Artificial Intelligence-Based Karst Cave Environment Data Analysis Method and System

Abstract

The present invention provides a method and system for analyzing karst cave environment data based on artificial intelligence. By obtaining the monitoring data of the karst cave environment of the first template and the second template and conducting multiple rounds of parameter learning, high-precision prediction of the evolution pattern and risk knowledge points of the karst cave environment is achieved. By constructing a prediction network for the cave environment evolution pattern and a prediction network for the cave environment risk, the evolution pattern characteristics of the cave environment and their corresponding risk knowledge points can be accurately estimated and predicted. During the network parameter learning process, through continuous iterative optimization, the prediction network gradually approaches the real situation until the network convergence requirement is met, thereby generating a cave environment risk prediction network and a cave environment evolution pattern prediction network with high-precision prediction capabilities. This not only improves the intelligent level of karst cave environment monitoring, but also realizes the accurate prediction of the karst cave environment evolution pattern and timely discovery and warning of potential environmental risk points.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence. Specifically, it relates to a method and system for analyzing karst cave environment data based on artificial intelligence. Background Art

[0002] Karst caves, as unique natural landforms on Earth, have a complex and variable internal environment, involving the interaction of multiple aspects such as geology, climate, hydrology, and ecology. For a long time, the monitoring and analysis of karst cave environments have mainly relied on manual field investigations and traditional instrument measurement methods. However, these methods have significant limitations, such as long monitoring periods, difficult data acquisition, poor real-time performance, and the inability to comprehensively reflect the dynamic changes in the cave environment.

[0003] In order to gain a deeper understanding of the evolution law of karst cave environments, related technologies have continuously explored new monitoring and analysis techniques. Traditional environmental monitoring methods often focus on the measurement of single or a few environmental parameters, making it difficult to comprehensively reveal the multi-dimensional characteristics of the cave environment and their interaction relationships. In addition, the evolution of the cave environment is often accompanied by a series of complex risk factors, such as geological disasters and ecological imbalances. The timely identification and early warning of these risk factors are crucial for the protection of cave resources and personnel safety.

[0004] Therefore, there is an urgent need for a method that can efficiently and accurately analyze karst cave environment data to achieve a comprehensive understanding of the cave environment evolution pattern and timely early warning of risk factors. Summary of the Invention

[0005] In view of the above-mentioned problems, in combination with the first aspect of the present invention, embodiments of the present invention provide a method for analyzing karst cave environment data based on artificial intelligence. The method includes:

[0006] Obtain the first template karst cave environment monitoring data and the second template karst cave environment monitoring data. The first template karst cave environment monitoring data carries the marked cave environment evolution pattern features and the marked cave environment risk knowledge points corresponding to the marked cave environment evolution pattern features. The second template karst cave environment monitoring data carries the marked cave environment risk knowledge points corresponding to the marked cave environment evolution pattern features. Both the first template karst cave environment monitoring data and the second template karst cave environment monitoring data contain monitoring records arranged in chronological order. Each monitoring record contains multiple environmental parameter values. The environmental parameter values include one or more combinations of air temperature parameter value, air humidity parameter value, carbon dioxide concentration parameter value, air pressure parameter value, cave internal structure change parameter value, oxygen concentration parameter value, wind speed parameter value, light intensity parameter value, methane concentration parameter value, hydrogen sulfide concentration parameter value, cave water flow velocity parameter value, cave water pH parameter value, soil humidity parameter value, rock temperature parameter value, radon gas concentration parameter value, microbial community quantity parameter value, cave wall roughness parameter value, cave top drip speed parameter value, negative ion concentration parameter value, and cave dust concentration parameter value;

[0007] In the (a + 1)-th round of network parameter learning task, use the cave environment evolution pattern prediction network generated in the a-th round of network parameter learning task to predict the second template karst cave environment monitoring data, and generate the first cave environment evolution pattern features. The cave environment evolution pattern prediction network is used to estimate the cave environment evolution pattern features in the karst cave environment monitoring data, where a is a positive integer;

[0008] Use the first template karst cave environment monitoring data to perform parameter learning on the cave environment risk prediction network generated in the a-th round of network parameter learning task, and generate the cave environment risk prediction network in the (a + 1)-th round of network parameter learning task. The cave environment risk prediction network is used to predict the cave environment risk knowledge points of the cave environment evolution pattern features for the karst cave environment monitoring data;

[0009] Use the basic cave environment evolution pattern prediction network in the (a + 1)-th round of network parameter learning task and the cave environment risk prediction network generated in the (a + 1)-th round of network parameter learning task to predict the first template karst cave environment monitoring data and the second template karst cave environment monitoring data, and generate the second cave environment evolution pattern features corresponding to the first template karst cave environment monitoring data and the third cave environment evolution pattern features corresponding to the second template karst cave environment monitoring data;

[0010] Based on the first feature loss between the marked cave environment evolution pattern features and the second cave environment evolution pattern features, and the second feature loss between the first cave environment evolution pattern features and the third cave environment evolution pattern features, parameter learning is performed on the basic cave environment evolution pattern prediction network to generate the cave environment evolution pattern prediction network in the (a + 1)-th round of network parameter learning task, until the cave environment risk prediction network and the cave environment evolution pattern prediction network with completed parameter learning are generated when the network convergence requirement is met.

[0011] In another aspect, an embodiment of the present invention further provides an artificial intelligence-based karst cave environment data analysis system, including a processor and a machine-readable storage medium. The machine-readable storage medium is connected to the processor. The machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0012] Based on the above aspects, the embodiments of the present application realize high-precision prediction of the karst cave environment evolution pattern and risk knowledge points by obtaining the first template and second template karst cave environment monitoring data containing rich environmental parameter values and using these data to perform multiple rounds of parameter learning on the network. This method can accurately estimate and predict the evolution pattern features of the cave environment and their corresponding risk knowledge points by constructing a cave environment evolution pattern prediction network and a cave environment risk prediction network. During the network parameter learning process, through continuous iterative optimization, the prediction network gradually approaches the real situation until the network convergence requirement is met, thereby generating a cave environment risk prediction network and a cave environment evolution pattern prediction network with high-precision prediction capabilities. This not only improves the intelligent level of karst cave environment monitoring, but also realizes the accurate prediction of the karst cave environment evolution pattern, timely discovery and warning of potential environmental risk points, and provides data support for the protection of cave resources and disaster warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is a schematic execution flow diagram of an artificial intelligence-based karst cave environment data analysis method provided by an embodiment of the present invention.

[0014] Figure 2 is a schematic hardware architecture diagram of an artificial intelligence-based karst cave environment data analysis system provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] The present invention will be specifically described below with reference to the accompanying drawings of the specification. Figure 1 is a schematic flow diagram of an artificial intelligence-based karst cave environment data analysis method provided by an embodiment of the present invention. The artificial intelligence-based karst cave environment data analysis method will be introduced in detail below.

[0016] Step S110, obtain the first template karst cave environmental monitoring data and the second template karst cave environmental monitoring data. The first template karst cave environmental monitoring data carries the marked cave environmental evolution pattern features and the marked cave environmental risk knowledge points corresponding to the marked cave environmental evolution pattern features. The second template karst cave environmental monitoring data carries the marked cave environmental risk knowledge points corresponding to the marked cave environmental evolution pattern features. Both the first template karst cave environmental monitoring data and the second template karst cave environmental monitoring data include monitoring records arranged in chronological order. Each monitoring record includes multiple environmental parameter values. The environmental parameter values include one or a combination of air temperature parameter values, air humidity parameter values, carbon dioxide concentration parameter values, air pressure parameter values, cave internal structure change parameter values, oxygen concentration parameter values, wind speed parameter values, light intensity parameter values, methane concentration parameter values, hydrogen sulfide concentration parameter values, cave water flow velocity parameter values, cave water pH parameter values, soil humidity parameter values, rock temperature parameter values, radon gas concentration parameter values, microbial community quantity parameter values, cave wall roughness parameter values, cave top drip velocity parameter values, negative ion concentration parameter values, and cave dust concentration parameter values.

[0017] In this embodiment, consider a karst cave area named "Karst Cave A". In this cave, in order to obtain the first template karst cave environmental monitoring data and the second template karst cave environmental monitoring data, multiple monitoring stations are pre-set in the cave.

[0018] For the environmental monitoring data of the first template karst cave, it is assumed that monitoring stations are set up near the entrance, in the middle, and deep inside Cave A respectively. Each monitoring station is equipped with advanced environmental monitoring instruments, which can accurately measure various environmental parameter values. For example, at the station near the entrance, monitoring and data recording are carried out at fixed time intervals (such as once an hour) starting from a certain moment. The air temperature parameter value fluctuates within a day, being relatively low in the morning, gradually rising as the sun rises, reaching a relatively high value at noon, and then starting to decline in the afternoon; the air humidity parameter value is affected by the external climate of the cave and the water vapor exchange inside the cave, and will increase significantly after rainy weather. The carbon dioxide concentration parameter value will change due to factors such as biological respiration and rock chemical reactions inside the cave. For example, when the microbial community activities inside the cave are frequent, the carbon dioxide concentration may increase. The air pressure parameter value will change with the air flow exchange inside and outside the cave. When there is strong wind outside the cave, the air pressure inside the cave will be correspondingly affected. The parameter value of the internal structure change of the cave can be obtained through regular laser scanning or high-precision measurement equipment. For example, it is found that there is a slight expansion of cracks in the rock in a certain area. The oxygen concentration parameter value has a certain relationship with the carbon dioxide concentration value because biological respiration consumes oxygen and produces carbon dioxide. The wind speed parameter value is different in different areas of the cave. The wind speed may be relatively high in the narrow cave passages. The light intensity parameter value is relatively high in the area near the cave entrance and very low deep inside the cave. The methane concentration parameter value may change due to the decomposition of organic matter inside the cave, and the hydrogen sulfide concentration parameter value is also related to the chemical processes and microbial activities inside the cave. The cave water flow velocity parameter value depends on the underground river or seepage situation inside the cave. For example, the water flow velocity may increase during the rainy season. The pH value of the cave water is affected by factors such as the rock composition inside the cave and the mineral content in the water flow. The soil humidity parameter value is also monitored in some areas covered with soil inside the cave, which will affect the growth of plants (if there are a small number of plants adapted to the cave environment). The rock temperature parameter value is related to the heat conduction and air temperature inside the cave. The radon concentration parameter value may vary due to the radioactive characteristics of the rock. The number parameter value of the microbial community is obtained through laboratory analysis of samples collected from different areas inside the cave. The cave wall roughness parameter value is obtained through a special roughness measurement instrument. The dripping speed parameter value at the cave top is related to the water seepage situation of the rock at the cave top. The negative ion concentration parameter value is affected by the air ionization situation inside the cave. The dust concentration parameter value inside the cave is related to factors such as air flow and rock weathering inside the cave.

[0019] For each monitoring record, it contains the above-mentioned multiple environmental parameter values. Moreover, these monitoring records are arranged in chronological order, thus forming the first-template karst cave environment monitoring data. Meanwhile, based on years of research experience and in-depth understanding of the cave, researchers annotated these monitoring data, marking the characteristics of the cave environment evolution pattern and the corresponding annotated cave environment risk knowledge points. For example, when the carbon dioxide concentration in the cave continuously increases, the air humidity increases, and the rock temperature drops, the marked characteristics of the cave environment evolution pattern are "the internal ecological balance of the cave begins to develop towards a biological community that is favorable for moisture-loving and high-carbon dioxide-tolerant environments, and at the same time, rock weathering may accelerate", and the corresponding annotated cave environment risk knowledge point is "the risk of erosion of some fragile stalagmites and stalactites in the cave by acidic substances (formed by high-concentration carbon dioxide dissolving in water to form carbonic acid) may increase, and the biological community adapted to low-carbon dioxide environments may face survival pressure".

[0020] For the second-template karst cave environment monitoring data, it is also obtained in karst cave A, but it may be from different monitoring sites or some supplementary monitoring methods are adopted. These data also contain monitoring records arranged in chronological order. Each record also contains multiple environmental parameter values and carries the annotated cave environment risk knowledge points corresponding to the same annotated cave environment evolution pattern characteristics in the first-template karst cave environment monitoring data. For example, at another monitoring site near the underground river in the cave, the change situation of the water flow velocity parameter value and the changes of other related environmental parameter values are recorded. Although these data may be different from the first-template karst cave environment monitoring data in the specific values and change trends of some parameters, they are all related to the same cave environment evolution pattern characteristics and related environmental risk knowledge points.

[0021] Step S120, in the (a + 1)-th round of network parameter learning task, use the cave environment evolution pattern prediction network generated in the a-th round of network parameter learning task to predict the second-template karst cave environment monitoring data, and generate the first cave environment evolution pattern characteristics. The cave environment evolution pattern prediction network is used to estimate the cave environment evolution pattern characteristics in the karst cave environment monitoring data, where a is a positive integer.

[0022] Assume that a takes the value of 1. Then in the 2nd round of network parameter learning task, the cave environment evolution pattern prediction network has been generated in the 1st round of network parameter learning task. This network is constructed based on a large amount of previous data and algorithms and is used to estimate the cave environment evolution pattern characteristics in the karst cave environment monitoring data.

[0023] For the second - template karst cave environmental monitoring data of karst cave A, input it into this cave environmental evolution pattern prediction network. Taking the data of the monitoring station near the underground river in the cave mentioned before as an example, the network will analyze according to the monitoring records containing various environmental parameter values such as air temperature, humidity, and water flow velocity, following its internal complex algorithms and models.

[0024] Suppose the algorithm model of the network contains a multi - layer neural network structure. In the first layer, it may extract features from each input environmental parameter value. For example, for the water flow velocity parameter value, it will analyze features such as the change amplitude and change frequency at different time periods; for the air temperature parameter value, it will analyze the correlation features between it and the water flow velocity and other parameters. Then, these preliminarily extracted features will be passed to the next - layer network structure for further analysis and integration.

[0025] After being processed by multiple layers of the network, a first cave environmental evolution pattern feature is finally generated. This feature may be described as "In the area near the underground river, due to the seasonal fluctuations of the water flow velocity and the co - variation of air temperature and humidity, the cave environment presents a periodic humidity regulation pattern, and this pattern affects the distribution and activities of the surrounding microbial communities, resulting in the evolution of the microbial communities towards species adapted to higher humidity and a specific temperature range". This first cave environmental evolution pattern feature is an estimate of the cave environmental evolution pattern obtained by the network through complex calculations and analyses based on the input second - template karst cave environmental monitoring data.

[0026] Step S130: Use the first - template karst cave environmental monitoring data to perform parameter learning on the cave environmental risk prediction network generated in the a - th round of network parameter learning tasks, and generate the cave environmental risk prediction network in the (a + 1) - th round of network parameter learning tasks. The cave environmental risk prediction network is used to predict the cave environmental risk knowledge points of the cave environmental evolution pattern features from the karst cave environmental monitoring data.

[0027] Continuing to take karst cave A as an example, use the first - template karst cave environmental monitoring data to perform parameter learning on the cave environmental risk prediction network generated in the 1st round of network parameter learning tasks to generate the cave environmental risk prediction network in the 2nd round of network parameter learning tasks.

[0028] Suppose in the monitoring data of the first template karst cave environment, the data of the monitoring site in the middle of Cave A shows that during a certain period, the carbon dioxide concentration value continues to rise, while the oxygen concentration value slightly decreases, and the cave internal structure change parameter value shows that some rocks have minor deformations. According to the marked characteristics of the cave environment evolution pattern "the increase in carbon dioxide concentration leads to changes in the chemical environment in the cave, which may affect the biological community and rock stability" and the corresponding marked cave environment risk knowledge points "biodiversity may decrease, and there is a risk of collapse of some cave structures".

[0029] The cave environment risk prediction network first extracts the local state change characteristics of the first local karst cave space (here it is the monitoring area in the middle of Cave A). For example, for the state change of the increase in carbon dioxide concentration, the network will analyze characteristics such as the rising rate, the duration of persistence, and the degree of association with other environmental parameter values (such as air humidity, temperature, etc.), and generate the first local state change characteristics corresponding to the first local karst cave space. Then, it performs spatial coding representation on this local space. For example, according to factors such as its geographical location and spatial range in the cave, it generates the first environmental block space data corresponding to the first local karst cave space.

[0030] Load the first local state change characteristics and the first environmental block space data into the cave environment risk prediction network generated in the first-round network parameter learning task. The network processes these data according to its internal algorithm model, such as a risk assessment model based on probability statistics or a classification model in machine learning algorithms that it may contain. After calculation, the first estimated cave environment evolution pattern and the estimated cave environment risk knowledge points corresponding to the first template karst cave environment monitoring data are obtained. Suppose the first estimated cave environment evolution pattern obtained is "the increase in carbon dioxide concentration will lead to changes in the structure of some biological communities in the cave and a decrease in the stability of the cave structure", and the estimated cave environment risk knowledge points are "some organisms in the cave that rely on a specific gas environment may become extinct, and the possibility of local collapse of the cave top increases".

[0031] Then, based on the third feature loss between the first estimated cave environment evolution pattern and the first candidate cave environment evolution pattern (i.e., the previously marked "the increase in carbon dioxide concentration leads to changes in the chemical environment in the cave, which may affect the biological community and rock stability"), and the marked cave environment risk knowledge points and the estimated cave environment risk knowledge points, parameter learning is performed on the cave environment risk prediction network generated in the first round of network parameter learning tasks. The third feature loss here can be determined by calculating the degree of difference between the two in various aspects of describing the cave environment evolution pattern (such as changes in the biological community, changes in rock stability, etc.). For example, if there is a difference in the degree of impact on biological species in the description of the biological community change in the first estimated cave environment evolution pattern and the first candidate cave environment evolution pattern, then this difference will be quantified as part of the third feature loss. Similarly, for the marked cave environment risk knowledge points and the estimated cave environment risk knowledge points, differences in aspects such as the specific types of risks (such as the risk of biological extinction, the risk of cave collapse) and the degree of risk (such as high, medium, low) will also be compared. In this way, the parameters of the cave environment risk prediction network are adjusted to generate the cave environment risk prediction network in the second round of network parameter learning tasks.

[0032] Step S140: Using the basic cave environment evolution pattern prediction network in the (a + 1)-th round of network parameter learning tasks and the cave environment risk prediction network generated in the (a + 1)-th round of network parameter learning tasks, predict the first template karst cave environment monitoring data and the second template karst cave environment monitoring data, and generate the second cave environment evolution pattern features corresponding to the first template karst cave environment monitoring data and the third cave environment evolution pattern features corresponding to the second template karst cave environment monitoring data.

[0033] In this embodiment, the basic cave environment evolution pattern prediction network in the second round of network parameter learning tasks and the cave environment risk prediction network generated in the second round of network parameter learning tasks are used to predict the first template karst cave environment monitoring data and the second template karst cave environment monitoring data of karst cave A.

[0034] For the monitoring data of the first template karst cave environment, first use the basic cave environment evolution pattern prediction network in the second round of network parameter learning tasks for prediction. This basic cave environment evolution pattern prediction network includes an environmental state change encoding sub-network and a fully connected mapping sub-network. Taking the monitoring data near the entrance of Cave A as an example, the environmental state change encoding sub-network extracts features from each monitoring record. For example, for a monitoring record at a certain moment, which contains parameter values such as air temperature, humidity, and wind speed, it extracts an initial feature vector sequence reflecting the changes in the cave environment state. This initial feature vector sequence may contain information such as the relative change amplitude of each parameter value and the proportional relationship with other parameters.

[0035] Then, based on this initial feature vector sequence, use a clustering algorithm to cluster the matching environmental states into the same category. For example, cluster the states with low air temperature, high humidity, and low wind speed into one category, and each category represents an environmental state node. According to the time sequence and the similarity between states, construct the intra-node transition relationship within the environmental state nodes and the inter-node transition relationship between environmental state nodes to generate a state transition diagram. This state transition diagram describes the dynamic change process of the environmental state near the cave entrance over time, such as the transition process from the cold and dry state in winter to the warm and humid state in spring.

[0036] Traverse this state transition diagram, starting from the initial state, and assign a unique code to the environmental state at each time point according to the time sequence and state transition relationship to generate a state sequence code. Arrange this state sequence code in chronological order to form an environmental state change trajectory. Use smoothing filtering to remove the noise and jitter in the environmental state change trajectory, such as eliminating the sudden fluctuations in temperature values caused by the small errors of the monitoring instruments. At the same time, use time series analysis to extract the periodic features and trend features in the environmental state change trajectory, such as finding that the air temperature has seasonal periodic changes every year and has a slow upward trend year by year. Use pattern recognition to identify the target change patterns or abnormal points in the environmental state change trajectory, such as identifying the situation where the wind speed increases abnormally due to large-scale construction outside the cave in a certain year. After obtaining the enhanced environmental state change trajectory features, load them into the fully connected mapping sub-network, and finally generate the second cave environment evolution pattern features corresponding to the first template karst cave environment monitoring data. This second cave environment evolution pattern feature may be described as "Near the cave entrance, over time, due to the combined effects of external environment and internal cave factors, the cave environment shows seasonal temperature and humidity fluctuations, and these fluctuations have a periodic impact on the biological community at the cave entrance. At the same time, in the long run, due to the increase in external interference factors, the environmental stability at the cave entrance shows a downward trend."

[0037] For the environmental monitoring data of the second template karst cave, the basic cave environmental evolution pattern prediction network in the second round of network parameter learning task and the cave environmental risk prediction network generated in the second round of network parameter learning task are used for prediction. First, the environmental state change encoding sub-network encodes and represents the environmental monitoring data of the second template karst cave. Taking the monitoring data near the underground river in the cave as an example, the environmental state change trajectory characteristics corresponding to the environmental monitoring data of the second template karst cave are generated. Then, the cave environmental risk prediction network generated in the second round of network parameter learning task extracts the local state change characteristics from the environmental state change trajectory characteristics corresponding to the environmental monitoring data of the second template karst cave, and generates the second local state change characteristics corresponding to the marked cave environmental evolution pattern characteristics. For example, for the part of the water flow velocity change in the environmental state change trajectory characteristics near the underground river, the local state change characteristics related to the change of the surrounding biological community are extracted.

[0038] Integrate the environmental state change trajectory characteristics and the local state change characteristics to generate enhanced trajectory characteristics. For example, integrate the trajectory characteristics related to the water flow velocity of the underground river and the local state change characteristics of the biological community change to obtain enhanced trajectory characteristics containing more information. Finally, load this enhanced trajectory characteristic into the fully connected mapping sub-network to generate the third cave environmental evolution pattern characteristics corresponding to the environmental monitoring data of the second template karst cave. This third cave environmental evolution pattern characteristic may be described as "in the area near the underground river, the change of water flow velocity is closely related to the surrounding biological community and the internal chemical environment of the cave. The seasonal fluctuation of water flow velocity leads to the periodic succession of the biological community, and this succession has an impact on the chemical substance cycle in the cave. At the same time, the cave structure has a tendency of local change under the long-term scouring of the water flow".

[0039] Step S150, based on the first feature loss between the marked cave environmental evolution pattern characteristics and the second cave environmental evolution pattern characteristics, and the second feature loss between the first cave environmental evolution pattern characteristics and the third cave environmental evolution pattern characteristics, perform parameter learning on the basic cave environmental evolution pattern prediction network to generate the cave environmental evolution pattern prediction network in the (a + 1)-th round of network parameter learning task, until the cave environmental risk prediction network and the cave environmental evolution pattern prediction network with completed parameter learning are generated when the network convergence requirement is met.

[0040] In this embodiment, based on the first feature loss between the labeled cave environment evolution pattern features and the second cave environment evolution pattern features, and the second feature loss between the first cave environment evolution pattern features and the third cave environment evolution pattern features, parameter learning is performed on the basic cave environment evolution pattern prediction network to generate the cave environment evolution pattern prediction network in the second round of network parameter learning tasks, until the cave environment risk prediction network and the cave environment evolution pattern prediction network with completed parameter learning are generated when the network convergence requirement is met.

[0041] Taking various cave environment evolution pattern features obtained in karst cave A before as an example, the labeled cave environment evolution pattern features of the first template karst cave environment monitoring data include the first candidate cave environment evolution pattern and the first local karst cave space. Suppose the first candidate cave environment evolution pattern is characterized as "due to the combined action of various environmental factors inside the cave, the biological community and the cave structure are in a dynamic balance state, but are prone to imbalance when affected by external disturbances", and the first local karst cave space is the middle area of cave A. The second cave environment evolution pattern features include the second estimated cave environment evolution pattern and the second estimated local karst cave space. Suppose the second estimated cave environment evolution pattern is "in the middle area of the cave, due to local environmental changes, the stability of the biological community is affected to a certain extent, but the change of the cave structure is not obvious", and the second estimated local karst cave space is also the middle area of cave A.

[0042] Calculate the first sub-feature loss between the first candidate cave environment evolution pattern and the second estimated cave environment evolution pattern. For example, in terms of the stability of the biological community, it is labeled as being prone to imbalance, while it is estimated as being affected to a certain extent, and the difference in this description needs to be quantified. In terms of the cave structure, it is labeled as no obvious change mentioned, while it is estimated as the change is not obvious, and this difference also needs to be quantified, so as to generate the first sub-feature loss. At the same time, calculate the second sub-feature loss between the first local karst cave space and the second estimated local karst cave space. Here, the quantification of differences in aspects such as spatial range and environmental features within the space may be involved, and finally the first feature loss is generated.

[0043] Similarly, for the first cave environmental evolution pattern feature and the third cave environmental evolution pattern feature, the first cave environmental evolution pattern feature includes the first estimated cave environmental evolution pattern and the first estimated local karst cave space, and the third cave environmental evolution pattern feature includes the third estimated cave environmental evolution pattern and the third estimated local karst cave space. Suppose the first estimated cave environmental evolution pattern is "the environmental changes near the underground river have a certain impact on the overall cave ecosystem", and the first estimated local karst cave space is the area near the underground river; the third estimated cave environmental evolution pattern is "the environmental changes near the underground river have an obvious impact on the distribution of some biological communities in the cave", and the third estimated local karst cave space is the area near the underground river. Calculate the third sub-feature loss between the first estimated cave environmental evolution pattern and the third estimated cave environmental evolution pattern, for example, quantify the differences in the descriptions of the impacts on the overall cave ecosystem and the distribution of some biological communities. At the same time, calculate the fourth sub-feature loss between the first estimated local karst cave space and the third estimated local karst cave space (here the loss may be relatively small because both are areas near the underground river), so as to generate the second feature loss.

[0044] Based on the first feature loss and the second feature loss, perform parameter learning on the basic cave environmental evolution pattern prediction network. For example, if the first feature loss indicates a large difference between the prediction and the annotation in terms of the stability of the biological community, then the network may adjust the parameter weights related to the biological community. If the second feature loss shows a deviation in the prediction of the ecological impact in the area near the underground river, then the network will correspondingly adjust the parameters related to the environmental factors in this area. By continuously iterating this process until the network convergence requirement is met, that is, both the first feature loss and the second feature loss are reduced to an acceptable range, at this time, generate the cave environmental risk prediction network and the cave environmental evolution pattern prediction network that have completed parameter learning.

[0045] After generating the cave environment risk prediction network and the cave environment evolution pattern prediction network that have completed parameter learning, assume that new target karst cave environment monitoring data is obtained. This data comes from a new monitoring area of karst cave A or more detailed data obtained by using a new monitoring technology in the original monitoring area. Load this target karst cave environment monitoring data into the cave environment evolution pattern prediction network and the cave environment risk prediction network that have completed parameter learning to obtain the target cave environment evolution pattern features corresponding to the target karst cave environment monitoring data and the cave environment risk knowledge points corresponding to the target cave environment evolution pattern features. For example, the target karst cave environment monitoring data shows that the hydrogen sulfide concentration in a certain area has suddenly increased, and at the same time, the number of microbial communities has changed abnormally. After being loaded into the network, the obtained target cave environment evolution pattern features may be "due to the sudden increase in the hydrogen sulfide concentration in this area, the microbial community structure has changed drastically, which may lead to an imbalance in the chemical environment in the cave", and the corresponding cave environment risk knowledge point is "the collapse of the microbial community may lead to the obstruction of the material cycle in the cave and may pose a risk of harmful gas accumulation".

[0046] Based on the above steps, the embodiment of the present application realizes high-precision prediction of the karst cave environment evolution pattern and risk knowledge points by obtaining the first template and second template karst cave environment monitoring data containing rich environmental parameter values and using these data to perform multiple rounds of parameter learning on the network. By constructing a cave environment evolution pattern prediction network and a cave environment risk prediction network, this method can accurately estimate and predict the evolution pattern features of the cave environment and the corresponding risk knowledge points. During the network parameter learning process, through continuous iterative optimization, the prediction network gradually approaches the real situation until the network convergence requirement is met, thereby generating a cave environment risk prediction network and a cave environment evolution pattern prediction network with high-precision prediction ability. This not only improves the intelligent level of karst cave environment monitoring but also realizes the accurate prediction of the karst cave environment evolution pattern and timely discovery and warning of potential environmental risk points, providing data support for the protection of cave resources and disaster warning.

[0047] In a possible implementation manner, step S140 includes:

[0048] Step S141, using the basic cave environment evolution pattern prediction network in the (a + 1)-th round of network parameter learning task to predict the first template karst cave environment monitoring data, and generating the second cave environment evolution pattern features corresponding to the first template karst cave environment monitoring data.

[0049] Taking the previously mentioned karst cave A as an example, in karst cave A, for example, the environmental monitoring data of the first template karst cave comes from monitoring stations in different areas of the cave, such as the entrance, the middle part, and the deep part. Taking the data of the station at the entrance as an example, the basic cave environmental evolution pattern prediction network processes the monitoring data at the entrance according to its internal algorithm. The environmental state change encoding sub-network in this network first extracts features for each monitoring record at the entrance, including multiple environmental parameter values such as air temperature parameter value, air humidity parameter value, carbon dioxide concentration parameter value, etc. Suppose that at a certain moment, the air temperature at the entrance is low, the humidity is high, and the carbon dioxide concentration is at a medium level. The environmental state change encoding sub-network will encode these parameter values according to their mutual relationships and comparisons with historical data, generating an initial feature vector sequence reflecting the change of the cave environmental state at the entrance at this moment. As time goes by, initial feature vector sequences at many such moments are obtained. Then, based on these initial feature vector sequences, a clustering algorithm is used to cluster similar environmental states into the same category. For example, the state with low temperature, high humidity, and stable carbon dioxide concentration is clustered into one category, and each category represents an environmental state node. Then, according to the time sequence and the similarity between states, the intra-node transfer relationship within the environmental state node and the inter-node transfer relationship between environmental state nodes are constructed, thus generating a state transition diagram. This state transition diagram details the dynamic change process of the environmental state at the cave entrance over time. Periodic change processes such as the temperature dropping and the humidity rising from day to night are reflected in the diagram. Finally, by traversing the state transition diagram, a unique code is assigned to the environmental state at each time point according to the time sequence and the state transition relationship to generate a state sequence code, and these state sequence codes are arranged in time sequence to form an environmental state change trajectory. Then the fully connected mapping sub-network processes this environmental state change trajectory, combines its internal predefined parameters and models, and finally generates the second cave environmental evolution pattern feature corresponding to the environmental monitoring data of the first template karst cave. This feature may be described as follows: at the cave entrance, due to the periodic changes of environmental parameters such as temperature, humidity, and carbon dioxide concentration and their interactions, there is a periodic adjustment in the biological community structure at the entrance, and there is a slow chemical change process on the cave wall under the long-term action of humidity and carbon dioxide.

[0050] Step S142: Use the basic cave environmental evolution pattern prediction network in the (a + 1)-th round of network parameter learning task and the cave environmental risk prediction network generated in the (a + 1)-th round of network parameter learning task to predict the second template karst cave environmental monitoring data, and generate the third cave environmental evolution pattern feature corresponding to the second template karst cave environmental monitoring data.

[0051] In a possible implementation manner, the basic cave environment evolution pattern prediction network includes an environmental state change encoding sub-network and a fully connected mapping sub-network.

[0052] Step S142 includes:

[0053] Step S1421: Use the environmental state change encoding sub-network to perform environmental state change encoding representation on the second template karst cave environment monitoring data, and generate the environmental state change trajectory feature corresponding to the second template karst cave environment monitoring data.

[0054] Step S1422: Use the cave environment risk prediction network generated in the (a + 1)-th round of network parameter learning task to extract local state change features from the environmental state change trajectory feature corresponding to the second template karst cave environment monitoring data, and generate the second local state change feature corresponding to the labeled cave environment evolution pattern feature.

[0055] Step S1423: Integrate the environmental state change trajectory feature and the local state change feature to generate an enhanced trajectory feature.

[0056] Step S1424: Load the enhanced trajectory feature into the fully connected mapping sub-network to generate the third cave environment evolution pattern feature corresponding to the second template karst cave environment monitoring data.

[0057] Taking the second template karst cave environment monitoring data in karst cave A as an example, assume that these data mainly come from the area near the underground river in the middle of the cave. First, the environmental state change encoding sub-network in the basic cave environment evolution pattern prediction network performs environmental state change encoding representation on the second template karst cave environment monitoring data. For the monitoring data in the area near the underground river in the middle, there are many environmental parameter values in each monitoring record. The environmental state change encoding sub-network will analyze and process these parameter values. For example, for the cave water flow velocity parameter value, it will consider factors such as the change range of the flow velocity in different seasons and the correlation between the flow velocity change and other environmental parameters. For the air humidity parameter value, it will analyze its response change when the underground river water flow fluctuates, etc. In this way, an initial feature vector sequence reflecting the environmental state change is generated for each monitoring moment, and then following the previously mentioned steps of clustering algorithm, constructing a state transition graph, generating a state sequence encoding, etc., finally generating the environmental state change trajectory feature corresponding to the second template karst cave environment monitoring data. This environmental state change trajectory feature can describe in detail the dynamic change of the environmental state in the area near the underground river in the middle over time. For example, when the flow velocity of the underground river increases greatly during the rainy season, the surrounding air humidity rises rapidly, and when the flow velocity slows down during the dry season, the humidity drops and the carbon dioxide concentration will increase slightly, etc., which are all reflected in this trajectory feature.

[0058] Next, the cave environment risk prediction network generated in the (a + 1)-th round of network parameter learning task is used to extract the local state change characteristics of the environmental state change trajectory features corresponding to the second template karst cave environment monitoring data. For example, for the part related to the biological community in the previously generated environmental state change trajectory features, the cave environment risk prediction network will deeply analyze the relationship between the numerical parameters of the biological community quantity, the numerical parameters of the microbial community quantity, etc. and other environmental parameters. If the pH value of the underground river water quality changes within a certain period of time, the cave environment risk prediction network will analyze the degree of influence of this change on the microbial community quantity. For example, when the pH value changes from weakly alkaline to weakly acidic, whether there is a tendency for the microbial community quantity to decrease and what is the rate of decrease, etc., so as to generate the second local state change characteristics corresponding to the marked cave environment evolution pattern features. This second local state change characteristics highlights the local state change situation related to the cave environment risk in the environmental state change trajectory features. In this example, it is the local state change of the influence of the pH value change of the underground river water quality on the microbial community quantity.

[0059] Then, the environmental state change trajectory features and the local state change characteristics are integrated to generate enhanced trajectory features. In the area near the underground river in the middle of Karst Cave A, the environmental state change trajectory features obtained previously, including the underground river flow velocity, the surrounding air humidity, the carbon dioxide concentration, etc., are integrated with the local state change characteristics reflecting the influence of the pH value on the microbial community quantity. During the integration process, the integration will be carried out according to the logical relationship and the mutual influence weight between the two. For example, if the underground river flow velocity has an indirect influence on the microbial community quantity, then during the integration, the trajectory features related to the flow velocity will be reasonably combined with the local state change characteristics related to the microbial community quantity according to this influence relationship to generate an enhanced trajectory feature containing more comprehensive information. This enhanced trajectory feature not only includes the overall change situation of the environmental state but also highlights the local change situation related to the cave environment risk.

[0060] Finally, load the enhanced trajectory features into the fully connected mapping sub-network to generate the third cave environment evolution pattern features corresponding to the second template karst cave environment monitoring data. The fully connected mapping sub-network processes the enhanced trajectory features according to the complex internal mapping relationships and predefined parameters. In the example of Karst Cave A, the fully connected mapping sub-network will consider various information in the enhanced trajectory features, such as the mutual relationships between factors like the underground river velocity, the change in the number of microbial communities, and the air humidity. Eventually, it generates the third cave environment evolution pattern features. This feature may be described as follows: In the area near the underground river in the middle of Karst Cave A, the change in the water flow velocity of the underground river indirectly changes the number of microbial communities by affecting factors such as the surrounding air humidity and the pH value of the water quality, thereby affecting the biological community structure of the entire area. And this impact may further lead to local adjustments in the chemical substance cycle inside the cave. At the same time, the scouring effect of the underground river on the cave wall will also have a certain impact on the stability of the cave structure when the water flow velocity changes.

[0061] In a possible implementation manner, step S1421 includes:

[0062] Step S1421-1, extract features from each monitoring record in the second template karst cave environment monitoring data to generate an initial feature vector sequence reflecting the changes in the cave environment state.

[0063] In this embodiment, it is assumed that the second template karst cave environment monitoring data is from the monitoring stations in a specific area in the middle of Cave A, and this area includes monitoring points at different positions such as the underground river and the surrounding cave walls and the ground. Each monitoring record contains multiple environmental parameter values, such as the air temperature parameter value, the air humidity parameter value, the carbon dioxide concentration parameter value, the cave water flow velocity parameter value, the number of microbial community parameter value, etc. For each monitoring record, the environmental state change encoding sub-network extracts features from these environmental parameter values. Taking the monitoring record at a certain moment as an example, the air temperature is 15 degrees Celsius, the air humidity is 80%, the carbon dioxide concentration is 0.05%, the cave water flow velocity is 0.5 m / s, and the number of microbial communities is a certain value. The environmental state change encoding sub-network will analyze the relationships between these parameter values and the historical data of this area and the preset standard values, such as the fluctuation range of the air temperature relative to the same period in history, the proportional relationship between the air humidity and the carbon dioxide concentration, the possible correlation between the cave water flow velocity and the number of microbial communities, etc. Through these analyses, an initial feature vector sequence containing multi-dimensional feature information is generated, and this sequence can initially reflect the characteristics of the cave environment state at this moment.

[0064] Step S1421-2: Based on the initial feature vector sequence, use a clustering algorithm to cluster the matched environmental states into the same category. Each category represents an environmental state node. According to the time sequence and the similarity between states, construct the intra-node transition relationship within the environmental state node and the inter-node transition relationship between environmental state nodes, and generate a state transition graph, which describes the dynamic change process of the cave environmental state over time.

[0065] For example, cluster the environmental state with a relatively low air temperature, a relatively high air humidity, a medium carbon dioxide concentration, and a slow cave water flow rate into one category, and this category becomes an environmental state node. According to the time sequence, when the air temperature gradually rises, the air humidity begins to drop, the carbon dioxide concentration fluctuates slightly, and the cave water flow rate increases, these environmental states can be clustered into another category and become another environmental state node. When constructing the intra-node transition relationship, for the states at different times within the same environmental state node, such as within the first-mentioned environmental state node, as time goes by, although the air temperature is low but there is a small increase, the air humidity is basically stable, and the carbon dioxide concentration has a slight fluctuation. This change relationship between the states at different times within the same category is the intra-node transition relationship. And the conversion between different environmental state nodes, such as the conversion from the node with low temperature, high humidity, and low flow rate to the node with high temperature, low humidity, and high flow rate, this conversion relationship is the inter-node transition relationship. The state transition graph constructed in this way details the dynamic change process of the cave environmental state over time, including the conversion order between different environmental states, the duration of each state, and the degree of association between states, etc.

[0066] Step S1421-3: Traverse the state transition graph. Starting from the initial state, according to the time sequence and the state transition relationship, assign a unique code to the environmental state at each time point to generate a state sequence code, and the unique code is used to uniquely identify the environmental state at that time point.

[0067] The initial state may be the environmental state at a certain specific moment, such as the environmental state in the middle area of Cave A in the morning. This state is defined as the initial state and is assigned a unique code, such as code 001. As time goes by, the environmental state at the next time point is assigned the next unique code, such as 002, etc., according to the previously constructed state transition relationship. This unique code can uniquely identify the environmental state at that time point. In this way, each time point's environmental state has a specific identifier, which is convenient for subsequent analysis and processing.

[0068] Step S1421-4: Arrange the state sequence codes in the time sequence to form an environmental state change trajectory.

[0069] These sequentially arranged state sequence encodings can intuitively display the change path of the cave environment state over time. Starting from the initial state encoding, the state encodings at subsequent time points are arranged in sequence to form a complete environmental state change trajectory. This trajectory can reflect how the cave environment changes from one state to another over a period of time, such as the change process from the low-temperature and high-humidity state in the morning to the high-temperature and low-humidity state at noon.

[0070] Step S1421-5, use smoothing filtering to remove the noise and jitter in the environmental state change trajectory, use time series analysis to extract the periodic features and trend features in the environmental state change trajectory, and use pattern recognition to identify the target change pattern or abnormal points in the environmental state change trajectory, so as to obtain the enhanced environmental state change trajectory features.

[0071] In actual monitoring data, due to the accuracy limitations of monitoring instruments or sudden interference factors in the environment, there may be some noise and jitter in the environmental state change trajectory. For example, the monitoring instrument may occasionally have small measurement errors, resulting in unreasonable fluctuations in the air temperature value at a certain time point. Smoothing filtering can remove these unreasonable fluctuations and make the environmental state change trajectory smoother and more stable. Through time series analysis, the periodic features in the cave environment state change trajectory can be found. For example, the air temperature and humidity may show periodic changes of low in the morning, high at noon, and low again in the evening within a day, or there are obvious periodic changes in the cave water flow velocity in different seasons of a year. At the same time, the trend features can also be extracted. For example, as the years increase, the carbon dioxide concentration in the cave has a slow upward trend. In addition, pattern recognition technology can be used to identify the target change pattern or abnormal points. For example, when large-scale rock collapses or underground river diversions occur in the cave, they will be manifested as abnormal points in the environmental state change trajectory, or a special target change pattern such as the simultaneous sharp decline of air humidity and the number of microbial communities within several consecutive time points can be identified. Through these operations, the enhanced environmental state change trajectory features are obtained, which can more accurately reflect the real change situation of the cave environment state and provide a more reliable data basis for subsequent analysis and prediction.

[0072] In a possible implementation manner, the labeled cave environment evolution pattern features of the first template karst cave environment monitoring data include a first candidate cave environment evolution pattern and a first local karst cave space. The first candidate cave environment evolution pattern represents the cave environment evolution pattern corresponding to the labeled cave environment evolution pattern features, and the first local karst cave space represents the karst cave environment partition of the labeled cave environment evolution pattern features in the first template karst cave environment monitoring data.

[0073] Step S130 includes:

[0074] Step S131, using the cave environment risk prediction network generated in the a-th round of network parameter learning task to extract the local state change features of the first local karst cave space, and generating the first local state change features corresponding to the first local karst cave space.

[0075] Step S132, performing spatial encoding representation on the first local karst cave space, and generating the first environmental block space data corresponding to the first local karst cave space.

[0076] Step S133, loading the first local state change features and the first environmental block space data into the cave environment risk prediction network generated in the a-th round of network parameter learning task, and obtaining the first estimated cave environment evolution pattern and the estimated cave environment risk knowledge points corresponding to the first template karst cave environment monitoring data.

[0077] Step S134, based on the third feature loss between the first estimated cave environment evolution pattern and the first candidate cave environment evolution pattern, and the labeled cave environment risk knowledge points and the estimated cave environment risk knowledge points, performing parameter learning on the cave environment risk prediction network generated in the a-th round of network parameter learning task, and generating the cave environment risk prediction network in the (a + 1)-th round of network parameter learning task.

[0078] In a possible implementation manner, Step S131 includes: obtaining a set threshold value, if the first scale corresponding to the first local state change features is not greater than the set threshold value, adding a zero vector with a second scale, and the sum of the first scale and the second scale is not less than the set threshold value.

[0079] In this embodiment, in the context of Karst Cave A, the labeled cave environment evolution pattern features of the first template karst cave environment monitoring data include the first candidate cave environment evolution pattern and the first local karst cave space. Among them, the first candidate cave environment evolution pattern represents the cave environment evolution pattern corresponding to the labeled cave environment evolution pattern features, and the first local karst cave space indicates the karst cave environment partition of the labeled cave environment evolution pattern features in the first template karst cave environment monitoring data.

[0080] Suppose in karst cave A, the first partial karst cave space is the area near the entrance of cave A. The environmental data of this area is included in the first template karst cave environmental monitoring data. The cave environmental risk prediction network begins to analyze this area. For example, the environmental parameter values in the area near the entrance include air temperature, air humidity, carbon dioxide concentration, wind speed, light intensity, etc. The network will analyze these parameter values to extract the characteristics of local state changes. For the air temperature parameter value, it will analyze the temperature change range and change frequency at different time periods. For example, within a certain period of time, from morning to noon, the air temperature rises from 10 degrees Celsius to 15 degrees Celsius, and information such as the rising range and rising rate of this temperature will be used as part of the characteristics of local state changes. For air humidity, if the humidity rises rapidly from 60% to 90% after rainfall, factors such as the numerical value of this sudden change in humidity and the time point of the change will also be taken into account in the consideration of the characteristics of local state changes. At the same time, for parameters such as carbon dioxide concentration, wind speed, and light intensity, the changes in their mutual relationships with other parameters in this local area will also be analyzed. For example, the diffusion situation of carbon dioxide concentration when the wind speed increases, and the relationship between the change in light intensity and the respiration of the microbial community (assuming it exists), which in turn affects the carbon dioxide concentration. In this process, it is also necessary to obtain a set threshold value. This set threshold value is a standard value obtained based on a large amount of previous data analysis or theoretical calculation, and is used to measure the scale of the characteristics of local state changes. If after extracting the characteristics of local state changes in the area near the entrance, it is found that the first scale corresponding to the first local state change characteristic is not greater than the set threshold value, it is necessary to add a zero-value vector of the second scale so that the sum of the first scale and the second scale is not less than the set threshold value. For example, assume that the set threshold value is a certain value, and the scale of the first local state change characteristic obtained through analysis is less than this value, then zero-value vectors will be added according to certain rules to meet the scale requirements. The addition of these zero-value vectors is to ensure the integrity and consistency of the data structure in subsequent calculations and analyses.

[0081] Next, for the first partial karst cave space, which is the area near the entrance of cave A, the spatial coding representation will consider multiple factors. For example, the geographical location information of this area, the orientation of the entrance, altitude, etc. Suppose the entrance faces east and the altitude is 500 meters, and this geographical information will be converted into a coding form. At the same time, the size of the spatial range of this area will also be coded. For example, the area of the area near the entrance is 100 square meters, and this area information will also become part of the coding. In addition, factors such as the connectivity of this area with other areas in the cave will also be considered. If the area near the entrance is connected to a main passage inside the cave, this connectivity relationship will also be coded into the first environmental block space data. By comprehensively coding these factors such as geography, spatial range, and connectivity, the first environmental block space data corresponding to the first partial karst cave space is generated.

[0082] Then, when the first local state change characteristics (including change characteristics of parameters such as air temperature, humidity, and carbon dioxide concentration) and the first environmental partitioned space data (including encoded information such as geographical location, spatial range, and connectivity) in the area near the entrance are loaded into this network, the network will perform calculations according to its predefined algorithm. For example, the network may perform multi-layer analysis and processing of these input data based on machine learning algorithms, such as neural network algorithms. In the first layer of the network, the local state change characteristics and spatial partitioned data of the input may be re-extracted and integrated again, converting different types of data into a unified and more easily analyzable feature representation form. Then, in subsequent layers, operations are performed based on these features and the existing model parameters within the network. After multi-layer calculations, the first estimated cave environment evolution pattern and estimated cave environment risk knowledge points corresponding to the first template karst cave environment monitoring data are obtained. Suppose the first estimated cave environment evolution pattern obtained is: Due to the changes in temperature and humidity in the area near the entrance, as well as the influence of connectivity with the interior of the cave, the biological community structure may be adjusted. Some organisms adapted to low-temperature and low-humidity environments may decrease, while organisms adapted to high-temperature and high-humidity environments may increase. The estimated cave environment risk knowledge points may be: Due to the adjustment of the biological community structure in the area near the entrance, the ecological balance may be affected to a certain extent. For example, some organisms have a protective effect on the cave wall surface (such as secreting special substances to prevent rock weathering), and a decrease in their number may increase the risk of cave wall weathering; at the same time, changes in the biological respiration may affect the carbon dioxide concentration, thereby affecting the chemical environment stability in the cave.

[0083] Finally, for the first estimated cave environment evolution pattern and the first candidate cave environment evolution pattern, for example, the first candidate cave environment evolution pattern is: significant changes in the temperature and humidity in the area near the entrance will cause significant changes in the biological community structure and will lead to obvious fluctuations in the chemical environment inside the cave. Compare this first candidate cave environment evolution pattern with the first estimated cave environment evolution pattern obtained previously and calculate the third feature loss. In terms of comparing the adjustment of the biological community structure, the first candidate pattern emphasizes significant changes, while the first estimated pattern is that adjustments may occur. This difference in description needs to be quantified as part of the feature loss. For the impact on the chemical environment inside the cave, one is obvious fluctuations and the other is the possible impact on stability. This difference also needs to be quantified into the third feature loss. At the same time, for the labeled cave environment risk knowledge points and the estimated cave environment risk knowledge points, the labeled risk knowledge points may clearly indicate the specific probability value that the reduction of a certain organism will lead to an increase in the risk of cave wall weathering, while the estimated risk knowledge points only mention an increased risk. Differences in the quantification of the risk level and the accuracy of the association of risk factors, etc., all need to be quantified. Through these quantified differences, that is, the third feature loss and the differences between the labeled cave environment risk knowledge points and the estimated cave environment risk knowledge points, parameter learning is performed on the cave environment risk prediction network generated in the a-th round of network parameter learning tasks. The network will adjust its internal parameters according to these differences, such as adjusting parameters such as the weight values and bias values in the neural network, so as to optimize the prediction ability of the network and finally generate the cave environment risk prediction network in the (a + 1)-th round of network parameter learning tasks, making this new network more accurate when predicting cave environment risk knowledge points from karst cave environment monitoring data.

[0084] In a possible implementation manner, before step S133, the method further includes:

[0085] Obtain the blended feature data, where the blended feature data is the feature data generated after blending the first local state change feature and the first environmental partition space data.

[0086] In this embodiment, in karst cave A, the first local state change feature includes the changes of various environmental parameters over time in a specific area (such as the area near the entrance of cave A), such as the fluctuation range of air temperature, the increasing or decreasing trend of air humidity, the change rate of carbon dioxide concentration, etc.; the first environmental block space data reflects the spatial attributes of this specific area, such as the geographical location (latitude, longitude, altitude, etc.), the size of the spatial range, the connectivity with other areas, etc. The fusion feature data is the feature data generated after the fusion of the first local state change feature and the first environmental block space data. This fusion is not a simple combination, but based on specific algorithms and rules. For example, for the area near the entrance, the change in air temperature may be related to the altitude of this area, and the fusion process will reflect this relationship in a quantitative way. Specifically, according to the physical principle or the empirical formula obtained from historical data statistics, the influence coefficient of altitude on air temperature is multiplied by the actual change value of air temperature, and then comprehensively calculated with the change values of other similarly processed environmental parameters and spatial attribute values to generate the fusion feature data. This fusion feature data contains the internal connection information between the environmental state change and the spatial attributes.

[0087] Step S133 includes: loading the fusion feature data, the first local state change feature, and the first environmental block space data into the cave environment risk prediction network generated in the a-th round of network parameter learning tasks to obtain the first estimated cave environment evolution pattern and the estimated cave environment risk knowledge points.

[0088] Taking a neural network structure as an example, when inputting the fused feature data, the first local state change feature, and the first environmental block space data, the input layer of the network first receives these data. In the first hidden layer of the network, neurons perform weighted summation operations on these input data. Each neuron has its own weight, and these weights are determined during the previous training or initialization process of the network. For example, for a feature value in the fused feature data that is associated with air temperature and altitude, it may be given a higher weight by a neuron because this feature value may be significant in predicting the risk of the cave environment. After the weighted summation in the first hidden layer, a non-linear transformation is performed through an activation function (such as the ReLU function) to obtain a new feature representation. These new feature representations will continue to perform similar operations in subsequent hidden layers, continuously extracting and integrating features. In the last layer of the network, that is, the output layer, the first estimated cave environment evolution pattern and the estimated cave environment risk knowledge points are generated based on the calculation results of the previous layers. The first estimated cave environment evolution pattern may be described as follows: in the area near the entrance of Cave A, due to the change trend of air temperature (obtained from the temperature information in the first local state change feature), the spatial attributes of this area (obtained from the first environmental block space data), and the fusion relationship between the two (reflected by the fused feature data), the biological community structure will adjust in the direction of adapting to temperature changes and the spatial environment. The number of some microbial communities adapted to low-temperature environments may decrease, while the number of microbial communities adapted to higher temperatures and specific spatial layouts (such as areas near ventilation openings where microorganisms are more likely to obtain oxygen, etc.) may increase. The estimated cave environment risk knowledge points may include that this adjustment of the biological community structure may affect the material cycle in the cave. For example, the organic matter decomposition process participated by microorganisms may be affected, thereby affecting the chemical environment stability in the cave; at the same time, the protective effect of the biological community on the cave wall (such as substances secreted by some microorganisms can inhibit rock weathering) may change due to the change of the community structure, increasing the risk of cave wall weathering.

[0089] In a possible implementation manner, before step S150, the method further includes:

[0090] Step A110, performing a first feature selection on the first cave environment evolution pattern feature and the third cave environment evolution pattern feature to generate first selected feature data.

[0091] In the example of karst cave A, both the first cave environmental evolution pattern feature and the third cave environmental evolution pattern feature contain complex information about the environmental evolution of different regions of the cave. The first feature selection is carried out based on specific selection criteria and algorithms. For example, the first cave environmental evolution pattern feature describes the environmental evolution of a certain area in the cave (such as the area near the underground river in the middle of cave A), including various aspects such as the impact of changes in the underground river water flow velocity on the surrounding biological community and the impact of changes in the cave wall temperature on the distribution of the microbial community; the third cave environmental evolution pattern feature describes the environmental evolution of another area in the cave (such as a specific area deep in cave A), involving aspects such as the impact of changes in oxygen concentration on biological respiration and the impact of changes in rock temperature on the stability of the cave structure. The first feature selection may screen according to the importance of each feature to the overall environmental evolution of the cave. For example, features directly related to changes in biodiversity (such as changes in the biological community structure, changes in key environmental parameters for the survival of organisms, etc.) will be preferentially selected, while some features with less impact on the overall environmental evolution or with a high degree of correlation with other features (which can be obtained through correlation analysis) may be excluded. In this way, the first selected feature data is generated, and this data set contains feature information that is crucial for predicting the cave environmental evolution pattern.

[0092] Step A120, load the blended feature data, the first local state change feature, and the first environmental block space data into the cave environmental risk prediction network generated in the (a + 1)-th round of network parameter learning task to obtain the target estimated cave environmental evolution pattern.

[0093] This process is similar to the cave environment risk prediction network generated by loading this data into the network parameter learning task of the a-th round before. However, since it is in the network parameter learning task of the (a + 1)-th round, the parameters of the network have been adjusted and optimized to a certain extent. When the input is the blended feature data, the first local state change feature, and the first environmental block space data, the network calculates according to its internal algorithm structure. For example, the weights in the network may have been updated according to the previous learning tasks, and the connection strength between neurons has changed, so that the network's processing method for the input data is different. After multiple layers of calculation by the network, the target estimated cave environment evolution pattern is obtained. This target estimated cave environment evolution pattern may be described as an estimate of the environment evolution pattern in a specific area of the cave (such as the area near the entrance mentioned above or other relevant areas) after considering the environment-space relationship in the blended feature data, the regional environmental dynamic changes in the first local state change feature, and the spatial attributes in the first environmental block space data. For example, in the area near the entrance, the target estimated cave environment evolution pattern may indicate that due to the interaction between a certain spatial factor not considered before (such as the special impact of the entrance shape on the air flow) and the environmental state changes (such as the seasonal changes in air temperature and humidity), the adjustment range and direction of the biological community structure are different from the previous estimates, and the degree of impact of this adjustment on the material cycle and chemical environment stability in the cave has also changed.

[0094] Step A130: Match the first selected feature data with the target estimated cave environment evolution pattern to generate second selected feature data that matches the target estimated cave environment evolution pattern.

[0095] In this embodiment, this matching process is to further screen and optimize the feature information related to the target estimated cave environment evolution pattern. For example, some features in the first selected feature data are directly related to the description in the target estimated cave environment evolution pattern. For example, the target estimated cave environment evolution pattern mentions the impact of the adjustment of the biological community structure on the material cycle, and the first selected feature data contains features related to the biological community structure. Then these relevant features will be screened out as the second selected feature data. This second selected feature data focuses more on the key features that match the target estimated cave environment evolution pattern and can be more effectively used in the subsequent parameter learning process.

[0096] Step S150 includes: Based on the first feature loss between the labeled cave environment evolution pattern features and the second cave environment evolution pattern features, and the second feature loss between the second selected feature data and the third cave environment evolution pattern features, perform parameter learning on the basic cave environment evolution pattern prediction network to generate the cave environment evolution pattern prediction network in the network parameter learning task of the (a + 1)-th round.

[0097] In this embodiment, the labeled cave environment evolution pattern features contain accurate labeling information of the cave environment evolution pattern, and the second cave environment evolution pattern features are an estimation of the cave environment evolution pattern obtained through network prediction. Calculate the first feature loss between them. For example, in the labeled cave environment evolution pattern features, it is pointed out that the change in the biological community structure in a certain area will lead to an accurate change range of the concentration of a specific substance (such as carbon dioxide) within a certain period of time, while there is a difference in the predicted change range of this substance concentration in the second cave environment evolution pattern features, and this difference will be quantified as part of the first feature loss. At the same time, for the second feature loss between the second selected feature data and the third cave environment evolution pattern features, the key features in the second selected feature data are compared with the relevant descriptions in the third cave environment evolution pattern features. For example, the feature of the protection of the cave wall by the biological community in the second selected feature data is related to the description of the stability of the cave wall in the third cave environment evolution pattern features. If there is an inconsistency between the two (such as the second selected feature data believes that the protection of the biological community is strong, while the third cave environment evolution pattern features show a decrease in the stability of the cave wall), this inconsistency will be quantified as part of the second feature loss. Based on these first feature losses and second feature losses, parameter learning is performed on the basic cave environment evolution pattern prediction network. The network will adjust its internal parameters according to these loss values, such as adjusting the connection weights and bias values between neurons. If the first feature loss indicates a large deviation in the prediction of the change in substance concentration caused by the change in biological community structure, the network may increase the connection weights of the neurons related to the biological community and substance concentration to more accurately reflect this relationship in subsequent predictions. By continuously adjusting the parameters until the network converges, the cave environment evolution pattern prediction network in the (a + 1)-th round of network parameter learning tasks is generated, so that it can more accurately reflect the actual situation when predicting the cave environment evolution pattern of the karst cave environment monitoring data.

[0098] In a possible implementation, the labeled cave environment evolution pattern features of the first template karst cave environment monitoring data include a first candidate cave environment evolution pattern and a first local karst cave space. The first candidate cave environment evolution pattern represents the cave environment evolution pattern corresponding to the labeled cave environment evolution pattern features, and the first local karst cave space represents the karst cave environment partition of the labeled cave environment evolution pattern features in the first template karst cave environment monitoring data. The first cave environment evolution pattern features include a first estimated cave environment evolution pattern and a first estimated local karst cave space. The second cave environment evolution pattern features include a second estimated cave environment evolution pattern and a second estimated local karst cave space. The third cave environment evolution pattern features include a third estimated cave environment evolution pattern and a third estimated local karst cave space.

[0099] Step S150 may further include:

[0100] Generate the first feature loss based on the first sub-feature loss between the first candidate cave environment evolution pattern and the second estimated cave environment evolution pattern, and the second sub-feature loss between the first local karst cave space and the second estimated local karst cave space.

[0101] Generate the second feature loss based on the third sub-feature loss between the first estimated cave environment evolution pattern and the third estimated cave environment evolution pattern, and the fourth sub-feature loss between the first estimated local karst cave space and the third estimated local karst cave space.

[0102] Perform parameter learning on the basic cave environment evolution pattern prediction network based on the first feature loss and the second feature loss to generate the cave environment evolution pattern prediction network in the (a + 1)-th round of network parameter learning tasks.

[0103] In karst cave A, assume that the first candidate cave environmental evolution model indicates that in the area near the entrance of cave A (i.e., the first local karst cave space), due to seasonal climate changes and air flow exchange inside and outside the cave, the air temperature and humidity will show periodic fluctuations. Such fluctuations will cause corresponding periodic changes in the microbial community structure. For example, the number of some microorganisms adapted to low temperature and high humidity environments will increase in winter, while the number of microorganisms adapted to high temperature and low humidity environments will increase in summer. At the same time, the change in the microbial community structure will affect the chemical substance cycle in the cave, such as the production and consumption rates of carbon dioxide. The second estimated cave environmental evolution model predicts that in the area near the entrance, although the change in the microbial community structure is related to temperature and humidity, the cycle and amplitude of the change are different from those of the first candidate cave environmental evolution model. For example, it is predicted that the increase amplitude of the number of microorganisms adapted to low temperature and high humidity environments in winter is smaller than the actually marked value. Calculating this difference involves quantitative comparisons in multiple aspects such as the change amplitude and cycle of the microbial number, so as to obtain the first sub-feature loss.

[0104] Regarding the second sub-feature loss between the first local karst cave space and the second estimated local karst cave space, assume that the first local karst cave space accurately defines the scope of the area near the entrance, including its boundaries with other areas, specific geographical coordinate ranges, etc. The second estimated local karst cave space may have deviations from the actual scope in the prediction. For example, the boundary definition is inaccurate or some areas that should not be included are included. By comparing the differences between the two in terms of spatial scope definition, geographical location accuracy, etc., these differences are quantified to generate the second sub-feature loss. Combining the first sub-feature loss and the second sub-feature loss, the first feature loss is obtained.

[0105] For the generation of the second feature loss, it is based on the third sub-feature loss between the first estimated cave environment evolution pattern and the third estimated cave environment evolution pattern, as well as the fourth sub-feature loss between the first estimated local karst cave space and the third estimated local karst cave space. For example, the first estimated cave environment evolution pattern describes the environmental evolution of the area near the underground river in the middle of Cave A (i.e., the first estimated local karst cave space), believing that the change in the water flow velocity of the underground river will cause changes in the surrounding soil moisture, which in turn affects the growth state of a small number of special plants (if any) growing in this area, and the change in the growth state of this plant will have an impact on the surrounding microbial community, changing the distribution and species composition of the microbial community. In the prediction of the same area by the third estimated cave environment evolution pattern, although it also believes that the underground river water flow velocity affects soil moisture and plant growth, there are differences in the degree of influence on plant growth state and the specific influence method on the microbial community compared with the first estimated cave environment evolution pattern. By quantifying these differences, such as comparing the prediction differences of plant growth indicators (such as height, number of leaves, etc.) and the prediction differences in the change of microbial community species composition, the third sub-feature loss is obtained.

[0106] For the fourth sub-feature loss between the first estimated local karst cave space and the third estimated local karst cave space, assume that the first estimated local karst cave space accurately describes the detailed spatial information of the area near the underground river in the middle, including the distance range from the underground river, the height of the terrain, etc. However, there are deviations in the description of these spatial information by the third estimated local karst cave space, such as inaccurate definition of the distance range from the underground river. By comparing and quantifying the differences in these spatial information, the fourth sub-feature loss is obtained. Combining the third sub-feature loss and the fourth sub-feature loss, the second feature loss is generated.

[0107] Based on the first feature loss and the second feature loss, parameter learning is performed on the basic cave environment evolution pattern prediction network to generate the cave environment evolution pattern prediction network in the (a + 1)-th round of network parameter learning tasks. The basic cave environment evolution pattern prediction network contains numerous parameters, such as weights and biases in neural networks, etc. If the first feature loss indicates that there are significant deviations in the prediction of the change in the microbial community structure in the prediction of the environment evolution pattern in the area near the entrance (such as a large first sub-feature loss), the network will adjust the weights of the input parameters related to the microbial community, such as the weights related to the number of microbial communities, air temperature, humidity, etc. If the second feature loss shows that there are deviations in the prediction of the impact of the underground river water flow velocity on plant and microbial communities in the prediction of the environment evolution pattern in the area near the underground river in the middle (such as a large third sub-feature loss), the network will adjust the weights of the input parameters related to the underground river water flow velocity, soil humidity, plant growth indicators, etc. By continuously adjusting the network parameters according to the first feature loss and the second feature loss until the network convergence requirement is met, at this time, the cave environment risk prediction network and the cave environment evolution pattern prediction network that have completed parameter learning are generated.

[0108] In a possible implementation manner, after step S150, it further includes:

[0109] Step S160, obtaining the target karst cave environment monitoring data.

[0110] Step S170, loading the target karst cave environment monitoring data into the cave environment evolution pattern prediction network and the cave environment risk prediction network that have completed parameter learning, to obtain the target cave environment evolution pattern features corresponding to the target karst cave environment monitoring data and the cave environment risk knowledge points corresponding to the target cave environment evolution pattern features.

[0111] After completing parameter learning, obtain the target karst cave environment monitoring data. This target karst cave environment monitoring data can come from a new monitoring area of karst cave A or data obtained by using new monitoring technologies or more refined monitoring frequencies for existing monitoring areas. For example, new monitoring stations may be set up deep in cave A, or the monitoring frequency in the area near the entrance may be increased from once per hour to once every half hour.

[0112] Load the target karst cave environmental monitoring data into the cave environmental evolution pattern prediction network and the cave environmental risk prediction network that have completed parameter learning, and obtain the target cave environmental evolution pattern features corresponding to the target karst cave environmental monitoring data and the cave environmental risk knowledge points corresponding to the target cave environmental evolution pattern features. For each monitoring record in the target karst cave environmental monitoring data, it contains multiple environmental parameter values, such as air temperature, humidity, carbon dioxide concentration, etc. When loaded into the cave environmental evolution pattern prediction network, the network processes these data according to the parameters it has learned. For example, if the target karst cave environmental monitoring data is from a new monitoring site deep in Cave A, the network will predict the target cave environmental evolution pattern features based on the special environmental conditions deep inside (such as lower temperature, higher carbon dioxide concentration, etc.) and the rules of the cave environmental evolution pattern learned previously. This feature may be described as deep in Cave A, due to the lower temperature and higher carbon dioxide concentration, the microbial community structure shows a special distribution state, and this state may be affected by the change of rock temperature over time.

[0113] At the same time, load the target karst cave environmental monitoring data into the cave environmental risk prediction network, and obtain the cave environmental risk knowledge points corresponding to the target cave environmental evolution pattern features. For example, based on the predicted special distribution state of the microbial community structure and the possible change trend, the cave environmental risk prediction network may conclude that deep in Cave A, this change in the microbial community structure may pose a risk to the survival of some rare microorganisms, thereby affecting the stability of the material cycle in the cave; or due to the interaction between the microbial community and the rock (such as the erosion or protection of the rock by the microorganisms), it may increase the risk of instability of the cave wall structure and other cave environmental risk knowledge points.

[0114] Figure 2 FIG. shows the hardware structure diagram of the artificial intelligence-based karst cave environmental data analysis system 100 provided by the embodiment of the present invention for implementing the above-mentioned artificial intelligence-based karst cave environmental data analysis method, as Figure 2 shown, the artificial intelligence-based karst cave environmental data analysis system 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

[0115] The machine-readable storage medium 120 can store data and / or instructions. In some embodiments, the machine-readable storage medium 120 can store data obtained from an external terminal. In some embodiments, the machine-readable storage medium 120 can store the data and / or instructions used by the artificial intelligence-based karst cave environmental data analysis system 100 to execute or use to complete the exemplary methods described in the present invention.

[0116] In a specific implementation process, one or more processors 110 execute computer-executable instructions stored in a machine-readable storage medium 120, enabling the processors 110 to execute the artificial intelligence-based karst cave environment data analysis method in the above method embodiments. The processors 110, the machine-readable storage medium 120, and the communication unit 140 are connected through a bus 130, and the processors 110 can be used to control the transceiver actions of the communication unit 140.

[0117] For the specific implementation process of the processors 110, reference can be made to the respective method embodiments executed by the above artificial intelligence-based karst cave environment data analysis system 100. Their implementation principles and technical effects are similar, and will not be elaborated here in this embodiment.

[0118] In addition, an embodiment of the present invention also provides a readable storage medium, in which computer-executable instructions are preset. When a processor executes the computer-executable instructions, the artificial intelligence-based karst cave environment data analysis method as described above is implemented.

[0119] It should be noted that, in order to simplify the description of the present invention disclosure and thus help the understanding of one or more embodiments of the invention, in the previous description of the embodiments of the present invention, multiple features are sometimes merged into one embodiment, drawing, or description thereof.

Claims

1. A karst cave environment data analysis method based on artificial intelligence, characterized in that: The method comprises: Acquire first template karst cave environment monitoring data and second template karst cave environment monitoring data, wherein the first template karst cave environment monitoring data carries marked cave environment evolution pattern characteristics and marked cave environment risk knowledge points corresponding to the marked cave environment evolution pattern characteristics, and the second template karst cave environment monitoring data carries marked cave environment risk knowledge points corresponding to the marked cave environment evolution pattern characteristics, and the first template karst cave environment monitoring data and the second template karst cave environment monitoring data both contain monitoring records arranged in chronological order, and each monitoring record contains multiple environmental parameter values, and the environmental parameter values ​​include one or more combinations of air temperature parameter values, air humidity parameter values, carbon dioxide concentration parameter values, air pressure parameter values, cave internal structure change parameter values, oxygen concentration parameter values, wind speed parameter values, light intensity parameter values, methane concentration parameter values, hydrogen sulfide concentration parameter values, cave water flow velocity parameter values, cave water pH parameter values, soil moisture parameter values, rock temperature parameter values, radon gas concentration parameter values, microbial community quantity parameter values, cave wall roughness parameter values, cave top dripping velocity parameter values, negative ion concentration parameter values, and cave dust concentration parameter values; In the a+1th round of network parameter learning task, the cave environment evolution pattern prediction network generated in the ath round of network parameter learning task is used to predict the second template karst cave environment monitoring data to generate a first cave environment evolution pattern feature, wherein the cave environment evolution pattern prediction network is used to estimate the cave environment evolution pattern feature in the karst cave environment monitoring data, and a is a positive integer; Using the first template karst cave environment monitoring data, parameter learning is performed on the cave environment risk prediction network generated in the a-th round of network parameter learning tasks to generate a cave environment risk prediction network in the a+1-th round of network parameter learning tasks, wherein the cave environment risk prediction network is used to predict cave environment risk knowledge points of cave environment evolution pattern characteristics based on the karst cave environment monitoring data; Using the basic cave environment evolution pattern prediction network in the a+1th round of network parameter learning task and the cave environment risk prediction network generated in the a+1th round of network parameter learning task, the first template karst cave environment monitoring data and the second template karst cave environment monitoring data are predicted to generate second cave environment evolution pattern features corresponding to the first template karst cave environment monitoring data and third cave environment evolution pattern features corresponding to the second template karst cave environment monitoring data; According to the first feature loss between the labeled cave environment evolution pattern feature and the second cave environment evolution pattern feature, as well as the second feature loss between the first cave environment evolution pattern feature and the third cave environment evolution pattern feature, parameter learning is performed on the basic cave environment evolution pattern prediction network to generate a cave environment evolution pattern prediction network in the a+1th round of network parameter learning task, until the network convergence requirements are met to generate a cave environment risk prediction network and a cave environment evolution pattern prediction network that have completed parameter learning.

2. The karst cave environment data analysis method based on artificial intelligence according to claim 1 is characterized in that: The method uses the basic cave environment evolution pattern prediction network in the a+1th round of network parameter learning task and the cave environment risk prediction network generated in the a+1th round of network parameter learning task to predict the first template karst cave environment monitoring data and the second template karst cave environment monitoring data, and generates the second cave environment evolution pattern characteristics corresponding to the first template karst cave environment monitoring data and the third cave environment evolution pattern characteristics corresponding to the second template karst cave environment monitoring data, including: Using the basic cave environment evolution pattern prediction network in the a+1th round of network parameter learning task to predict the first template karst cave environment monitoring data, generating the second cave environment evolution pattern features corresponding to the first template karst cave environment monitoring data; The second template karst cave environment monitoring data is predicted using the basic cave environment evolution pattern prediction network in the a+1th round of network parameter learning tasks and the cave environment risk prediction network generated in the a+1th round of network parameter learning tasks to generate the third cave environment evolution pattern characteristics corresponding to the second template karst cave environment monitoring data.

3. The karst cave environment data analysis method based on artificial intelligence according to claim 2 is characterized in that: The basic cave environment evolution pattern prediction network includes an environment state change encoding subnetwork and a fully connected mapping subnetwork; The method of using the basic cave environment evolution pattern prediction network in the a+1th round of network parameter learning task and the cave environment risk prediction network generated in the a+1th round of network parameter learning task to predict the second template karst cave environment monitoring data to generate the third cave environment evolution pattern features corresponding to the second template karst cave environment monitoring data includes: Using the environmental state change encoding subnetwork to encode the environmental state change of the second template karst cave environmental monitoring data, and generate environmental state change trajectory features corresponding to the second template karst cave environmental monitoring data; Using the cave environment risk prediction network generated in the a+1th round of network parameter learning task, extracting local state change features from the environmental state change trajectory features corresponding to the second template karst cave environment monitoring data, and generating second local state change features corresponding to the annotated cave environment evolution pattern features; Integrating the environmental state change trajectory features and the local state change features to generate enhanced trajectory features; Loading the enhanced trajectory features into the fully connected mapping subnetwork to generate the third cave environment evolution pattern features corresponding to the second template karst cave environment monitoring data; The step of encoding the environmental state change of the second template karst cave environment monitoring data using the environmental state change encoding subnetwork to generate environmental state change trajectory features corresponding to the second template karst cave environment monitoring data includes: Extracting features from each monitoring record in the second template karst cave environment monitoring data to generate an initial feature vector sequence reflecting changes in the cave environment state; Based on the initial feature vector sequence, a clustering algorithm is used to cluster the matching environmental states into the same category, each category represents an environmental state node, and according to the time sequence and the similarity between the states, the intra-node transfer relationship within the environmental state node and the inter-node transfer relationship between the environmental state nodes are constructed to generate a state transition diagram, which describes the dynamic change process of the cave environmental state over time; Traversing the state transition diagram, starting from the initial state, assigning a unique code to the environment state at each time point according to the time sequence and the state transition relationship, and generating a state sequence code, wherein the unique code is used to uniquely identify the environment state at the time point; Arranging the state sequence codes in chronological order to form an environmental state change trajectory; Smoothing filtering is used to remove noise and jitter in the environmental state change trajectory, and time series analysis is used to extract periodic characteristics and trend characteristics in the environmental state change trajectory. Pattern recognition is used to identify target change patterns or abnormal points in the environmental state change trajectory to obtain enhanced environmental state change trajectory characteristics.

4. The karst cave environment data analysis method based on artificial intelligence according to any one of claims 1 to 3, characterized in that: The annotated cave environment evolution pattern characteristics of the first template karst cave environment monitoring data include a first candidate cave environment evolution pattern and a first local karst cave space, wherein the first candidate cave environment evolution pattern represents the cave environment evolution pattern corresponding to the annotated cave environment evolution pattern characteristics, and the first local karst cave space represents the karst cave environment partition of the annotated cave environment evolution pattern characteristics in the first template karst cave environment monitoring data; The method of using the first template karst cave environment monitoring data to perform parameter learning on the cave environment risk prediction network generated in the a-th round of network parameter learning task to generate the cave environment risk prediction network in the a+1-th round of network parameter learning task includes: Using the cave environment risk prediction network generated in the a-th round of network parameter learning task, extracting local state change features of the first local karst cave space to generate first local state change features corresponding to the first local karst cave space; Performing spatial coding representation on the first local karst cave space to generate first environmental block spatial data corresponding to the first local karst cave space; Loading the first local state change feature and the first environmental block spatial data into the cave environment risk prediction network generated in the a-th round of network parameter learning task, obtaining a first estimated cave environment evolution pattern and an estimated cave environment risk knowledge point corresponding to the first template karst cave environment monitoring data; According to the third feature loss between the first estimated cave environment evolution pattern and the first candidate cave environment evolution pattern, as well as the labeled cave environment risk knowledge points and the estimated cave environment risk knowledge points, parameter learning is performed on the cave environment risk prediction network generated in the a-th round of network parameter learning task to generate the cave environment risk prediction network in the a+1-th round of network parameter learning task.

5. The karst cave environment data analysis method based on artificial intelligence according to claim 4 is characterized in that: The method of extracting local state change features of the first local karst cave space using the cave environment risk prediction network generated in the a-th round of network parameter learning task to generate first local state change features corresponding to the first local karst cave space includes: Get the set threshold value; If the first scale corresponding to the first local state change feature is not greater than the set threshold value, a zero-value vector of a second scale is added, and the sum of the first scale and the second scale is not less than the set threshold value.

6. The karst cave environment data analysis method based on artificial intelligence according to claim 4 is characterized in that: Before the first local state change feature and the first environmental block spatial data are loaded into the cave environment risk prediction network generated in the a-th round network parameter learning task, and a first estimated cave environment evolution pattern and an estimated cave environment risk knowledge point corresponding to the first template karst cave environment monitoring data are obtained, the method further includes: Acquire blending feature data, where the blending feature data is feature data generated by blending the first local state change feature and the first environmental block spatial data; The step of loading the first local state change feature and the first environmental block spatial data into the cave environment risk prediction network generated in the a-th round of network parameter learning task to obtain a first estimated cave environment evolution pattern and estimated cave environment risk knowledge points corresponding to the first template karst cave environment monitoring data includes: The blending feature data, the first local state change feature and the first environmental block spatial data are loaded into the cave environment risk prediction network generated in the a-th round of network parameter learning task to obtain the first estimated cave environment evolution pattern and the estimated cave environment risk knowledge points.

7. The karst cave environment data analysis method based on artificial intelligence according to claim 6 is characterized in that: Before performing parameter learning on the basic cave environment evolution pattern prediction network based on the first feature loss between the labeled cave environment evolution pattern feature and the second cave environment evolution pattern feature, and the second feature loss between the first cave environment evolution pattern feature and the third cave environment evolution pattern feature, and generating the cave environment evolution pattern prediction network in the a+1th round of network parameter learning task, the method further includes: Performing first feature selection on the first cave environment evolution pattern feature and the third cave environment evolution pattern feature to generate first selected feature data; Loading the blending feature data, the first local state change feature and the first environmental block spatial data into the cave environment risk prediction network generated in the a+1th round of network parameter learning task to obtain a target estimated cave environment evolution pattern; Matching the first selected feature data with the target estimated cave environment evolution pattern to generate second selected feature data matching the target estimated cave environment evolution pattern; The method of performing parameter learning on the basic cave environment evolution pattern prediction network based on the first feature loss between the labeled cave environment evolution pattern feature and the second cave environment evolution pattern feature, and the second feature loss between the first cave environment evolution pattern feature and the third cave environment evolution pattern feature, to generate the cave environment evolution pattern prediction network in the a+1th round of network parameter learning task, includes: According to the first feature loss between the labeled cave environment evolution pattern feature and the second cave environment evolution pattern feature, as well as the second feature loss between the second selected feature data and the third cave environment evolution pattern feature, parameter learning is performed on the basic cave environment evolution pattern prediction network to generate a cave environment evolution pattern prediction network in the a+1th round of network parameter learning task.

8. The karst cave environment data analysis method based on artificial intelligence according to any one of claims 1 to 3, characterized in that: The annotated cave environment evolution pattern characteristics of the first template karst cave environment monitoring data include a first candidate cave environment evolution pattern and a first local karst cave space, the first candidate cave environment evolution pattern represents the cave environment evolution pattern corresponding to the annotated cave environment evolution pattern characteristics, the first local karst cave space represents the karst cave environment partition of the annotated cave environment evolution pattern characteristics in the first template karst cave environment monitoring data, the first cave environment evolution pattern characteristics include a first estimated cave environment evolution pattern and a first estimated local karst cave space, the second cave environment evolution pattern characteristics include a second estimated cave environment evolution pattern and a second estimated local karst cave space, and the third cave environment evolution pattern characteristics include a third estimated cave environment evolution pattern and a third estimated local karst cave space; The method of performing parameter learning on the basic cave environment evolution pattern prediction network based on the first feature loss between the labeled cave environment evolution pattern feature and the second cave environment evolution pattern feature, and the second feature loss between the first cave environment evolution pattern feature and the third cave environment evolution pattern feature, to generate the cave environment evolution pattern prediction network in the a+1th round of network parameter learning task, includes: generating the first feature loss according to the first sub-feature loss between the first candidate cave environment evolution pattern and the second estimated cave environment evolution pattern, and the second sub-feature loss between the first local karst cave space and the second estimated local karst cave space; generating the second characteristic loss according to the third sub-characteristic loss between the first estimated cave environment evolution pattern and the third estimated cave environment evolution pattern, and the fourth sub-characteristic loss between the first estimated local karst cave space and the third estimated local karst cave space; Based on the first feature loss and the second feature loss, parameter learning is performed on the basic cave environment evolution pattern prediction network to generate the cave environment evolution pattern prediction network in the a+1th round of network parameter learning task.

9. The karst cave environment data analysis method based on artificial intelligence according to any one of claims 1 to 3, characterized in that: The method further comprises: performing parameter learning on the basic cave environment evolution pattern prediction network according to the first feature loss between the labeled cave environment evolution pattern feature and the second cave environment evolution pattern feature, and the second feature loss between the first cave environment evolution pattern feature and the third cave environment evolution pattern feature, generating a cave environment evolution pattern prediction network in the a+1th round of network parameter learning task, until the cave environment risk prediction network and the cave environment evolution pattern prediction network with completed parameter learning are generated when the network convergence requirements are met, and further comprising: Obtaining environmental monitoring data of target karst caves; The target karst cave environment monitoring data is loaded into the cave environment evolution pattern prediction network and the cave environment risk prediction network that have completed parameter learning, and the target cave environment evolution pattern characteristics corresponding to the target karst cave environment monitoring data and the cave environment risk knowledge points corresponding to the target cave environment evolution pattern characteristics are obtained.

10. A karst cave environment data analysis system based on artificial intelligence, characterized in that: The artificial intelligence-based karst cave environment data analysis system includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the artificial intelligence-based karst cave environment data analysis method described in any one of claims 1 to 9 above.

Images