Safety monitoring and early warning method for underground energy storage space

Abstract

The invention discloses a safety monitoring and early warning method for an underground energy storage space, and the method comprises the steps: obtaining the original monitoring data of the underground energy storage space, and carrying out the self-adaptive signal processing and correlation analysis of the original monitoring data, and obtaining the monitoring data; the original monitoring data comprises original temperature data and original pressure data; transmitting the monitoring data through a data compression and priority scheduling mechanism; storing the transmitted monitoring data, and performing fusion analysis on the monitoring data to obtain spatial distribution characteristics of temperature and pressure; judging whether an abnormal condition exists or not in real time according to the spatial distribution characteristics; performing prediction according to the stored monitoring data to obtain a predicted monitoring data change trend; and performing mining analysis on the stored monitoring data to obtain a knowledge base, and judging the monitoring data and the change trend of the monitoring data according to the knowledge base to obtain a safety monitoring early warning result.

Description

Technical Field

[0001] The present invention belongs to the technical field of energy storage safety early warning, and in particular relates to a safety monitoring and early warning method for underground energy storage spaces. Background Art

[0002] Monitoring and early warning of pressure and high temperatures in underground energy storage spaces present numerous technical challenges. First, the complex and ever-changing environment of underground energy storage spaces, with uneven temperature and pressure distribution, makes it difficult for traditional single-point monitoring methods to fully and accurately reflect the status of the entire space. Second, the complex internal structure of underground energy storage spaces, with numerous pipes, equipment, and supporting structures, can interfere with and attenuate monitoring signals, affecting the accuracy and reliability of monitoring data. Furthermore, the harsh internal environment of underground energy storage spaces, with factors such as high temperatures, high pressures, and corrosive media, can damage monitoring equipment, shortening its lifespan and increasing maintenance costs. Furthermore, accidents in underground energy storage spaces pose serious consequences. Explosions, leaks, and other incidents pose significant threats to personnel safety and the environment, placing extremely high demands on the real-time, reliable, and stable nature of monitoring and early warning systems. Finally, the transmission, storage, and analysis of massive amounts of monitoring data present significant challenges. Efficient data management and analysis methods are needed to improve the speed and accuracy of data processing, enabling timely detection of anomalies and issuing early warnings. Summary of the Invention

[0003] In order to solve the above technical problems, the present invention proposes a safety monitoring and early warning method for underground energy storage space to solve the problems existing in the above-mentioned prior art.

[0004] To achieve the above objectives, the present invention provides a safety monitoring and early warning method for underground energy storage space, comprising:

[0005] Obtaining original monitoring data of the underground energy storage space, performing adaptive signal processing and correlation analysis on the original monitoring data to obtain monitoring data; the original monitoring data includes original temperature data and original pressure data;

[0006] Transmitting the monitoring data through data compression and priority scheduling mechanism;

[0007] The transmitted monitoring data is stored and fused and analyzed to obtain the spatial distribution characteristics of temperature and pressure; and whether there is an abnormality in real time based on the spatial distribution characteristics;

[0008] Make predictions based on the stored monitoring data to obtain the predicted monitoring data change trend;

[0009] The stored monitoring data is mined and analyzed to obtain a knowledge base, and the monitoring data and the changing trend of the monitoring data are judged based on the knowledge base to obtain a safety monitoring early warning result.

[0010] Optionally, obtaining raw monitoring data of underground energy storage space includes:

[0011] A plurality of temperature sensors and pressure sensors are arranged in the underground energy storage space by a distributed multi-point monitoring method, data is collected by the temperature sensors and pressure sensors, and the collected data is preprocessed to obtain original monitoring data.

[0012] Optionally, the monitoring data acquisition process includes:

[0013] The original monitoring data is preprocessed, and the preprocessed monitoring data is amplified by an adaptive gain control algorithm to obtain an amplified signal. According to the internal structure of the underground energy storage space, a signal propagation attenuation model is constructed, and the amplified signal is corrected by the signal propagation attenuation model to obtain a corrected monitoring signal. The correlation coefficients between different corrected monitoring signals are calculated by a cross-correlation analysis algorithm to obtain a correlation matrix. The correlation matrix is subjected to threshold judgment processing to obtain a connectivity relationship graph. The corrected monitoring signal is identified according to the connectivity relationship graph to obtain an abnormal interference measurement point. The monitoring signal of the abnormal interference measurement point is processed by an interference suppression algorithm to obtain a reliable monitoring signal. The reliable monitoring signal and other monitoring signals are fused to obtain monitoring data.

[0014] Optionally, the process of transmitting the monitoring data includes:

[0015] The monitoring data is compressed and scheduled according to priority scheduling rules. During the scheduling process, the compressed monitoring data is divided into data packets, and the transmission priority of the data packets is determined. The data packets are sent to a time-sensitive network for transmission according to the transmission priority. The data packets transmitted by the time-sensitive network are received through a high-speed industrial Ethernet, and the data packets are sorted according to the receiving time. The data packets are spliced according to the sorting results to obtain the transmitted monitoring data, and the transmitted monitoring data is decompressed to obtain the transmitted monitoring data.

[0016] Optionally, the process of obtaining the spatial distribution characteristics of temperature and pressure includes:

[0017] The transmitted monitoring data is stored in a distributed database and batch processed using a parallel computing framework. During batch processing, the monitoring data is preprocessed and fused using a spatial interpolation algorithm to obtain the spatial distribution characteristics of temperature and pressure.

[0018] A threshold value is determined based on the spatial distribution characteristics of the temperature and pressure to obtain an abnormal area, and the abnormal area is analyzed to obtain the location, time and severity of the abnormal area.

[0019] Optionally, the process of making predictions based on stored monitoring data includes:

[0020] The current monitoring data is predicted by a machine learning algorithm to obtain a predicted monitoring data change trend; wherein the machine learning algorithm is trained based on historical monitoring data.

[0021] Optionally, the process of mining and analyzing the stored monitoring data includes:

[0022] The stored monitoring data is used as historical monitoring data, accident data is extracted from the historical monitoring data, wherein the accident data includes temperature and pressure at the time of the accident, the accident data is clustered and analyzed to obtain patterns and regularities of different types of accidents, a knowledge base is constructed, and the accident patterns and regularities are stored in the knowledge base, wherein the knowledge base includes accident types, key features, and trigger conditions, wherein the key features include;

[0023] Obtaining a characteristic vector of the monitoring data, wherein the characteristic vector includes a change amount and an average value of temperature and pressure;

[0024] The feature vector and the monitoring data change trend are matched according to key features and trigger conditions to obtain the corresponding accident type to obtain a safety monitoring early warning result.

[0025] On the other hand, the present invention provides a safety monitoring and early warning system for underground energy storage space, which is used to execute the above method.

[0026] Compared with the prior art, the present invention has the following advantages and technical effects:

[0027] The present invention discloses a safety monitoring and early warning method for underground energy storage spaces. In view of the problems that the underground energy storage space environment is complex and changeable, the internal structure is complex, and the monitoring signal is susceptible to interference, the present invention adopts a distributed multi-point monitoring method and deploys multiple temperature and pressure sensors to achieve comprehensive monitoring coverage. The signal quality is improved by an adaptive signal processing algorithm, and high-speed industrial Ethernet is used to ensure data transmission. Distributed databases and big data platforms are used to achieve efficient data storage and analysis, and machine learning algorithms are used to establish a prediction model to provide early warning for abnormal situations. A digital twin model is constructed to carry out simulation and deduction of accident conditions and optimize disaster prevention and mitigation measures. The present invention realizes intelligent and precise monitoring and early warning of underground energy storage spaces, and improves the safety and reliability of energy storage systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0029] Figure 1 This is a flow chart of a safety monitoring and early warning method for underground energy storage space according to an embodiment of the present invention;

[0030] Figure 2 This is a structural diagram of a safety monitoring and early warning system for underground energy storage space according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0032] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0033] like Figure 1 In this embodiment, a safety monitoring and early warning method and system for underground energy storage space may specifically include:

[0034] S101. In view of the complex and changeable characteristics of the underground energy storage space environment, a distributed multi-point monitoring method is adopted, and multiple temperature and pressure sensors are deployed in the space to achieve comprehensive monitoring coverage of the entire space.

[0035] Given the complex and ever-changing nature of underground energy storage environments, a distributed multi-point monitoring approach is employed. Multiple temperature and pressure sensors are deployed throughout the space to capture real-time data from each sensor. Wireless communication technology transmits the temperature and pressure data collected by each sensor to a central monitoring system, enabling centralized data management and storage. Data preprocessing techniques are employed to remove noise and address outliers in the collected temperature and pressure data, improving data quality and reliability. Based on preset temperature and pressure thresholds, the processed data is analyzed and evaluated in real time to identify abnormalities exceeding the threshold. If an abnormality is detected, an alarm is triggered, notifying relevant personnel via text message, email, or other means, and appropriate measures are implemented according to pre-defined emergency response plans. Data visualization technology is used to display temperature and pressure data in real time on the monitoring interface in the form of graphs and heat maps, providing personnel with an intuitive understanding of the environmental conditions within the space.

[0036] For example, an underground energy storage space environmental monitoring system employs a distributed multi-point monitoring approach, deploying multiple temperature and pressure sensors throughout the space. For example, within a 1,000 cubic meter energy storage space, 20 temperature sensors and 10 pressure sensors can be evenly distributed, forming a grid-like monitoring network. This layout comprehensively captures the temperature and pressure distribution within the space, avoiding blind spots. Data collected by the sensors is transmitted to a central monitoring system via wireless communication. Low-power wide-area network technologies such as ZigBee and LoRa can be used for reliable data transmission. The central monitoring system uses a distributed database to store massive amounts of monitoring data, ensuring data security and scalability. Data preprocessing is a key step in improving data quality. Median filtering can be used to remove sudden noise from temperature data, and a moving average method can be used to smooth the pressure data curve. Interpolation can be used to correct outliers that significantly deviate from the normal range. These preprocessing operations significantly improve the accuracy of subsequent analysis. Real-time data analysis is based on pre-set thresholds. For example, the temperature threshold can be set between -10°C and 50°C, and the pressure threshold can be set between 0.8 MPa and 1.2 MPa. When a threshold is detected, the system immediately triggers an alarm. Alarm information can be sent to relevant personnel via SMS, email, and other means, ensuring timely resolution of potential risks. Data visualization technology provides monitoring personnel with an intuitive display of environmental conditions. Line charts can be used to display temporal trends in temperature and pressure, while heat maps can be used to visually display temperature distribution within a space. These visualization methods help quickly identify abnormal areas and patterns of change.

[0037] S102: To address the complex internal structure of underground energy storage spaces and the susceptibility of monitoring signals to interference and attenuation, an adaptive signal processing algorithm is used to filter, amplify, and correct the monitoring signals, improving the signal-to-noise ratio and dynamic range. Correlation analysis of monitoring signals at different measurement points identifies and eliminates abnormal interference, ensuring the reliability of monitoring data.

[0038] The system obtains raw monitoring signals from within the underground energy storage space and uses an adaptive filtering algorithm to remove noise interference from the signals, producing a filtered monitoring signal. The filtered monitoring signal is amplified using an adaptive gain control algorithm to improve the dynamic range and signal-to-noise ratio, resulting in an amplified monitoring signal. A signal propagation attenuation model is established based on the internal structure of the underground energy storage space. Adaptive correction is performed on the amplified monitoring signal to compensate for attenuation during signal propagation, resulting in a corrected monitoring signal. Corrected monitoring signals are obtained from different measurement points within the underground energy storage space. A cross-correlation analysis algorithm is used to calculate the correlation coefficients between the signals at different measurement points, generating a correlation matrix between the measurement points. Based on the correlation matrix, the correlation matrix is binarized according to a preset correlation threshold to generate a connectivity graph between the measurement points. Anomalous interference measurement points are identified using this connectivity graph. For each identified anomalous interference measurement point, an adaptive interference suppression algorithm is used to process its monitoring signal, suppressing the anomalous interference components and producing a reliable monitoring signal. The reliable monitoring signal from the anomalous interference measurement point is fused with the monitoring signals from other measurement points to obtain complete monitoring data within the underground energy storage space, providing data support for subsequent safety assessments and early warnings.

[0039] For example, the raw monitoring signals from underground energy storage spaces often contain various noise interferences, such as environmental noise and equipment vibration. Adaptive filtering algorithms can dynamically adjust filtering parameters based on signal characteristics to effectively remove noise. For example, a minimum mean square error (LMS) adaptive filter continuously adjusts the filter coefficients to minimize the mean square error between the output signal and the desired signal, effectively suppressing noise. The filtered signal may have a small amplitude, affecting the accuracy of subsequent processing. Adaptive gain control algorithms automatically adjust the amplification factor based on signal strength to improve the signal's dynamic range. For example, a higher gain is used for signal segments with smaller amplitudes, while a lower gain is used for signal segments with larger amplitudes to avoid signal distortion. The complex structure of underground energy storage spaces can cause signal attenuation during propagation. Establishing a signal propagation attenuation model, such as an exponential attenuation model that accounts for factors such as distance attenuation and dielectric absorption, can effectively compensate for the signal. For example, if a measurement point is 100 meters from the signal source and the attenuation model calculates a 20% attenuation of the signal strength, the signal at that measurement point can be amplified by 1.25 times to compensate. Cross-correlation analysis can reveal correlations between signals at different measurement points. By calculating the cross-correlation coefficients of signals at different measurement points, a correlation matrix reflecting the relationships between them can be obtained. Binarization with a correlation threshold (e.g., 0.8) can intuitively visualize the connectivity between the measurement points. For example, if the correlation coefficient between two measurement points is greater than 0.8, they are considered strongly connected; otherwise, they are weakly connected or have no connectivity. Identifying anomalous interference points is crucial for ensuring the reliability of monitoring data. By analyzing the connectivity graph, anomalous measurement points with weak or no connectivity to the majority of measurement points can be identified. For example, in a monitoring network consisting of 50 measurement points, if a measurement point has strong connectivity with only three measurement points, while the other measurement points have strong connectivity with an average of 20 measurement points, it may be experiencing anomalous interference. For identified anomalous interference points, adaptive interference suppression algorithms can be used. For example, adaptive notch filters can be used to estimate the frequency characteristics of the interfering signal in real time and dynamically adjust the filter parameters to effectively suppress interference components in specific frequency bands. This approach can effectively remove anomalous interference while preserving the useful signal. Finally, the processed abnormal measurement point signals are fused with signals from other measurement points using methods such as weighted averaging and Kalman filtering. For example, weights can be assigned based on the reliability of each measurement point, with higher weights assigned to highly reliable points. This yields more accurate and comprehensive monitoring data. This fused data provides reliable data support for safety assessments and early warnings of underground energy storage spaces, helping to promptly identify potential risks and implement appropriate prevention and control measures.

[0040] S103: To ensure low latency and high reliability of monitoring data transmission, high-speed industrial Ethernet and time-sensitive network technologies are used. Data compression and priority scheduling mechanisms are used to reduce data volume and transmission load.

[0041] Acquire monitoring data, compress the monitoring data according to preset data compression rules to obtain compressed monitoring data, and reduce the data level of the monitoring data; determine the transmission priority of the compressed monitoring data according to preset priority scheduling rules, and adopt different scheduling strategies for monitoring data of different priorities; divide the compressed monitoring data into multiple data packets, and send the data packets to the time-sensitive network for transmission according to the determined transmission priority; receive the data packets transmitted by the time-sensitive network through high-speed industrial Ethernet, and sort the data packets in order of reception time; splice the data packets according to the sorting results of the data packets to obtain complete compressed monitoring data; decompress the received compressed monitoring data according to the preset data decompression rules to obtain the original monitoring data; input the decompressed monitoring data into the monitoring data real-time processing module, analyze and process the monitoring data in real time, and promptly detect abnormal situations and trigger early warnings.

[0042] For example, in underground energy storage space monitoring systems, data compression is a key technology for reducing data transmission volume. For example, lossless compression algorithms such as Huffman coding or the LZW algorithm can be used to compress monitoring data such as temperature and pressure. For example, if the original temperature data is 20.5°C, 21.3°C, or 20.8°C, compression can convert it into a more compact binary representation, significantly reducing the data volume. Data transmission priority scheduling is crucial to ensuring the timely transmission of critical information. Priority can be set based on data type and importance, such as assigning safety-related pressure data the highest priority, followed by temperature data, and then humidity data. Preemptive scheduling can be used for high-priority data to ensure it is transmitted first during network congestion. Time-Sensitive Networking (TSN) technology ensures deterministic data transmission. For example, pressure data can be divided into multiple small 64-byte packets and transmitted over a TSN network. TSN reserves bandwidth and time slots for these packets, ensuring they arrive at their destination on schedule, minimizing network latency and jitter. High-speed industrial Ethernet networks such as EtherCAT or PROFINET enable high-speed and reliable data reception. These protocols support transmission rates up to 100Mbps or even 1Gbps, enabling rapid reception of data packets from the TSN network. Received data packets are sorted according to timestamps to ensure data timing accuracy. Packet concatenation is a critical step in restoring complete monitoring data. For example, if a pressure data packet is received, the system identifies the data type and sequence based on the packet header information, concatenating multiple pressure data packets in the correct order to reconstruct a complete pressure monitoring data sequence. Data decompression is the reverse of compression. If Huffman coding is used to compress temperature data, decompression uses the same coding tree to restore the compressed binary data to the original temperature value sequence, such as 20.5°C, 21.3°C, 20.8°C, and so on. Finally, the decompressed monitoring data enters the real-time processing module, which may include anomaly detection algorithms, such as those based on statistical methods. For example, for pressure data, if the pressure value at a given moment exceeds three standard deviations of the historical average, the system will immediately trigger an alert, alerting operators to the potential anomaly. This series of processing steps forms a complete data transmission and processing chain. From data acquisition, compression, transmission, reception, decompression, and analysis, each link has been carefully designed to ensure the efficient and reliable operation of the underground energy storage space monitoring system. In this way, the system can transmit large amounts of monitoring data within limited network bandwidth and promptly identify potential safety hazards, providing strong support for the safe operation of energy storage spaces.

[0043] S104. To achieve efficient storage and management of monitoring data, a distributed database and big data platform are selected to support rapid retrieval and analysis of massive amounts of data. Multi-point monitoring data is integrated and analyzed to obtain the spatial distribution characteristics of temperature and pressure and determine whether there are any abnormalities.

[0044] A distributed database is used to store massive amounts of monitoring data, improving data read and write efficiency through data sharding and parallel processing. Based on the distributed database, a big data processing platform is constructed, utilizing parallel computing frameworks such as MapReduce to perform batch analysis of monitoring data. Multi-point monitoring data is obtained from the distributed database, and the data from different monitoring points is cleaned and preprocessed to eliminate the influence of noise and outliers. A spatial interpolation algorithm is used to fuse the multi-point monitoring data to obtain the continuous distribution characteristics of temperature and pressure throughout the monitoring area. Based on preset temperature and pressure thresholds, anomaly detection is performed on the fused distribution data to determine whether there are areas outside the normal range. If an abnormal area is detected, the spatiotemporal distribution characteristics of the anomaly points are further analyzed to determine the location, time, and severity of the anomaly. The anomaly analysis results are compared with historical data, and time series analysis and machine learning algorithms are used to predict the development trend of the anomaly, providing support for early warning and decision-making.

[0045] For example, distributed databases are a key technology for processing massive amounts of monitoring data. For example, in a temperature and pressure monitoring system, data can be sharded and stored by time or location. For example, a year's data can be divided into 12 months and stored on different nodes, or nationwide monitoring points can be distributed by province on different servers. This approach significantly improves data read and write efficiency. For example, to query data for a specific month in a specific province, only the corresponding node needs to be accessed. The Hadoop ecosystem can be used to build a big data processing platform. HDFS is used to store raw data, HBase serves as a distributed database, and Spark performs data analysis. For example, when analyzing temperature data for a whole region over a year, the task can be broken down into multiple Map tasks to process the data for each sub-region in parallel, and then the results are aggregated through Reduce tasks, significantly reducing processing time. Data cleaning and preprocessing are crucial steps to ensure analysis quality. For temperature data, a reasonable range can be set, such as -50°C to 50°C, with values outside this range considered outliers. For pressure data, a moving average method can be used to eliminate short-term fluctuations and highlight long-term trends. These methods can effectively eliminate interference caused by factors such as equipment failures and transmission errors. Spatial interpolation algorithms can be used to construct a continuous distribution field for temperature and pressure. For example, kriging is a commonly used geostatistical method that accounts for spatial autocorrelation and can better reflect the spatial variation characteristics of data. For example, given temperature data for Beijing, Tianjin, and Shijiazhuang, kriging interpolation can be used to estimate the temperature distribution across the entire North China Plain. Anomaly detection is the core of the early warning system. Thresholds can be set based on historical data, such as determining an anomaly when the summer maximum temperature in a region exceeds 40°C or the winter minimum temperature falls below -20°C. Abnormal pressure may indicate a geological disaster; for example, a sudden increase in pressure in a region may indicate an earthquake precursor. Analyzing spatiotemporal distribution characteristics helps locate the source of anomalies and assess their impact. For example, if an abnormally high temperature is detected around a chemical park, it may be caused by a chemical leak. By analyzing the spatial distribution and temporal evolution of anomalies, pollution sources can be tracked and diffusion trends can be predicted.

[0046] Trend forecasting based on historical data can be achieved using time series models such as ARIMA or deep learning methods such as LSTM networks. For example, analyzing temperature changes in a region over the past decade can predict future climate trends and provide a basis for formulating long-term environmental policies. Machine learning algorithms such as random forests can also be used for multi-factor analysis, such as studying the combined impact of factors such as temperature, humidity, and wind speed on air quality. The implementation of this system can bring significant technical benefits. Real-time monitoring and rapid analysis enable timely detection and resolution of environmental anomalies, significantly reducing potential risks. Multi-dimensional data fusion and intelligent analysis enhance the scientific nature and foresight of decision-making. Moreover, with the accumulation of data and algorithm optimization, the system's forecasting accuracy will continue to improve, providing increasingly powerful support for environmental protection and resource management.

[0047] The spatial interpolation algorithm is used to fuse the multi-point monitoring data to obtain the continuous distribution characteristics of temperature and pressure in the entire monitoring area.

[0048] Based on the pre-established monitoring point location information, the temperature and pressure data of each monitoring point are obtained; the least squares method is used to fit the obtained temperature and pressure data to obtain the spatial interpolation function of temperature and pressure; the monitoring area is divided into several grids, and the temperature and pressure estimates of each grid center point are calculated according to the spatial interpolation function; it is determined whether the temperature and pressure estimates of each grid center point exceed the preset threshold range. If so, the grid is marked as an abnormal area; cluster analysis is performed on the abnormal area to obtain the distribution characteristics and range boundaries of the abnormal area; based on the distribution characteristics and range boundaries of the abnormal area, the continuous distribution characteristics of temperature and pressure in the monitoring area are determined; the continuous distribution characteristics of temperature and pressure in the monitoring area are presented in a visual manner to facilitate analysis and decision-making.

[0049] For example, in a monitoring system, temperature and pressure data must first be acquired from each monitoring point. This data is typically collected by sensors distributed throughout the monitoring area and transmitted to a data center via a communications network. For example, an area might have 100 monitoring points, each equipped with a temperature and pressure sensor, collecting data every 10 minutes. After acquiring the data, spatial interpolation using the least squares method is a common approach. The least squares method fits the data by minimizing the sum of squared errors to obtain spatial distribution functions for temperature and pressure. This method effectively handles data from unevenly distributed monitoring points and provides continuous estimates for the entire area. The monitoring area is divided into a grid for ease of calculation and analysis. For example, an industrial park is divided into a 1000×1000 grid, with each grid representing an area of 10 square meters. Using the spatial interpolation function obtained above, temperature and pressure estimates can be calculated for each grid center point. This creates a high-resolution distribution map of the entire area. Next, the temperature and pressure at each grid point must be determined to determine if they are abnormal. Preset thresholds may be based on safety standards or historical data, such as a temperature limit of 50°C or a pressure limit of 10 MPa. If the estimated value at a grid point exceeds these thresholds, it is marked as an anomaly. Clustering analysis of anomaly areas can help identify the spatial distribution characteristics of the anomaly. Common clustering algorithms such as K-means or DBSCAN can group adjacent anomaly grid points together, thereby determining the extent and shape of the anomaly area. For example, a circular high-temperature area with a diameter of approximately 100 meters may be found in the northeast corner of the area. Based on the clustering results, the continuous distribution characteristics of temperature and pressure throughout the monitoring area can be determined. This includes not only the location and extent of the anomaly area, but also the distribution trend of normal areas. For example, the temperature may gradually decrease from the center of the area to the periphery, while the pressure is higher around certain production equipment. Finally, visualizing the analysis results is crucial. Heat maps can be used to display the spatial distribution of temperature and pressure, using different colors to represent different value ranges. Anomalous areas can be highlighted with special markers or outlines. This intuitive visualization helps managers quickly identify potential problem areas and make timely decisions and responses. Through this series of steps, the monitoring system effectively transforms discrete monitoring point data into a continuous distribution characteristic for the entire area and identifies anomalies. This method not only improves the accuracy and coverage of monitoring, but also provides strong support for anomaly detection and early warning, which is of great significance for ensuring safe operations.

[0050] S105. To proactively identify potential abnormal risks, we use machine learning algorithms to establish temperature and pressure prediction models, estimating future trends. By mining and analyzing historical data, we identify patterns and patterns in accident occurrences and build a knowledge base. These patterns and patterns are then compared with real-time monitoring data. When similar abnormal patterns are detected, appropriate warning mechanisms are triggered.

[0051] Historical temperature and pressure data was obtained and preprocessed, including missing value and outlier handling and data normalization, to obtain a dataset suitable for modeling. Based on this preprocessed historical data, a temperature and pressure prediction model was established using time series prediction algorithms such as ARIMA and LSTM. Through model training and optimization, a model with high prediction accuracy was obtained.

[0052] Accident characteristics such as temperature and pressure at the time of the accident are extracted from historical data. Cluster analysis is performed on the accident data using clustering algorithms such as K-means and DBSCAN to identify patterns and regularities associated with different accident types. These patterns and regularities are stored in a knowledge base, which contains information such as accident type, key characteristics, and triggering conditions. Knowledge representation methods such as ontologies and rules are used to effectively organize and manage this knowledge. Real-time temperature and pressure monitoring data is acquired, processed, and features extracted in real time to generate feature vectors that match the prediction model and the knowledge base. This real-time feature vector is input into the prediction model to obtain predicted values for temperature and pressure over a period of time. Simultaneously, the feature vector is matched against accident patterns in the knowledge base to determine whether there are any abnormal risks. If the predicted value exceeds the normal range, or if the real-time data closely matches the accident pattern, an early warning mechanism is triggered, generating an alert and notifying relevant personnel. Simultaneously, automatic or semi-automatic countermeasures are implemented based on the action plan in the knowledge base to prevent the accident.

[0053] For example, first, historical temperature and pressure data is acquired. This data may come from multiple sensors, recording measurements at different points in time. For example, a chemical plant may have recorded temperature and pressure data every hour for the past year. This data is preprocessed to address missing values. For example, if data for a particular time point is missing, the average of the data from the preceding and following time points can be used to fill the missing values. For outliers, such as a sudden and unreasonable temperature increase at a certain point in time, this can be identified and corrected by setting a threshold or using statistical methods (such as the 3σ principle). This data can be replaced with the average of the data from adjacent time points or by using more complex interpolation methods. Data normalization converts data of different dimensions to the same scale. For example, temperature data can be converted from degrees Celsius to a value between 0 and 1, and pressure data can be processed similarly. This eliminates dimensionality effects and improves model training efficiency and accuracy. Next, a time series prediction algorithm is used to build a model. For example, the long short-term memory (LSTM) network is a special type of recurrent neural network that is particularly well-suited for processing time series data. For example, a LSTM model can be trained using the past year's temperature and pressure data as a training set to predict temperature and pressure changes over the next 24 hours. During the model training process, the network parameters are continuously adjusted to minimize the error between the model prediction value and the actual value.

[0054] Next, accident characteristics are extracted from historical data. For example, before a certain accident, both temperature and pressure showed a rapid upward trend. Using the K-means clustering algorithm to analyze these accident data, accidents can be classified into several categories, such as high-temperature, high-pressure, and low-temperature, low-pressure. Each type of accident has its own specific patterns. For example, before a high-temperature, high-pressure accident, temperature and pressure typically rise rapidly within a short period of time. These accident patterns and patterns are stored in a knowledge base. A knowledge base can be thought of as a structured database for storing and managing accident-related information. For example, a knowledge base might include a rule: "If the temperature rises by more than 50 degrees Celsius and the pressure rises by more than 1 MPa within one hour, a high-temperature, high-pressure accident is likely to have occurred." Using knowledge representation methods, such as ontologies, accident types, key characteristics, and triggering conditions can be formally described. For example, the accident type is "high-temperature, high-pressure accident," the key characteristics are "temperature rises by more than 50 degrees Celsius within one hour" and "pressure rises by more than 1 MPa within one hour," and the triggering conditions are "temperature rises by more than 50 degrees Celsius within one hour" and "pressure rises by more than 1 MPa within one hour," making it easier for computers to understand and reason about them. Real-time temperature and pressure monitoring data is acquired and processed and feature extracted in real time. For example, sensors collect temperature and pressure data once every minute, and the system calculates features such as the rate of change and average value of temperature and pressure in real time. These features are combined into a feature vector, such as [temperature = 80°C, pressure = 2 MPa, temperature rate of change = 5°C / minute, pressure rate of change = 0.2 MPa / minute]. This real-time feature vector is input into the prediction model to predict temperature and pressure for a period of time in the future. For example, based on the current feature vector, the LSTM model predicts that the temperature will rise by 30°C and the pressure will rise by 1 MPa within the next hour. Simultaneously, the feature vector is matched with accident patterns in the knowledge base. For example, if the current feature vector has a high similarity with the feature vector of a high-temperature and high-pressure accident, the warning mechanism is triggered. If the predicted value exceeds the normal range, such as if the temperature is predicted to exceed the safety threshold within the next hour, or if the real-time data closely matches the accident pattern, such as if the current feature vector has a similarity of more than 90% with the feature vector of a high-temperature and high-pressure accident, the warning mechanism is triggered. It generates warning messages, such as "Warning: Temperatures are predicted to exceed safety thresholds within the next hour. A high-temperature, high-pressure accident may occur!" and sends them to relevant personnel, such as regional safety managers. Simultaneously, based on the solutions in the knowledge base, it automatically or semi-automatically implements countermeasures, such as providing ventilation or prompting operators to perform relevant operations, to prevent accidents. This approach helps prevent accidents before they occur, nipping them in the bud and ensuring production safety.

[0055] The real-time feature vector is input into the prediction model to obtain the predicted values of temperature and pressure in the future.

[0056] Real-time temperature and pressure data is collected by sensors and preprocessed to generate standardized temperature and pressure feature vectors. The time window size is determined based on the predicted time period, and the feature vectors are divided into time windows to generate a series of feature vector sequences. This feature vector sequence is then input into a long short-term memory (LSTM) neural network model. Through model training, the mapping relationship between the feature vector sequence and future temperature and pressure values is learned. During model training, a cross-validation method is used to divide the dataset into a training set and a validation set. Model performance is continuously optimized by adjusting model hyperparameters until the prediction error on the validation set reaches a preset threshold. The trained LSTM model is saved for subsequent temperature and pressure predictions. When a prediction is required, the real-time temperature and pressure feature vectors are input into the model to generate predicted temperature and pressure values for the future period. To improve prediction accuracy, an ensemble learning method can be used to train multiple LSTM models with different structures and parameters. The prediction results of each model are weighted averaged to obtain the final predicted value. Based on the error between the predicted and actual values, the model is updated and optimized online to continuously improve its prediction performance. At the same time, the prediction results are compared with the preset temperature and pressure thresholds. If the predicted value exceeds the threshold range, the early warning mechanism is triggered and relevant personnel are notified to handle it.

[0057] For example, real-time temperature and pressure data collection is crucial in industrial production. The temperature and pressure within underground energy storage spaces directly impact product quality and production safety. Sensors can collect data multiple times per second, ensuring real-time and accurate monitoring. Data preprocessing is fundamental to model training. Raw data may contain noise and outliers, requiring filtering and normalization. For example, the temperature data for an underground energy storage space ranges from 0-50°C and the pressure ranges from 0-10 MPa. These can be normalized to a range of 0-1 to facilitate model learning. The choice of time window affects the accuracy and timeliness of predictions. For rapidly changing processes, a smaller time window, such as 5 minutes, may be required; for slowly changing processes, a larger time window, such as 1 hour, may be selected. A suitable time window can capture the temporal characteristics of the data and improve prediction accuracy. The LSTM model excels in processing time series data. It can learn long-term dependencies and is suitable for time-dependent data such as temperature and pressure. For example, in underground energy storage spaces, temperature and pressure changes often exhibit hysteresis and periodicity, which LSTM can effectively capture. Cross-validation is an effective method for evaluating model performance. K-fold cross-validation can be used: divide the dataset into K parts, using K-1 parts for training and 1 part for validation, repeating this cycle K times. This fully utilizes limited data and avoids overfitting. For example, if an underground energy storage space has one year of historical data, 12-fold cross-validation can be used, with each month's data used as a validation set. Adjusting model hyperparameters is key to improving prediction accuracy. Methods such as grid search or random search can be used to find the optimal hyperparameter combination. For example, the number of hidden layers, number of neurons, and learning rate in an LSTM model all affect model performance. In practice, it may be necessary to try dozens or even hundreds of combinations before finding the optimal configuration. Ensemble learning can further improve prediction accuracy. Multiple LSTM models can be trained, each using different initialization parameters or network architectures. For example, five LSTM models can be trained, each with 1, 2, 3, 4, and 5 hidden layers, and then their predictions can be weighted averaged. This approach can reduce the bias of individual models and improve the stability of the overall prediction. Online model updates are crucial for maintaining prediction performance. Over time, process conditions may change, causing model performance to degrade. Regularly retraining the model with new data, or employing incremental learning methods, allows the model to adapt to new operating conditions. For example, fine-tuning the model weekly using the most recent week's data ensures model timeliness while avoiding the computational overhead of frequent, large-scale training. Early warning mechanisms are a crucial component of a predictive system. Safety thresholds for temperature and pressure can be set based on historical data and expert experience. When predicted values exceed these thresholds, the system automatically issues an alarm. For example, the safe temperature range for an underground energy storage space is 80-120°C, and the pressure range is 0.5-2 MPa.If the forecast shows that the temperature will reach 125°C in 30 minutes, the system will issue an alarm in advance so that the operator can take timely cooling measures to prevent safety accidents.

[0058] S106. To verify and optimize early warning strategies and emergency response plans, a digital twin model of the underground energy storage space is constructed. This digital twin model is a virtual replica of the actual energy storage space, including its geometry, physical properties, and operating status. Real-time monitoring data is used to update the digital twin model and simulate accident conditions. The digital twin model allows for testing different control strategies in a virtual environment and evaluating their effectiveness, thereby enabling the implementation of optimal disaster prevention and mitigation measures in the actual system.

[0059] Obtain the geometric structure, physical properties and operating status data of the underground energy storage space and build its digital twin model; update the digital twin model in real time through real-time monitoring of the acquired data to keep it synchronized with the actual system; set various accident conditions in the digital twin model to simulate the state changes when an accident occurs; design a variety of control strategies and disaster prevention and mitigation measures for each accident condition; simulate and deduce each control strategy and disaster prevention and mitigation measure in the digital twin model and evaluate their effectiveness; determine the optimal control strategy and disaster prevention and mitigation measures for various accident conditions based on the simulation results; apply the early warning strategies and emergency plans tested and optimized in the digital twin model to the actual system to improve the safety and reliability of the underground energy storage space.

[0060] For example, first, comprehensive basic information about the underground energy storage space is required, including its 3D geometry, physical parameters of the rock and soil layers, and current operating status. For example, an underground salt cavern gas storage reservoir requires a precise 3D model. 3D laser scanning technology can be used to measure the cavern's internal spatial configuration in detail, including its length, width, height, volume, and irregular shape details, thereby constructing a geometric model with centimeter-level accuracy. Furthermore, drilling and coring are required for rock mechanics experiments to determine key physical parameters of the surrounding rock, such as Young's modulus, Poisson's ratio, compressive strength, and permeability. For example, the average Young's modulus of the surrounding rock of a salt cavern gas storage reservoir was measured to be 25 GPa, Poisson's ratio 0.25, compressive strength 30 MPa, and permeability 0.01 mD. Furthermore, sensors installed at the wellhead are required to monitor the reservoir's current operating status, including pressure, temperature, and injected or withdrawn gas flow rates, in real time. For example, the current reservoir pressure is 15 MPa, the temperature is 40°C, and the gas injection rate is 100,000 cubic meters per hour. Based on this detailed data, a digital twin model of the underground energy storage space can be constructed. This model is not just a static three-dimensional model, but rather a dynamic system that reflects the state of the storage space in real time. A digital twin model can be understood as a virtual, dynamic replica of a physical entity in the digital world. For example, specialized energy storage simulation software can be used to input the collected geometric data, physical parameters, and operating status data into the software to construct a digital twin model that corresponds exactly to the actual salt cavern gas storage facility. To maintain consistency between the digital twin model and the actual system, a real-time data update mechanism is necessary. For example, on-site pressure, temperature, flow rate, and other sensors can be connected to the digital twin model. Through a data interface, the latest monitoring data can be transmitted to the model at regular intervals (for example, every five minutes), driving model updates. This ensures that parameters such as pressure and temperature in the digital twin model remain consistent with those of the actual gas storage facility. Various potential accident conditions can then be simulated within the digital twin model. For example, simulations can simulate reservoir volume reduction due to formation creep, caprock rupture due to excessive injection and production rates, and natural gas leaks due to seal failure. For caprock rupture, for example, the model can set the injection pressure to exceed the caprock rupture pressure to observe the rupture's location, extent, and impact on the reservoir's safety. For each accident scenario, various control strategies and mitigation measures are required. For example, to address the risk of caprock rupture, various control strategies can be designed, such as reducing the injection and production rate, optimizing the injection and production well layout, and implementing artificial supports. To evaluate the effectiveness of these measures, simulations are conducted within the digital twin model. For example, in a caprock rupture scenario, the effects of reducing the injection rate by 20%, 40%, and 60% can be tested. The development of the caprock rupture at these different rates can be observed and recorded, along with the extent of its impact on the reservoir's safety.By comparing simulation results from different control strategies, the optimal solution can be identified. For example, a comparison revealed that reducing the injection rate by 40% can maintain gas storage efficiency while reducing the risk of caprock rupture to an acceptable level. This strategy can then be identified as the optimal control strategy for addressing caprock rupture risk. Finally, the optimization strategy validated by the digital twin model is applied to the actual system. For example, the optimized injection and production rate parameters are distributed to the on-site automatic control system to guide actual production operations. Furthermore, the digital twin model is integrated with the on-site monitoring system to enable real-time early warning and emergency response. This allows the digital twin model to rapidly analyze and provide decision support should an anomaly occur in the actual system, improving the safety and reliability of underground energy storage and preventing risks before they occur.

[0061] like Figure 2 As shown, the present invention provides a safety monitoring and early warning system for underground energy storage space, which mainly includes:

[0062] The environmental monitoring module is used to address the complex and changeable characteristics of the underground energy storage space environment. It adopts a distributed multi-point monitoring method and deploys multiple temperature and pressure sensors in the space to achieve comprehensive monitoring coverage of the entire space.

[0063] The signal processing module is used to address the problem of complex internal structures of underground energy storage spaces and the susceptibility of monitoring signals to interference and attenuation. It uses adaptive signal processing algorithms to filter, amplify, and correct monitoring signals, improving the signal-to-noise ratio and dynamic range of the signals. By performing correlation analysis on monitoring signals at different measurement points, it identifies and eliminates abnormal interference to ensure the reliability of monitoring data.

[0064] The data transmission module is used to ensure low latency and high reliability of monitoring data transmission. It adopts high-speed industrial Ethernet and time-sensitive network technology, and reduces data volume and transmission load through data compression and priority scheduling mechanism.

[0065] The data storage and analysis module is used to achieve efficient storage and management of monitoring data. It uses a distributed database and big data platform to support rapid retrieval and analysis of massive data. It can also perform fusion analysis on multi-point monitoring data to obtain the spatial distribution characteristics of temperature and pressure and determine whether there are any abnormal conditions.

[0066] The risk warning module is used to detect possible abnormal risks in advance. It uses machine learning algorithms to establish temperature and pressure prediction models to estimate the changing trends in the future. By mining and analyzing historical data, it summarizes the patterns and regularities of accident occurrences and forms a knowledge base. These patterns and regularities are compared with real-time monitoring data. When similar abnormal patterns are detected, the corresponding warning mechanism is triggered.

[0067] The digital twin and optimization module is used to build a digital twin model of the underground energy storage space to verify and optimize early warning strategies and emergency plans. The digital twin model is a virtual replica of the actual energy storage space, including the geometric structure, physical properties and operating status of the space. The digital twin model is updated using real-time monitoring data to simulate accident conditions. Through the digital twin model, different control strategies can be tested in a virtual environment and their effects evaluated, thereby implementing the optimal disaster prevention and mitigation measures in the actual system.

[0068] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A safety monitoring and early warning method for underground energy storage space, characterized in that: include: Obtaining original monitoring data of the underground energy storage space, performing adaptive signal processing and correlation analysis on the original monitoring data to obtain monitoring data; The original monitoring data includes original temperature data and original pressure data; Transmitting the monitoring data through data compression and priority scheduling mechanism; The transmitted monitoring data is stored, and the monitoring data is fused and analyzed to obtain spatial distribution characteristics of temperature and pressure; Determine in real time whether there are abnormal conditions based on spatial distribution characteristics; Make predictions based on the stored monitoring data to obtain the predicted monitoring data change trend; The stored monitoring data is mined and analyzed to obtain a knowledge base, and the monitoring data and the changing trend of the monitoring data are judged based on the knowledge base to obtain a safety monitoring early warning result.

2. The method according to claim 1, characterized in that The original monitoring data of underground energy storage space includes: A plurality of temperature sensors and pressure sensors are arranged in the underground energy storage space by a distributed multi-point monitoring method, data is collected by the temperature sensors and pressure sensors, and the collected data is preprocessed to obtain original monitoring data.

3. The method according to claim 1, characterized in that The acquisition process of the monitoring data includes: The original monitoring data is preprocessed, and the preprocessed monitoring data is amplified by an adaptive gain control algorithm to obtain an amplified signal. According to the internal structure of the underground energy storage space, a signal propagation attenuation model is constructed, and the amplified signal is corrected by the signal propagation attenuation model to obtain a corrected monitoring signal. The correlation coefficients between different corrected monitoring signals are calculated by a cross-correlation analysis algorithm to obtain a correlation matrix. The correlation matrix is subjected to threshold judgment processing to obtain a connectivity relationship graph. The corrected monitoring signal is identified according to the connectivity relationship graph to obtain an abnormal interference measurement point. The monitoring signal of the abnormal interference measurement point is processed by an interference suppression algorithm to obtain a reliable monitoring signal. The reliable monitoring signal and other monitoring signals are fused to obtain monitoring data.

4. The method according to claim 1, wherein The process of transmitting the monitoring data includes: The monitoring data is compressed and scheduled according to priority scheduling rules. During the scheduling process, the compressed monitoring data is divided into data packets, and the transmission priority of the data packets is determined. The data packets are sent to a time-sensitive network for transmission according to the transmission priority. The data packets transmitted by the time-sensitive network are received through a high-speed industrial Ethernet, and the data packets are sorted according to the receiving time. The data packets are spliced according to the sorting results to obtain the transmitted monitoring data, and the transmitted monitoring data is decompressed to obtain the transmitted monitoring data.

5. The method according to claim 1, characterized in that The process of obtaining the spatial distribution characteristics of temperature and pressure includes: The transmitted monitoring data is stored in a distributed database and batch processed using a parallel computing framework. During batch processing, the monitoring data is preprocessed and fused using a spatial interpolation algorithm to obtain the spatial distribution characteristics of temperature and pressure. A threshold value is determined based on the spatial distribution characteristics of the temperature and pressure to obtain an abnormal area, and the abnormal area is analyzed to obtain the location, time and severity of the abnormal area.

6. The method according to claim 1, characterized in that The process of making predictions based on stored monitoring data includes: The current monitoring data is predicted by a machine learning algorithm to obtain a predicted monitoring data change trend; wherein the machine learning algorithm is trained based on historical monitoring data.

7. The method according to claim 1, characterized in that The process of mining and analyzing the stored monitoring data includes: The stored monitoring data is used as historical monitoring data, accident data is extracted from the historical monitoring data, wherein the accident data includes temperature and pressure at the time of the accident, the accident data is clustered and analyzed to obtain patterns and regularities of different types of accidents, a knowledge base is constructed, and the accident patterns and regularities are stored in the knowledge base, wherein the knowledge base includes accident types, key features, and trigger conditions, wherein the key features include; Obtaining a characteristic vector of the monitoring data, wherein the characteristic vector includes a change amount and an average value of temperature and pressure; The feature vector and the monitoring data change trend are matched according to key features and trigger conditions to obtain the corresponding accident type to obtain a safety monitoring early warning result.

8. A safety monitoring and early warning system for underground energy storage space, characterized in that: Used to execute the method according to any one of claims 1 to 7.

Images