A mine filling retaining wall pressure field real-time prediction method and system

Abstract

The present application provides a kind of mine filling retaining wall pressure field real-time prediction method and system, belong to mine filling engineering safety monitoring and intelligent prediction technical field, including: scanning typical stope three-dimensional model, the real-time filling flow multi-source monitoring data of key point position and retaining wall pressure are collected, and field time series dataset is constructed;Based on three-dimensional model and multi-source monitoring data, simulate the filling process, trace back the filling time of paste body at any height position, in-situ property and cumulative maintenance time, generate dynamic layered attribute characteristics;Construct training sample with dynamic layered attribute characteristics as input and retaining wall pressure as label;Neural ordinary differential equation is used to establish a spatial continuous prediction model of retaining wall pressure;Supervised training is carried out using training samples;For other target stope, generate dynamic layered attribute characteristics input into the trained prediction model, and continuously predict the pressure in the full height range of the retaining wall.The present application can real-time predict the mine filling retaining wall pressure field.

Description

Technical Field

[0001] This invention belongs to the field of safety monitoring and intelligent prediction technology for mine backfilling projects, and specifically refers to a method and system for real-time prediction of pressure field of mine backfilling retaining walls. Background Technology

[0002] In modern underground mining, backfilling mining methods are widely used due to their effectiveness in controlling ground pressure, reducing surface subsidence, and improving ore recovery. In cemented backfilling stopes, backfill paste is transported underground via pipelines and gradually forms a backfill body under the constraint of retaining walls. As a key load-bearing component, the stress state of the retaining walls directly affects the safety of backfilling operations. Therefore, accurate, continuous, and real-time prediction of retaining wall pressure is a key technical issue in intelligent mine backfilling.

[0003] However, the formation and evolution of retaining wall pressure are significantly complex. On the one hand, retaining wall pressure is not only determined by the filling height, but also influenced by multiple factors such as filling flow rate, paste concentration, cement-sand ratio, downtime, paste formation sequence, and stoppage geometry. On the other hand, under conditions of multi-stage filling and intermittent downtime, the paste at different heights has different filling times and formation durations, resulting in a non-uniform spatial distribution of the paste. This spatiotemporal coupling effect makes the retaining wall pressure distribution dynamic and uncertain.

[0004] Currently, engineering sites mainly rely on three types of methods: empirical formulas, numerical simulation, and traditional data-driven methods. Empirical formulas are simple in form but difficult to reflect changes in working conditions; numerical simulation has a long modeling cycle and cannot meet the needs of real-time prediction; traditional data-driven methods usually ignore spatial continuity and differences in the process of paste formation, resulting in insufficient ability to characterize the continuous evolution of pressure during the shutdown phase. Summary of the Invention

[0005] To address the technical problems existing in the prior art, the present invention provides a method and system for real-time prediction of the pressure field of mine backfill retaining walls, the technical solution of which is as follows: On the one hand, a method for real-time prediction of the pressure field of a mine backfill retaining wall is provided, the method comprising: S1. For a small number of typical mining sites, scan their three-dimensional models and collect real-time filling flow multi-source monitoring data of key points, as well as retaining wall pressure of key points, to construct a field time series dataset of retaining wall pressure of key points. The real-time filling flow multi-source monitoring data of key points includes the filling time of paste, in-situ attributes, and cumulative curing time. S2. Based on the three-dimensional model and real-time filling flow multi-source monitoring data, the filling process is continuously simulated. For any predicted height position in the mining area, the filling time, in-situ attributes and cumulative curing time of the paste at any predicted height position are traced back to generate dynamic layered attribute features. The dynamic layered attribute features include height position, filling time, in-situ attributes and cumulative curing time. S3. Based on the aforementioned on-site time-series dataset and dynamic hierarchical attribute features, construct training samples with the dynamic hierarchical attribute features of key points as input and the retaining wall pressure of key points as labels. S4. A continuous prediction model for the pressure space of the retaining wall is established using the constant differential equation of the God: with height as the continuous independent variable and hidden state variables as the evolution object, the rate of change of the hidden state variables with respect to height is parameterized by the neural network. Boundary reference conditions are given at the liquid level. The hidden state variables at each height are obtained by integration along the height direction. Then, the hidden state variables are mapped to the pressure of the retaining wall through the mapping network from the hidden state variables to the pressure, so as to realize the continuous prediction of the pressure field along the height. S5. Supervised training is performed on the prediction model using the training samples to obtain the trained prediction model. S6. For the online pressure field prediction stage of other target mining areas, dynamic hierarchical attribute features of key points are generated based on the current collected data and input into the trained prediction model to continuously predict the pressure across the entire height range of the retaining wall.

[0006] Optionally, the in-situ attributes refer to attribute information directly related to the state of the paste corresponding to the target height position, including the paste concentration, ash-sand ratio, filling batch parameters corresponding to the paste filling time at that position, and the mining section or structural features related to that position.

[0007] Optionally, the cumulative curing time refers to the time elapsed since the paste entered a certain height position until the current moment, used to characterize the differences in the formation sequence of the paste and its state evolution process: For pastes in the solidification and hardening stage, the cumulative curing time reflects the duration of the curing process; for pastes that have completed solidification, the cumulative curing time is used to characterize their continued existence and evolutionary state after formation.

[0008] Optionally, the construction process of the divine ordinary differential equation is as follows: Input feature vector for:

[0009] in, C represents the cumulative curing time of the infill material at height h, and C represents the infill concentration of the infill material at height h. The ratio of ash to sand; The cumulative maintenance time is defined as:

[0010] in, This is the moment when the paste at height h enters the corresponding position in the mining area; Based on the input feature vector Construct the continuous evolution relationship of the hidden state variable Z(h) along the height direction:

[0011] Where h represents the height coordinate along the retaining wall, and Z represents the 16-dimensional hidden state variable. For a neural network used to fit the rate of change of hidden states along the height direction These are the parameters to be trained.

[0012] Optionally, the constant differential equation uses the height of the liquid level in the filling material as a boundary reference condition: from the current height of the liquid level in the filling material... Integrating begins at point [0,0,0……,0], with the hidden state variables initially set to [0,0,0……,0], represented as:

[0013] Given the boundary reference conditions, the ordinary differential equation ODE is integrated from top to bottom along the height of the retaining wall to obtain the hidden state quantity Z(h) at any height position h. Then, the hidden state quantity is mapped to the pressure through a network. The pressure prediction value is obtained by mapping. :

[0014] in, This is the intermediate height variable during the integration process.

[0015] On the other hand, a real-time prediction system for the pressure field of a mine backfill retaining wall is provided, the system comprising: The first construction module is used to scan the three-dimensional model of a small number of typical mining sites, and collect real-time filling flow multi-source monitoring data of key points, as well as retaining wall pressure of key points, to construct a field time series dataset of retaining wall pressure of key points. The real-time filling flow multi-source monitoring data of key points includes the filling time of paste, in-situ attributes, and cumulative curing time. The generation module is used to continuously simulate the filling process based on the three-dimensional model and real-time filling flow multi-source monitoring data, and to generate dynamic layered attribute features by backtracking the filling time, in-situ attributes and cumulative curing time of the paste at any predicted height position in the mining area. The dynamic layered attribute features include height position, filling time, in-situ attributes and cumulative curing time. The second construction module is used to construct training samples based on the on-site time-series dataset and dynamic hierarchical attribute features, with the dynamic hierarchical attribute features of key points as input and the retaining wall pressure of key points as labels. A module is established to build a continuous spatial prediction model of the pressure field of the retaining wall using the neural network: with height as the continuous independent variable and hidden state variables as the evolution objects, the rate of change of the hidden state variables with respect to height is parameterized by the neural network. Boundary reference conditions are given at the liquid level, and the hidden state variables at each height are obtained by integration along the height direction. Then, the hidden state variables are mapped to the pressure field of the retaining wall through the mapping network from the hidden state variables to the pressure, so as to realize the continuous prediction of the pressure field along the height. The training module is used to perform supervised training on the prediction model using the training samples to obtain the trained prediction model. The prediction module is used for the online pressure field prediction stage of other target mining areas. It generates dynamic hierarchical attribute features of key points based on the currently collected data and inputs them into the trained prediction model to continuously predict the pressure across the entire height range of the retaining wall.

[0016] Optionally, the in-situ attributes refer to attribute information directly related to the state of the paste corresponding to the target height position, including the paste concentration, ash-sand ratio, filling batch parameters corresponding to the paste filling time at that position, and the mining section or structural features related to that position.

[0017] Optionally, the cumulative curing time refers to the time elapsed since the paste entered a certain height position until the current moment, used to characterize the differences in the formation sequence of the paste and its state evolution process: For pastes in the solidification and hardening stage, the cumulative curing time reflects the duration of the curing process; for pastes that have completed solidification, the cumulative curing time is used to characterize their continued existence and evolutionary state after formation.

[0018] Optionally, the construction process of the divine ordinary differential equation is as follows: Input feature vector for:

[0019] in, C represents the cumulative curing time of the infill material at height h, and C represents the infill concentration of the infill material at height h. The ratio of ash to sand; The cumulative maintenance time is defined as:

[0020] in, This is the moment when the paste at height h enters the corresponding position in the mining area; Based on the input feature vector Construct the continuous evolution relationship of the hidden state variable Z(h) along the height direction:

[0021] Where h represents the height coordinate along the retaining wall, and Z represents the 16-dimensional hidden state variable. For a neural network used to fit the rate of change of hidden states along the height direction These are the parameters to be trained.

[0022] Optionally, the constant differential equation uses the height of the liquid level in the filling material as a boundary reference condition: from the current height of the liquid level in the filling material... Integrating begins at point [0,0,0……,0], with the hidden state variables initially set to [0,0,0……,0], represented as:

[0023] Given the boundary reference conditions, the ordinary differential equation ODE is integrated from top to bottom along the height of the retaining wall to obtain the hidden state quantity Z(h) at any height position h. Then, the hidden state quantity is mapped to the pressure through a network. The pressure prediction value is obtained by mapping. :

[0024] in, For the intermediate height variable in the integration process

[0025] On the other hand, an electronic device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, which is loaded and executed by the processor to realize the above-mentioned method for real-time prediction of pressure field of mine backfill retaining wall.

[0026] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement the above-described method for real-time prediction of pressure field of mine backfill retaining wall.

[0027] The beneficial effects of the technical solution provided by this invention include at least the following: 1. This invention is based on multi-source monitoring data from the field, which can more realistically reflect the actual working conditions of the project.

[0028] 2. This invention solves the problem of spatial non-uniformity of paste state at different height positions by utilizing the dynamic layering property characteristics of the filling body.

[0029] 3. This invention uses the constant differential equation to predict the continuous pressure field over the entire height range of the retaining wall.

[0030] 4. This invention is applicable to both the continuous filling stage and the stop filling stage, enabling continuous prediction and risk warning of the retaining wall pressure field during both stages.

[0031] 5. This invention can be applied to other target mining areas after model training has been completed in a small number of typical mining areas. Even when pressure monitoring devices are not deployed in other target mining areas or when real-time pressure data is lacking, the pressure field prediction of the entire height of the retaining wall can still be achieved based on real-time operating data and structural parameters, thereby reducing the need for on-site pressure monitoring deployment. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a flowchart of a method for real-time prediction of pressure field of mine backfill retaining wall provided by an embodiment of the present invention; Figure 2 This is a block diagram of a real-time prediction system for the pressure field of a mine backfill retaining wall provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0034] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0035] This invention provides a method for real-time prediction of the pressure field of a mine backfill retaining wall. This method can be implemented by an electronic device, which can be a terminal or a server. Figure 1 The diagram shown is a flowchart of the method. The processing flow may include the following steps: S1. For a small number of typical mining sites, scan their three-dimensional models and collect real-time filling flow multi-source monitoring data of key points, as well as retaining wall pressure of key points, to construct a field time series dataset of retaining wall pressure of key points. The real-time filling flow multi-source monitoring data of key points includes the filling time of paste, in-situ attributes, and cumulative curing time. Optionally, in this embodiment of the invention, data acquisition is performed by a data acquisition layer, including: (1) Data acquisition unit for filling station operation, used to acquire production parameters such as filling start time, end time, filling round, filling flow rate, paste concentration and ash-sand ratio; (2) Retaining wall pressure monitoring unit, used to acquire retaining wall pressure monitoring values; (3) The mining area structure information input unit is used to process the three-dimensional model of the mining area, collect and record structural information such as the geometric dimensions of the mining area, the dimensions of the retaining wall, and the cross-sectional shape.

[0036] After data collection and preprocessing, the data is uniformly incorporated into the field time-series dataset. Preprocessing includes timestamp alignment, outlier removal, missing value imputation, smoothing, and standardization. These steps create a field time-series dataset suitable for subsequent dynamic hierarchical attribute representation of infill bodies, model training, and online inference.

[0037] Optionally, the in-situ attributes refer to attribute information directly related to the state of the paste corresponding to the target height position, including the paste concentration, ash-sand ratio, filling batch parameters corresponding to the paste filling time at that position, and the mining section or structural features related to that position.

[0038] Optionally, the cumulative curing time refers to the time elapsed since the paste entered a certain height position until the current moment, used to characterize the differences in the formation sequence of the paste and its state evolution process: For pastes in the solidification and hardening stage, the cumulative curing time reflects the duration of the curing process; for pastes that have completed solidification, the cumulative curing time is used to characterize their continued existence and evolutionary state after formation.

[0039] S2. Based on the three-dimensional model and real-time filling flow multi-source monitoring data, the filling process is continuously simulated. For any predicted height position in the mining area, the filling time, in-situ attributes and cumulative curing time of the paste at any predicted height position are traced back to generate dynamic layered attribute features. The dynamic layered attribute features include height position, filling time, in-situ attributes and cumulative curing time. S3. Based on the aforementioned on-site time-series dataset and dynamic hierarchical attribute features, construct training samples with the dynamic hierarchical attribute features of key points as input and the retaining wall pressure of key points as labels. S4. A continuous prediction model for the pressure space of the retaining wall is established using the constant differential equation of the God: with height as the continuous independent variable and hidden state variables as the evolution object, the rate of change of the hidden state variables with respect to height is parameterized by the neural network. Boundary reference conditions are given at the liquid level. The hidden state variables at each height are obtained by integration along the height direction. Then, the hidden state variables are mapped to the pressure of the retaining wall through the mapping network from the hidden state variables to the pressure, so as to realize the continuous prediction of the pressure field along the height. Optionally, the construction process of the divine ordinary differential equation is as follows: Input feature vector for:

[0040] in, C represents the cumulative curing time of the infill material at height h, and C represents the infill concentration of the infill material at height h. The ratio of ash to sand; The cumulative maintenance time is defined as:

[0041] in, This is the moment when the paste at height h enters the corresponding position in the mining area; Based on the input feature vector Construct the continuous evolution relationship of the hidden state variable Z(h) along the height direction:

[0042] Where h represents the height coordinate along the retaining wall, and Z represents the 16-dimensional hidden state variable. For a neural network used to fit the rate of change of hidden states along the height direction These are the parameters to be trained.

[0043] Optionally, the constant differential equation uses the height of the liquid level in the filling material as a boundary reference condition: from the current height of the liquid level in the filling material... Integrating begins at point [0,0,0……,0], with the hidden state variables initially set to [0,0,0……,0], represented as:

[0044] Given the boundary reference conditions, the ordinary differential equation ODE is integrated from top to bottom along the height of the retaining wall to obtain the hidden state quantity Z(h) at any height position h. Then, the hidden state quantity is mapped to the pressure through a network. The pressure prediction value is obtained by mapping. :

[0045] in, This is the intermediate height variable during the integration process.

[0046] S5. Supervised training is performed on the prediction model using the training samples to obtain the trained prediction model. Optionally, a single training sample is composed of operating condition input features. With real labels Composition, including tags The pressure on the retaining wall was measured on-site.

[0047] The model training uses mean squared error (MSE) as the target loss function, expressed as:

[0048] In the formula: The predicted pressure on the retaining wall output by the model; θ represents the true value of the measured retaining wall pressure; θ represents all optimizable parameters of the model; N is the total number of samples in the training set.

[0049] Minimize the loss using the gradient descent algorithm. The model parameters θ are iteratively updated to complete the training of the prediction model.

[0050] S6. For the online pressure field prediction stage of other target mining areas, dynamic hierarchical attribute features of key points are generated based on the current collected data and input into the trained prediction model to continuously predict the pressure across the entire height range of the retaining wall.

[0051] Optionally, during the online operation phase, real-time monitoring data is acquired and preprocessed to calculate the filling time, in-situ properties, and cumulative curing time of the paste at each height location. The obtained features are then input into the prediction model, and the continuous pressure field distribution across the entire height range of the retaining wall is obtained through integral solving.

[0052] During the continuous filling phase, the model updates the pressure prediction results in real time based on changes in the liquid level of the filling body, the filling flow rate, and the paste concentration. During the stop filling phase, although the liquid level of the filling body remains basically unchanged, the cumulative curing time of the paste at each height position still increases over time. Therefore, the model can still make dynamic predictions on the pressure field based on changes in dynamic attributes.

[0053] Subsequently, the prediction results are compared with the design thresholds and safety criteria. When the warning conditions are met, the stress assessment results are output and the alarm system is activated.

[0054] Optionally, embodiments of the present invention can be applied to cross-mining site migration: In one or more existing typical stopes with a small number of samples, a prediction model established using the constant differential equation is trained using training samples to obtain a trained prediction model. After training, the model is deployed to other target stopes. During the operation of other target stopes, if no pressure monitoring device is installed or real-time pressure data is lacking, it is only necessary to collect information such as the liquid level height of the filling body, the filling flow rate, the paste concentration, the ash-sand ratio, and the stope structure parameters to generate dynamic stratified attribute features of key points. These features can then be input into the trained model to continuously predict the pressure field across the entire height range of the retaining wall in other target stopes.

[0055] To improve cross-mining application capabilities, during the training phase, data from multiple different mining sites can be selected to jointly construct a training set, enabling the model to learn the common patterns of retaining wall pressure evolution under different mining site conditions.

[0056] If other target mining areas have a small amount of historical pressure data, this small amount of data can be used to perform lightweight calibration of the model to further improve prediction accuracy.

[0057] like Figure 2 As shown in the figure, this invention also provides a real-time prediction system for the pressure field of a mine backfill retaining wall, the system comprising: The first construction module 210 is used to scan the three-dimensional model of a small number of typical mining sites, and collect real-time filling flow multi-source monitoring data of key points, as well as retaining wall pressure of key points, to construct a field time series dataset of retaining wall pressure of key points. The real-time filling flow multi-source monitoring data of key points includes the filling time of paste, in-situ attributes, and cumulative curing time. The generation module 220 is used to continuously simulate the filling process based on the three-dimensional model and real-time filling flow multi-source monitoring data, and to generate dynamic layered attribute features by backtracking the filling time, in-situ attributes and cumulative curing time of the paste at any predicted height position in the mining area. The dynamic layered attribute features include height position, filling time, in-situ attributes and cumulative curing time. The second construction module 230 is used to construct training samples based on the on-site time-series dataset and dynamic hierarchical attribute features, with the dynamic hierarchical attribute features of key points as input and the retaining wall pressure of key points as labels. Module 240 is established to build a continuous spatial prediction model of the retaining wall pressure using the constant differential equation of the God: with height as the continuous independent variable and hidden state variables as the evolution object, the rate of change of the hidden state variables with respect to height is parameterized by the neural network. Boundary reference conditions are given at the liquid level, and the hidden state variables at each height are obtained by integration along the height direction. Then, the hidden state variables are mapped to the retaining wall pressure through the mapping network from the hidden state variables to the pressure, so as to realize the continuous prediction of the pressure field along the height. The training module 250 is used to perform supervised training on the prediction model using the training samples to obtain the trained prediction model. The prediction module 260 is used for the online pressure field prediction stage of other target mining areas. It generates dynamic hierarchical attribute features of key points based on the currently collected data and inputs them into the trained prediction model to continuously predict the pressure across the entire height range of the retaining wall.

[0058] The real-time prediction system for the pressure field of a mine backfill retaining wall provided in this embodiment of the invention has a functional structure that corresponds to the real-time prediction method for the pressure field of a mine backfill retaining wall provided in this embodiment of the invention, and will not be described again here.

[0059] Figure 3 This is a schematic diagram of the structure of an electronic device 300 provided in an embodiment of the present invention. The electronic device 300 may vary considerably due to different configurations or performance. It may include one or more central processing units (CPUs) 301 and one or more memories 302. The memory 302 stores at least one instruction, which is loaded and executed by the processor 301 to implement the steps of the above-mentioned method for real-time prediction of pressure field of mine backfill retaining wall.

[0060] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including instructions that can be executed by a processor in a terminal to complete the aforementioned method for real-time prediction of the pressure field of a mine backfill retaining wall. For example, the computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device.

[0061] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0062] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for real-time prediction of the pressure field of a mine backfill retaining wall, characterized in that, The method includes: S1. For a small number of typical mining sites, scan their three-dimensional models and collect real-time filling flow multi-source monitoring data of key points, as well as retaining wall pressure of key points, to construct a field time series dataset of retaining wall pressure of key points. The real-time filling flow multi-source monitoring data of key points includes the filling time of paste, in-situ attributes, and cumulative curing time. S2. Based on the three-dimensional model and real-time filling flow multi-source monitoring data, the filling process is continuously simulated. For any predicted height position in the mining area, the filling time, in-situ attributes and cumulative curing time of the paste at any predicted height position are traced back to generate dynamic layered attribute features. The dynamic layered attribute features include height position, filling time, in-situ attributes and cumulative curing time. S3. Based on the aforementioned on-site time-series dataset and dynamic hierarchical attribute features, construct training samples with the dynamic hierarchical attribute features of key points as input and the retaining wall pressure of key points as labels. S4. A continuous prediction model for the pressure space of the retaining wall is established using the constant differential equation of the God: with height as the continuous independent variable and hidden state variables as the evolution object, the rate of change of the hidden state variables with respect to height is parameterized by the neural network. Boundary reference conditions are given at the liquid level. The hidden state variables at each height are obtained by integration along the height direction. Then, the hidden state variables are mapped to the pressure of the retaining wall through the mapping network from the hidden state variables to the pressure, so as to realize the continuous prediction of the pressure field along the height. S5. Supervised training is performed on the prediction model using the training samples to obtain the trained prediction model. S6. For the online pressure field prediction stage of other target mining areas, dynamic hierarchical attribute features of key points are generated based on the current collected data and input into the trained prediction model to continuously predict the pressure across the entire height range of the retaining wall.

2. The method according to claim 1, characterized in that, The in-situ attributes refer to attribute information directly related to the state of the paste corresponding to the target height position, including the paste concentration, ash-sand ratio, filling batch parameters, and the mining section or structural features related to the position, corresponding to the paste filling time.

3. The method according to claim 1, characterized in that, The cumulative curing time refers to the time elapsed since the paste entered a certain height position until the current moment, used to characterize the differences in the formation time of the paste and its state evolution process: For pastes in the solidification and hardening stage, the cumulative curing time reflects the duration of the curing process; for pastes that have completed solidification, the cumulative curing time is used to characterize their continued existence and evolutionary state after formation.

4. The method according to claim 1, characterized in that, The process of constructing the divine ordinary differential equation is as follows: Input feature vector for: ； in, C represents the cumulative curing time of the infill material at height h, and C represents the infill concentration of the infill material at height h. The ratio of ash to sand; The cumulative maintenance time is defined as: ； in, This is the moment when the paste at height h enters the corresponding position in the mining area; Based on the input feature vector Construct the continuous evolution relationship of the hidden state variable Z(h) along the height direction: ； Where h represents the height coordinate along the retaining wall, and Z represents the 16-dimensional hidden state variable. For a neural network used to fit the rate of change of hidden states along the height direction, These are the parameters to be trained.

5. The method according to claim 4, characterized in that, The aforementioned constant differential equation uses the height of the liquid level in the filling body as a boundary reference condition: from the current height of the liquid level in the filling body... Integrating begins at point [0,0,0……,0], with the hidden state variables initially set to [0,0,0……,0], represented as: Given the boundary reference conditions, the ordinary differential equation ODE is integrated from top to bottom along the height of the retaining wall to obtain the hidden state quantity Z(h) at any height position h. Then, the hidden state quantity is mapped to the pressure through a network. The pressure prediction value is obtained by mapping. : ； ； in, This is the intermediate height variable during the integration process.

6. A real-time prediction system for the pressure field of a mine backfill retaining wall, characterized in that, The system includes: The first construction module is used to scan the three-dimensional model of a small number of typical mining sites, and collect real-time filling flow multi-source monitoring data of key points, as well as retaining wall pressure of key points, to construct a field time series dataset of retaining wall pressure of key points. The real-time filling flow multi-source monitoring data of key points includes the filling time of paste, in-situ attributes, and cumulative curing time. The generation module is used to continuously simulate the filling process based on the three-dimensional model and real-time filling flow multi-source monitoring data, and to generate dynamic layered attribute features by backtracking the filling time, in-situ attributes and cumulative curing time of the paste at any predicted height position in the mining area. The dynamic layered attribute features include height position, filling time, in-situ attributes and cumulative curing time. The second construction module is used to construct training samples based on the on-site time-series dataset and dynamic hierarchical attribute features, with the dynamic hierarchical attribute features of key points as input and the retaining wall pressure of key points as labels. A module is established to build a continuous spatial prediction model of the pressure field of the retaining wall using the neural network: with height as the continuous independent variable and hidden state variables as the evolution objects, the rate of change of the hidden state variables with respect to height is parameterized by the neural network. Boundary reference conditions are given at the liquid level, and the hidden state variables at each height are obtained by integration along the height direction. Then, the hidden state variables are mapped to the pressure field of the retaining wall through the mapping network from the hidden state variables to the pressure, so as to realize the continuous prediction of the pressure field along the height. The training module is used to perform supervised training on the prediction model using the training samples to obtain the trained prediction model. The prediction module is used for the online pressure field prediction stage of other target mining areas. It generates dynamic hierarchical attribute features of key points based on the current collected data and inputs them into the trained prediction model to continuously predict the pressure across the entire height range of the retaining wall.

7. The system according to claim 6, characterized in that, The in-situ attributes refer to attribute information directly related to the state of the paste corresponding to the target height position, including the paste concentration, ash-sand ratio, filling batch parameters, and the mining section or structural features related to the position, corresponding to the paste filling time.

8. The system according to claim 6, characterized in that, The cumulative curing time refers to the time elapsed since the paste entered a certain height position until the current moment, used to characterize the differences in the formation time of the paste and its state evolution process: For pastes in the solidification and hardening stage, the cumulative curing time reflects the duration of the curing process; for pastes that have completed solidification, the cumulative curing time is used to characterize their continued existence and evolutionary state after formation.

9. The system according to claim 6, characterized in that, The process of constructing the divine ordinary differential equation is as follows: Input feature vector for: ； in, C represents the cumulative curing time of the infill material at height h, and C represents the infill concentration of the infill material at height h. The ratio of ash to sand; The cumulative maintenance time is defined as: ； in, This is the moment when the paste at height h enters the corresponding position in the mining area; Based on the input feature vector Construct the continuous evolution relationship of the hidden state variable Z(h) along the height direction: ； Where h represents the height coordinate along the retaining wall, and Z represents the 16-dimensional hidden state variable. For a neural network used to fit the rate of change of hidden states along the height direction These are the parameters to be trained.

10. The system according to claim 9, characterized in that, The aforementioned constant differential equation uses the height of the liquid level in the filling body as a boundary reference condition: from the current height of the liquid level in the filling body... Integrating begins at point [0,0,0……,0], with the hidden state variables initially set to [0,0,0……,0], represented as: ； Given the boundary reference conditions, the ordinary differential equation ODE is integrated from top to bottom along the height of the retaining wall to obtain the hidden state quantity Z(h) at any height position h. Then, the hidden state quantity is mapped to the pressure through a network. The pressure prediction value is obtained by mapping. : ； ； in, This is the intermediate height variable during the integration process.

Images