, the data types include image types and non-image types, wherein the non-image types include but are not limited to digital data, character data, and the like. Therefore, the sample database can be classified into an image sample library and a non-image sample library. The image processing method refers to algorithms such as image conversion and image processing, which adjust all monitoring data to the corresponding standard format after image processing by setting the size, format, and other information in the standard format.

[0059] Information type refers to information corresponding to non-image data types, such as digital type, character type, etc. Data label refers to label information corresponding to information type, such as digital label, character label, etc. Style control parameter refers to preset style information corresponding to each data label, such as font, border, color, etc. The style control parameter adjusts the retrieval data corresponding to each data label to a standard format. The heterogeneous data combination generated by the standard format can improve the efficiency of calculation, reading, capturing, deleting, comparing, exchanging and searching, and provide diversity in data processing.

[0060] As a preferred technical solution of the present invention, generating a prediction and early warning model includes the following steps:

[0061] The heterogeneous data combination is classified into first type data and second type data based on corresponding data types, and the first type data and the second type data are grouped and arranged based on a preset arrangement method to generate a plurality of corresponding first data blocks and second data blocks, and all the first data blocks and all the second data blocks are combined to generate a first data combination.

[0062] The learning model is a restricted Boltzmann machine network model, which includes a visible layer and a hidden layer. The first data combination is stored in the visible layer, and the output value corresponding to the hidden layer is obtained. All output values ​​are learned based on the contrastive divergence algorithm and a second data combination is generated. The sum of square errors between the first data combination and the second data combination are calculated and set as the first error of the first data combination. The parameters of the learning model are adjusted based on the first error until the first error corresponding to the first data combination is minimized, and the learning model after parameter adjustment is set as a prediction and early warning model.

[0063] Specifically, in this embodiment, the first type of data refers to data of image type in the heterogeneous data combination, and the second type of data refers to data of non-image type in the heterogeneous data combination. The preset arrangement method refers to arranging in sequence by corresponding time features. In the heterogeneous data combination, the first type of data and the second type of data are arranged in sequence by standard format and time features, but the first type of data and the second type of data will form adjacent blocks with different arrangements in the table file corresponding to the heterogeneous data combination. Therefore, the blocks composed of adjacent first type data are set as first data blocks, and the blocks composed of adjacent second type data are set as second data blocks. The first data combination is generated by combining the blocks, wherein the first data combination is the initial heterogeneous data combination.

[0064] The restricted Boltzmann machine network model corresponding to the learning model can realize the training and learning of the model through deep learning, and accurately generate the corresponding prediction and warning model, wherein the restricted Boltzmann machine (RBM) network model belongs to the unsupervised learning algorithm. The visible layer and the hidden layer are the input layer and the feature layer composed of visible neurons and hidden neurons respectively, and the neurons between the two layers are fully connected. After the first data combination is input into the visible layer, the full connection between the neurons enables the hidden layer to output the corresponding output value through the probability distribution learning method, wherein the output value is 0 or 1. The probability distribution learning method refers to using the sigmoid function to calculate the activation probability of each hidden neuron and determine the state of each hidden neuron. The probability above the threshold corresponds to the output value of the hidden neuron being 1 (activated), and the probability below the threshold is 0 (not activated).

[0065] The contrastive divergence algorithm is used to learn all output values ​​to generate a second data combination. The error, i.e., the first error, can be calculated by comparing the first input data combination and the second data combination of the reconstructed data. Then, the parameters of the neural network model are updated according to the first error. The parameters include weights and biases. The weights refer to the connection strength between the visible layer neurons and the hidden layer neurons. The bias refers to the bias vector corresponding to each hidden neuron and the visible neuron. The contrastive divergence algorithm is a fast algorithm for restricted Boltzmann machines, which can improve the training rate of restricted Boltzmann machine network models. The learning model corresponding to the minimum first error belongs to a model adjusted after training and learning, and has a more accurate data prediction and analysis function. Therefore, the model is set as a prediction and early warning model.

[0066] As a preferred technical solution of the present invention, adjusting the parameters of the learning model based on the first error includes the following steps:

[0067] If the first error is greater than the first threshold, the feature correlation between the first data block and any second data block in the first data combination is obtained, and the second data block with the largest feature correlation is adjusted to a position adjacent to the first data block. This step is repeated until all first data blocks and all second data blocks are rearranged and combined to generate a new first data combination, which is set as the updated data combination.

[0068] The feature change amount between the updated data combination and the first data combination is calculated, the numerical value corresponding to the feature change amount is set as the change ratio corresponding to the parameters of the learning model, and the parameters of the learning model are adjusted based on the change ratio.

[0069] Specifically, in this embodiment, if the first error is greater than the first threshold, it means that the training effect of the learning model is poor. Therefore, it is necessary to rearrange the arrangement positions of each first data block and the second data block in the first data combination, so that the reading order of each block in the first data combination changes, and each first data block and the second data block are evenly distributed, which can shorten the training and learning time of the learning model and improve the timeliness of the learning model. Feature correlation refers to the correlation between the feature engineering of the data. The feature engineering of the first data block and any second data block can be obtained by the principal component analysis method, and the correlation value corresponding to the linear correlation degree between any two feature engineerings can be measured by the Pearson correlation coefficient algorithm, wherein the feature engineering refers to the feature vector corresponding to the data. Adjacent position means that the arrangement positions of each block in the heterogeneous data combination are adjacent, including but not limited to vertically adjacent, horizontally adjacent, etc.

[0070] By repeating this step, the arrangement positions of all blocks in the first data combination can be adjusted to form an updated data combination, so that blocks with high feature correlation are arranged adjacently, thereby improving the rapidity of training and learning of the learning model. At the same time, the changing trend of the first error can be quickly monitored.

[0071] The feature change refers to the change in the correlation between the feature engineering corresponding to the first data combination and the correlation between the feature engineering corresponding to the updated data combination calculated by the principal component analysis method and the Pearson correlation coefficient algorithm. For example, the maximum feature correlation between the feature engineering generated by the first data combination is F1, and the maximum feature correlation between the feature engineering generated by the updated data combination is F2. Then the feature change is: (|F2-F1|) / F1, where the feature engineering uses the principal component analysis method for automatic learning feature acquisition. The change ratio of the parameter is set by the feature change. For example, if the feature change is 0.8, the parameters of the learning model are adjusted by increasing or decreasing the change ratio of 0.8. It can quickly provide a correlation basis for the adjustment of the learning model.

[0072] As a preferred technical solution of the present invention, the prediction and early warning model outputs the state prediction result of the radioactive tailings pond within the prediction time based on the state characteristics, including the following steps:

[0073] The state features are set with feature labels based on the type of monitoring equipment, all state features are clustered to generate multiple clustering results, the feature label corresponding to the cluster center in any clustering result is obtained and set as the prediction label, the prediction and early warning model outputs the prediction value corresponding to the prediction label within the prediction time based on all state features contained in the clustering result, all prediction values ​​are combined in turn, the prediction value combination corresponding to all prediction labels is generated, and set as the state prediction result.

[0074] Specifically, in this embodiment, the state feature is a data feature generated by the prediction and early warning model through the type of data, the dimension of the data, the complexity of the data and the purpose of the analysis in the heterogeneous data combination, so the number of state features is multiple. The data information of the same monitoring device in the heterogeneous data combination is correlated, so feature labels can be set for each state feature according to the type of different monitoring devices. All state features are clustered by the K-Means clustering method to generate a clustering result of a specified number of clusters, where the number of clusters is consistent with the number of monitoring devices, and the clustering result is composed of multiple state features. The cluster center refers to the state feature located at the center of the clustering result, and the feature label of the state feature is set as the prediction label. The prediction and early warning model can analyze and predict all the state features contained in the clustering result, and output the prediction value corresponding to the prediction label. By combining the prediction values ​​corresponding to all prediction labels, the state prediction result of the radioactive tailings pond can be generated.

[0075] As a preferred technical solution of the present invention, the early warning monitoring of the radioactive tailings pond based on the state prediction result, accuracy and preset safety threshold includes the following steps:

[0076] A validation set is extracted from the heterogeneous data combination based on the time characteristics, the accuracy of the prediction and early warning model is obtained based on the validation set, and the state fluctuation range of the state prediction result is generated based on the accuracy.

[0077] Obtain the safety threshold of the monitoring equipment corresponding to any prediction tag, combine all safety thresholds to generate a warning status data group for the radioactive tailings pond, and if the status fluctuation range contains any value in the warning status data group, obtain the corresponding prediction time point, generate safety warning information, and send the safety warning information and prediction time point to the management terminal to complete the warning monitoring of the radioactive tailings pond.

[0078] Specifically, in this embodiment, the heterogeneous data combination is used as training data. Through the time features corresponding to the data information contained in the heterogeneous data combination, a certain proportion of data information can be extracted from the heterogeneous data combination and set as a validation set, wherein the validation set is a part of the data used to evaluate the accuracy of the model and prevent the model from overfitting. The accuracy rate is a numerical value used to measure the accuracy of the prediction and early warning model, and the state fluctuation range refers to the floating change range corresponding to the state prediction result. For example, if the accuracy rate is 0.8 and the state prediction result is (10, 5, 3), then the state fluctuation range is {(10±10×0.8), (5±5×0.8), (3±3×0.8)}.

[0079] The safety thresholds of the monitoring devices corresponding to all the prediction labels are combined to generate a warning status data group, wherein the monitoring devices corresponding to each value in the warning status data group are the same as the monitoring devices corresponding to each value in the state fluctuation range. For example, the warning status data group is (5, 10, 6), wherein the first value 5 in the warning status data group belongs to the first value (10±10×0.8) in the state fluctuation range, which means that the monitoring device corresponding to the first value has the possibility of being at the safety threshold, and the corresponding monitoring device and the prediction time point are combined into safety warning information. Among them, the prediction time point refers to the time point corresponding to the prediction time. The management terminal refers to a mobile terminal used to monitor and manage the radioactive tailings pond in a timely manner. Through this step, the automatic monitoring of the radioactive tailings pond can be accurately and quickly realized, thereby improving the safety of the radioactive tailings pond.

[0080] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0081] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the above-mentioned program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0082] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0083] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for those of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

[0084] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A radioactive tailings pond automated monitoring method based on multi-source heterogeneous data fusion, characterized in that: The method comprises the following steps: S1: Setting monitoring equipment for the radioactive tailings pond, wherein the monitoring equipment collects monitoring data of the radioactive tailings pond based on a preset monitoring mode, sets a transmission mode, and transmits all the monitoring data to a monitoring platform based on the transmission mode, wherein the monitoring equipment refers to sensor equipment for all-weather, automated monitoring of the radioactive tailings pond, including equipment types corresponding to GNSS monitoring, deep displacement monitoring, groundwater level monitoring, infiltration line monitoring, radon exhalation rate monitoring, and ground gamma radiation monitoring; S2: The monitoring platform obtains the data type and time characteristics of the monitoring data, and constructs a fusion model, wherein the fusion model fuses all the monitoring data based on the data type and the time characteristics to generate a heterogeneous data combination; S3: constructing a learning model, inputting the heterogeneous data combination into the learning model for learning and training, and generating a prediction and early warning model, wherein the prediction and early warning model extracts state features corresponding to the radioactive tailings pond from the heterogeneous data combination based on the time features; S4: outputting a state prediction result of the radioactive tailings pond within a prediction time based on the state characteristics, obtaining the accuracy of the prediction and early warning model, and completing early warning monitoring of the radioactive tailings pond based on the state prediction result, the accuracy and a preset safety threshold; The generation of the prediction and early warning model includes the following steps: classifying the heterogeneous data combination into first type data and second type data based on the corresponding data type, grouping and arranging the first type data and the second type data based on a preset arrangement method to generate a plurality of corresponding first data blocks and second data blocks, combining all first data blocks and all second data blocks to generate a first data combination; the learning model is a restricted Boltzmann machine network model, the restricted Boltzmann machine network model includes a visible layer and a hidden layer, storing the first data combination in the visible layer, and obtaining the output value corresponding to the hidden layer, learning all the output values ​​based on the contrastive divergence algorithm and generating a second data combination, calculating the sum of squares error between the first data combination and the second data combination, and setting it as the first error of the first data combination, adjusting the parameters of the learning model based on the first error until the first error corresponding to the first data combination is minimized, and setting the learning model after parameter adjustment as the prediction and early warning model; The method of adjusting the parameters of the learning model based on the first error includes the following steps: if the first error is greater than a first threshold, obtaining the feature correlation between the first data block and any second data block in the first data combination, adjusting the second data block with the largest feature correlation to an adjacent position of the first data block, repeating this step until all the first data blocks and all the second data blocks are rearranged and combined to generate a new first data combination, and setting it as an updated data combination; calculating the feature change amount between the updated data combination and the first data combination, setting the numerical value corresponding to the feature change amount as the change ratio corresponding to the parameters of the learning model, and adjusting the parameters of the learning model based on the change ratio; wherein, the feature engineering of the first data block and any second data block is respectively obtained by the principal component analysis method, and then the correlation numerical value corresponding to the linear correlation degree between any two feature engineerings is measured by the Pearson correlation coefficient algorithm to obtain the feature correlation; the adjacent positions include longitudinal adjacent and transverse adjacent; The prediction and early warning model outputs the state prediction result of the radioactive tailings pond within the prediction time based on the state characteristics, including the following steps: setting feature labels for the state characteristics based on the type of the monitoring equipment, clustering all the state characteristics to generate multiple clustering results, obtaining the feature label corresponding to the cluster center in any clustering result, and setting it as the prediction label, the prediction and early warning model outputs the prediction value corresponding to the prediction label within the prediction time based on all the state characteristics contained in the clustering result, combining all the prediction values ​​in sequence, generating a prediction value combination corresponding to all the prediction labels, and setting it as the state prediction result; Wherein, completing the early warning monitoring of the radioactive tailings pond based on the state prediction result, the accuracy rate and the preset safety threshold comprises the following steps: Extracting a validation set from the heterogeneous data combination based on the time feature, acquiring the accuracy of the prediction and early warning model based on the validation set, and generating a state fluctuation range of the state prediction result based on the accuracy; Obtain the safety threshold of the monitoring equipment corresponding to any prediction tag, combine all the safety thresholds to generate a warning status data group for the radioactive tailings pond, if the status fluctuation range contains any value in the warning status data group, obtain the corresponding prediction time point, generate safety warning information, send the safety warning information and the prediction time point to the management terminal, and complete the warning monitoring of the radioactive tailings pond.

2. The method according to claim 1, characterized in that The transmitting of all the monitoring data to the monitoring platform based on the transmission mode comprises the following steps: Based on the monitoring mode, the collection intervals corresponding to all the monitoring devices are obtained respectively, the least common multiple values ​​of all the collection intervals are obtained, the multiples of any collection interval are calculated based on the least common multiple values, the multiples are set to the number of transmission nodes of the monitoring device in the transmission mode corresponding to any of the collection intervals, the corresponding transmission nodes are selected based on the number of transmission nodes to connect to generate a transmission path for the monitoring device, and the monitoring data corresponding to the monitoring device is transmitted to the monitoring platform based on the transmission path.

3. The method according to claim 2, characterized in that The generating of heterogeneous data combination comprises the following steps: The monitoring platform stores the monitoring data transmitted using the same transmission path in the same database and sets it as a sample database; Setting a data extraction function, wherein the fusion model extracts the corresponding monitoring data from all sample databases based on the data extraction function and sets the data as retrieval data; A standard format is set, a table file is constructed based on the standard format, all the search data are adjusted based on the standard format, and are sequentially written into the table file based on the time feature, and the data information contained in the table file is set as the heterogeneous data combination.

4. The method according to claim 3, characterized in that: The step of adjusting all the retrieved data based on the standard format comprises the following steps: Based on the data type, all the sample databases are divided into an image sample database and a non-image sample database. If the sample database is the image sample database, the search data is converted into the corresponding standard format based on an image processing method. If the sample database is the non-image sample library, the information type of the search data is obtained, a data tag is constructed based on the information type, a style control parameter is set by the data tag, and the search data is converted into the corresponding standard format based on the style control parameter.

Images