Mine pressure prediction method based on mixing of large and small models

Abstract

The invention discloses a mine pressure prediction method based on mixing of large and small models, and the method comprises the steps: obtaining original data collected by at least one underground sensor, and carrying out the processing of the original data, and obtaining a multi-modal spatial-temporal feature; based on the multi-modal spatial-temporal characteristics, determining a target scene type to which the scene belongs through a large language model; determining a corresponding scheduling model set from a small model library based on the target scene type and the multi-modal spatial-temporal characteristics; and obtaining a target mine pressure prediction result through a scheduling model set and a large language model based on the multi-modal spatial-temporal characteristics and the target scene type. According to the method, the adaptability and the accuracy of the mine pressure prediction result are improved, the interpretability and the engineering availability are enhanced, a clear reference basis is provided for an underground dispatcher, and the credibility and the decision-making efficiency of the mine pressure prediction result are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mine pressure prediction, in particular to a mine pressure prediction method and device based on large-small model mixing and a storage medium. BACKGROUND

[0002] With the deepening of the intelligent construction of mines, mine pressure prediction technology, as a key means to ensure mining safety, relies on intelligent analysis methods of support pressure and microseismic monitoring sensor data. Among them, the underground mining environment has high dynamicity and complexity, the interaction of factors such as working face advancing speed, geological structure change, equipment operating state, etc. makes the mine pressure data present significant nonlinear, time series fluctuation and spatial heterogeneity characteristics.

[0003] In related technologies, mine pressure prediction models (such as LSTM) are difficult to adapt to changing data distribution under different working faces, geological conditions or sudden working conditions (such as fault crossing and roof pressure), cannot actively perceive environmental changes, errors are easy to accumulate in long-term prediction, and the accuracy of mine pressure prediction results is reduced. At the same time, the mine pressure prediction results obtained by related technologies lack explainability, are difficult to gain the trust of personnel, and restrict the practical application of prediction results in safety decision-making. SUMMARY

[0004] The present application aims to at least partially solve one of the technical problems in the related art.

[0005] To this end, the present application proposes a mine pressure prediction method based on large-small model mixing, which can perform semantic analysis and identify the current target scene type through a large language model, and call a target small model in the dispatch model set matched with the target scene type to perform mine pressure prediction, thereby improving the adaptability and accuracy of the mine pressure prediction result, enhancing the explainability and engineering usability, providing clear reference for underground dispatchers, and improving the trustworthiness and decision-making efficiency of the mine pressure prediction result.

[0006] Another object of the present application is to propose a mine pressure prediction device based on large-small model mixing.

[0007] To achieve the above-mentioned purpose, the present application proposes a mine pressure prediction method based on large-small model mixing, which comprises: obtaining original data collected by at least one sensor underground and processing the original data to obtain multi-modal spatio-temporal features; determining the target scene type through a large language model based on the multi-modal spatio-temporal features; determining the corresponding dispatch model set from the small model library based on the target scene type and the multi-modal spatio-temporal features; Based on the multi-modal spatio-temporal features and the target scene type, a target mine pressure prediction result is obtained through the scheduling model set and the large language model.

[0008] The mine pressure prediction method based on the large and small model hybrid of the embodiment of the application can further have the following additional technical features: In an embodiment of the application, the processing of the original data to obtain multi-modal spatio-temporal features comprises: The original data is resampled by using a sliding window mechanism to obtain resampled data; The resampled data is aligned to obtain aligned data; The aligned data is embedded into a first feature vector by using an encoder, and the numerical information in the aligned data is determined as a second feature vector; The first feature vector and the second feature vector are spliced to obtain multi-modal spatio-temporal features.

[0009] In an embodiment of the application, the determination of the target scene type through the large language model based on the multi-modal spatio-temporal features comprises: A semantic prompt word corresponding to the multi-modal spatio-temporal features is determined; The multi-modal spatio-temporal features and the semantic prompt word are input into a large language model to obtain a first probability distribution corresponding to the scene type; Based on the first probability distribution, the target scene type is determined.

[0010] In an embodiment of the application, the determination of the corresponding scheduling model set from the small model library based on the target scene type and the multi-modal spatio-temporal features comprises: The target scene type and the multi-modal spatio-temporal features are input into a target model selector to obtain a second probability distribution of small models in the small model library for scheduling; Based on the second probability distribution, the label order of the target small model for scheduling is determined; The weight coefficient corresponding to the target small model is obtained, and the label order of the target small model and the weight coefficient of the target small model are determined as a scheduling model set.

[0011] In an embodiment of the application, the obtaining of the target mine pressure prediction result through the scheduling model set and the large language model based on the multi-modal spatio-temporal features and the target scene type comprises: The target small models in the scheduling model set are sequentially scheduled, and first mine pressure prediction results of each target small model are obtained based on the multi-modal spatio-temporal features; weighting and summing the weight coefficients of each target small model and the first mine pressure prediction result to obtain a second mine pressure prediction result; based on the target scene type, determining a corresponding historical scheduling model set and input semantics; based on the historical scheduling model set and the input semantics, generating an explanation content in natural language through the large language model; determining the explanation content and the second mine pressure prediction result as a target mine pressure prediction result.

[0012] In an embodiment of the present application, the method further comprises: obtaining a real collection value corresponding to the target mine pressure prediction result; based on the target mine pressure prediction result and the real collection value, determining a prediction error; based on the prediction error, determining whether to adjust the large model and the target small model in the scheduling model set; if it is determined to adjust the large model and the target small model in the scheduling model set, adjusting the large model and the target small model in the scheduling model set based on the target mine pressure prediction result and the real collection value.

[0013] In an embodiment of the present application, the determination of whether to adjust the large model and the target small model in the scheduling model set based on the prediction error comprises: if the prediction error exceeds a preset threshold, it is determined to adjust the large model and the target small model in the scheduling model set.

[0014] In an embodiment of the present application, the adjustment of the large model and the target small model in the scheduling model set based on the target mine pressure prediction result and the real collection value comprises: based on the target mine pressure prediction result and the real collection value, determining a loss value through a loss function; based on the loss value, adjusting network parameters in the target small model to obtain an adjusted small model; based on the target mine pressure prediction result and the real collection value, fine-tuning the large model to obtain a fine-tuned large model.

[0015] Another aspect of the present application proposes a mine pressure prediction device based on a large-small model hybrid, which comprises: a processing module for obtaining original data collected by at least one sensor underground and processing the original data to obtain multi-modal spatio-temporal features; The first determination module is configured to determine a target scene type by using a large language model based on the multi-modal spatio-temporal features. The second determination module is configured to determine a corresponding scheduling model set from a small model library based on the target scene type and the multi-modal spatio-temporal features. The prediction module is configured to obtain a target mine pressure prediction result by using the scheduling model set and the large language model based on the multi-modal spatio-temporal features and the target scene type.

[0016] The mine pressure prediction method and device based on the large-small model hybrid according to the embodiments of the present application include the following steps: obtaining original data collected by at least one sensor underground and processing the original data to obtain multi-modal spatio-temporal features; determining a target scene type by using a large language model based on the multi-modal spatio-temporal features; determining a corresponding scheduling model set from a small model library based on the target scene type and the multi-modal spatio-temporal features; and obtaining a target mine pressure prediction result by using the scheduling model set and the large language model based on the multi-modal spatio-temporal features and the target scene type. Thus, the present application can perform semantic analysis and identify the current target scene type by using a large language model, and call a target small model in the scheduling model set matched with the target scene type to perform mine pressure prediction, thereby improving the adaptability and accuracy of the mine pressure prediction result, enhancing the explainability and engineering usability, providing clear reference basis for underground dispatchers, and improving the trustworthiness and decision efficiency of the mine pressure prediction result.

[0017] Additional aspects and advantages of the present application will be made apparent by the following description and the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0018] The above and / or additional aspects and advantages of the present application will become apparent and be readily understood from the following description, taken in conjunction with the accompanying drawings, in which: Figure 1 is a flowchart of a mine pressure prediction method based on a large-small model hybrid according to an embodiment of the present application; Figure 2 is a structural diagram of a mine pressure prediction device based on a large-small model hybrid according to an embodiment of the present application. DETAILED DESCRIPTION

[0019] It should be noted that the embodiments and features in the present application can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0020] In order to make the personnel in the technical field better understand the present application scheme, the technical scheme in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by the person skilled in the art without creative labor should belong to the protection scope of the present application.

[0021] The size model hybrid-based mine pressure prediction method and device proposed according to the embodiments of the present application are described below with reference to the drawings.

[0022] Figure 1 The flowchart of the size model hybrid-based mine pressure prediction method of the embodiments of the present application is shown in the figure.

[0023] As shown in the figure, the method can include the following steps: Figure 1 Step 101, acquiring original data collected by at least one sensor underground and processing the original data to obtain multi-modal spatio-temporal features. In an embodiment of the present application, the acquisition of the original data collected by at least one sensor underground can include support pressure, hydraulic system state, coal mining machine current and position, roof type, advancing speed and engineering log.

[0024] In an embodiment of the present application, after acquiring the original data collected by at least one sensor underground, the original data can be processed to obtain multi-modal spatio-temporal features to construct a unified input space and solve the problems of mixed data sources, different frequencies and time sequence misplacement in mine pressure prediction.

[0025] In an embodiment of the present application, the method of processing the original data to obtain multi-modal spatio-temporal features can include the following steps:

[0026] Step 1011, resampling the original data by using a sliding window mechanism to obtain resampled data; Step 1012, aligning the resampled data to obtain aligned data; Step 1013, embedding the text information in the aligned data into a first feature vector by using an encoder and determining the numerical information in the aligned data as a second feature vector; Step 1014, splicing the first feature vector and the second feature vector to obtain multi-modal spatio-temporal features. In an embodiment of the present application, the original data is resampled by using a sliding window mechanism to solve the difference in data sampling frequency and time delay.

[0027]

[0028] ​In an embodiment of the present application, the method for aligning the resampled data to obtain aligned data can include: aligning the resampled data by timestamp synchronization, missing interpolation, and normalization processing to obtain aligned data.

[0029] Further, in an embodiment of the present application, after obtaining the first feature vector and the second feature vector through the above steps, the first feature vector and the second feature vector can be spliced to obtain a multi-modal spatio-temporal feature , so as to provide high-quality input data for subsequent scene recognition and mine pressure prediction.

[0030] In an embodiment of the present application, the method for obtaining the multi-modal spatio-temporal feature based on the multi-modal data can include the following steps: may be:

[0031] , wherein, represents a multi-modal spatio-temporal feature constructed at ; represents time series data features collected by a support pressure sensor; represents device (such as a coal mining machine and a conveyor) operating state features (such as current, position, and speed); represents geological environment features (such as roof type and support form); represents text description features (such as safety logs and working condition records).

[0032] Step 102, determining a target scene type to which the multi-modal spatio-temporal feature belongs through a large language model based on the multi-modal spatio-temporal feature.

[0033] In an embodiment of the present application, after obtaining the multi-modal spatio-temporal feature through the above steps, a target scene type to which the multi-modal spatio-temporal feature belongs can be determined through a large language model based on the multi-modal spatio-temporal feature.

[0034] Specifically, in an embodiment of the present application, the method for determining a target scene type to which the multi-modal spatio-temporal feature belongs through a large language model based on the multi-modal spatio-temporal feature can include the following steps: Step 1021, determining a semantic prompt word corresponding to the multi-modal spatio-temporal feature; Step 1022, inputting the multi-modal spatio-temporal feature and the semantic prompt word into a large language model to obtain a first probability distribution corresponding to the scene type; Step 1023, determining a target scene type to which the multi-modal spatio-temporal feature belongs based on the first probability distribution.

[0035] In one embodiment of the present invention, semantic cue words corresponding to multimodal spatiotemporal features can be determined based on experience and needs. Also, in another embodiment of the present invention, semantic cue words corresponding to multimodal spatiotemporal features can be determined using a cue word model. The cue word model can be an existing model, such as the HiTime hierarchical multimodal model.

[0036] Furthermore, in one embodiment of the present invention, after determining the semantic prompt words corresponding to the multimodal spatiotemporal features, the multimodal spatiotemporal features and semantic prompt words can be input into a large language model to obtain a first probability distribution corresponding to the scene type. The large language model is obtained by fine-tuning using training data from that domain.

[0037] In one embodiment of the present invention, the first probability distribution corresponding to the scene type obtained through the above steps can be expressed as:

[0038] in, Represented as multimodal spatiotemporal features Belongs to scene type The probability of; Represented as the first Scenario types (such as periodic pressure, fault effects, etc.); Indicates by parameters A large language model for control is used for functions that match multimodal spatiotemporal features and semantic cue words with scene types; This represents the standard normalization function, which outputs the probability distribution.

[0039] Furthermore, in one embodiment of the present invention, after obtaining the first probability distribution through the above steps, the target scene type can be determined based on the first probability distribution. Specifically, in one embodiment of the present invention, the scene type corresponding to the highest probability in the first probability distribution can be determined as the target scene type.

[0040] Step 103: Based on the target scene type and multimodal spatiotemporal characteristics, determine the corresponding scheduling model set from the small model library.

[0041] In one embodiment of the present invention, after determining the target scene type and multimodal spatiotemporal characteristics through the above steps, the corresponding scheduling model set can be determined from the small model library based on the target scene type and multimodal spatiotemporal characteristics.

[0042] Specifically, in one embodiment of the present invention, the method for determining the corresponding scheduling model set from the small model library based on the target scene type and multimodal spatiotemporal characteristics may include the following steps: Step 1031, input the target scene type and the multi-modal spatio-temporal feature into the target model selector, to obtain a second probability distribution of the small models in the small model library for scheduling; Step 1032, determine the label order of the target small model for scheduling based on the second probability distribution; Step 1033, obtain the weight coefficient corresponding to the target small model, and determine the label order of the target small model and the weight coefficient of the target small model as the scheduling model set.

[0043] In an embodiment of the present application, the small model library can include small models with lightweight and efficient structures, and each small model is pre-trained under different scenarios according to historical data. In an embodiment of the present application, the small model library can include at least one of LSTM, Mamba, TCN and iTransformer.

[0044] In an embodiment of the present application, the target model selector can be obtained by training.

[0045] In an embodiment of the present application, the second probability distribution obtained by the target model selector can include small models in the small model library and corresponding probabilities.

[0046] In an embodiment of the present application, after obtaining the second probability distribution through the above steps, the method of determining the label order of the target small model for scheduling based on the second probability distribution can include: sorting the probabilities in the second probability distribution in descending order, and determining the top pre-set number of small models as the target small models, and sequentially labeling the target small models in descending order to obtain the label order of the target small models.

[0047] In an embodiment of the present application, the pre-set number can be set as needed, such as 1 or 3. When the pre-set number is 1, a single model can be matched; when the pre-set number is greater than 1, multiple models can be matched.

[0048] In an embodiment of the present application, after obtaining the target small model through the above steps, the weight coefficient corresponding to the target small model can be obtained, and the label order of the target small model and the weight coefficient of the target small model are determined as the scheduling model set. In an embodiment of the present application, the weight coefficient corresponding to each target small model input by the user can be obtained, and the sum of the weight coefficients of all target small models is 1.

[0049] In an embodiment of the present application, the scheduling model set obtained through the above steps can be represented as:

[0050] wherein, Represented as a set of scheduling models, it can specify the order in which target sub-models are called; Represented as the first A small target model; This represents the weight coefficient of the j-th target mini-model that is scheduled; This indicates the number of target small models invoked in this round of scheduling; and This achieves normalized weighting.

[0051] Step 104: Based on multimodal spatiotemporal features and target scene type, obtain the target mine pressure prediction result through a set of scheduling models and a large language model.

[0052] In one embodiment of the present invention, after obtaining the multimodal spatiotemporal features, target scene type and scheduling model set through the above steps, the target mineral pressure prediction result can be obtained based on the multimodal spatiotemporal features and target scene type, through the scheduling model set and large language model.

[0053] Specifically, in one embodiment of the present invention, the method for obtaining target mineral pressure prediction results based on multimodal spatiotemporal features and target scene type through a set of scheduling models and a large language model may include the following steps: Step 1041: Sequentially schedule the target sub-models in the scheduling model set, and obtain the first mineral pressure prediction result of each target sub-model based on the multimodal spatiotemporal characteristics; Step 1042: The weighted sum of the weight coefficients of each target small model and the first mine pressure prediction result is performed to obtain the second mine pressure prediction result; Step 1043: Based on the target scenario type, determine the corresponding set of historical scheduling models and input semantics; Step 1044: Based on the historical scheduling model set and input semantics, generate natural language explanations using a large language model; Step 1045: Combine the explanation with the second mine pressure prediction result to determine the target mine pressure prediction result.

[0054] In one embodiment of the present invention, target small models in the scheduling model set are scheduled sequentially based on multimodal spatiotemporal features. Obtain the future time steps output by each target small model First mine pressure prediction results .

[0055] Furthermore, in one embodiment of the present invention, the weight coefficients of each target small model are used... Corresponding to the first mine pressure prediction results By performing a weighted summation, the second mine pressure prediction result is obtained. ,in,

[0056] wherein, represents the prediction value of the instantaneous mine pressure; represents the prediction result of the target small model on the input multi-modal spatio-temporal features of the first mine pressure prediction result; represents the weight of the prediction result of the target small model in the total output.

[0057] Further, in an embodiment of the present application, the corresponding historical scheduling model set and input semantics can be determined based on the target scene type. In an embodiment of the present application, the input semantics can be the historical output semantics corresponding to the historical scheduling set.

[0058] In an embodiment of the present application, the natural language explanation content generated by the large language model can include the current working condition description, the model selection reason, the prediction trend analysis and the risk prompt suggestion. Based on this, the corresponding explanation content can be output by the large language model, which enhances the explainability and engineering usability of the model, provides clear reference basis for the underground dispatcher, and improves the trust degree and decision efficiency of the prediction result.

[0059] In an embodiment of the present application, after obtaining the explanation content and the second mine pressure prediction result through the above steps, the explanation content and the second mine pressure prediction result can be determined as the target mine pressure prediction result, so that the target small model combination prediction under the scene driving can be realized through the large and small model hybrid architecture, and the problem of precision degradation caused by fixed model is solved.

[0060] Further, in an embodiment of the present application, the above method can further include the following steps: Step 105, obtaining the real acquisition value of the target mine pressure prediction result corresponding time; Step 106, determining the prediction error based on the target mine pressure prediction result and the real acquisition value; Step 107, determining whether to adjust the target small model in the large model and the scheduling model set based on the prediction error; Step 108, if it is determined to adjust the target small model in the large model and the scheduling model set, adjusting the target small model in the large model and the scheduling model set based on the target mine pressure prediction result and the real acquisition value.

[0061] In an embodiment of the present application, after obtaining the target mine pressure prediction result, the real acquisition value of the target mine pressure prediction result corresponding time can be obtained , and the prediction error is determined based on the target mine pressure prediction result and the real acquisition value determining the prediction error .

[0062] In an embodiment of the present application, the method for determining whether to adjust the target small model in the large model and the scheduling model set based on the prediction error can include: if the prediction error exceeds the preset threshold, determining to adjust the target small model in the large model and the scheduling model set; and if the prediction error does not exceed the preset threshold, determining not to adjust the target small model in the large model and the scheduling model set.

[0063] Further, in an embodiment of the present application, after determining to adjust the target small model in the large model and the scheduling model set through the above steps, the target small model in the large model and the scheduling model set can be adjusted based on the target mine pressure prediction result and the real collection value.

[0064] Specifically, in an embodiment of the present application, the method for adjusting the target small model in the large model and the scheduling model set based on the target mine pressure prediction result and the real collection value can include the following steps: Step 1081, determining a loss value through a loss function based on the target mine pressure prediction result and the real collection value; Step 1082, adjusting network parameters in the target small model based on the loss value to obtain an adjusted small model; Step 1083, fine-tuning the large model based on the target mine pressure prediction result and the real collection value to obtain a fine-tuned large model.

[0065] In an embodiment of the present application, the content of the above steps 1081-1082 can refer to the prior art, and the present embodiment will not be described here.

[0066] In an embodiment of the present application, the large model can be fine-tuned based on the target mine pressure prediction result and the real collection value to obtain a fine-tuned large model, thereby constructing a complete closed loop from input recognition, model prediction to error feedback, and improving long-term adaptability and generalization ability.

[0067] The mine pressure prediction method based on size model mixing can acquire original data collected by at least one sensor underground, and process the original data to obtain multi-modal spatio-temporal features; based on the multi-modal spatio-temporal features, a target scene type is determined through a large language model; based on the target scene type and the multi-modal spatio-temporal features, a corresponding scheduling model set is determined from a small model library; and based on the multi-modal spatio-temporal features and the target scene type, a target mine pressure prediction result is obtained through the scheduling model set and the large language model. Thus, the current target scene type can be semantically analyzed and recognized through the large language model, and a target small model in the scheduling model set matched with the target scene type is called to perform mine pressure prediction, so that the adaptability and accuracy of the mine pressure prediction result are improved, the explainability and engineering usability are enhanced, definite reference bases are provided for underground dispatchers, and the trustworthiness and decision efficiency of the mine pressure prediction result are improved.

[0068] Figure 2 is a structural schematic diagram of the mine pressure prediction device 200 based on size model mixing according to an embodiment of the present application.

[0069] As shown in Figure 2 , the device can include: a processing module 201 configured to acquire original data collected by at least one sensor underground, and process the original data to obtain multi-modal spatio-temporal features; a first determination module 202 configured to determine a target scene type through a large language model based on the multi-modal spatio-temporal features; a second determination module 203 configured to determine a corresponding scheduling model set from a small model library based on the target scene type and the multi-modal spatio-temporal features; a prediction module 204 configured to obtain a target mine pressure prediction result through the scheduling model set and the large language model based on the multi-modal spatio-temporal features and the target scene type.

[0070] In an embodiment of the present application, the processing module 201 is specifically configured to: resample the original data using a sliding window mechanism to obtain resampled data; align the resampled data to obtain aligned data; embed the aligned data in text form into a first feature vector through an encoder, and determine numerical information in the aligned data as a second feature vector; splice the first feature vector and the second feature vector to obtain the multi-modal spatio-temporal features.

[0071] In an embodiment of the present application, the first determination module 202 is specifically configured to: determine a semantic prompt word corresponding to the multi-modal spatio-temporal features; Input the multi-modal spatio-temporal features and the semantic prompt word into the large language model to obtain a first probability distribution corresponding to the scene type; Based on the first probability distribution, a target scene type is determined.

[0072] In an embodiment of the present application, the second determination module 203 is specifically configured to: Input the target scene type and the multi-modal spatio-temporal features into the target model selector to obtain a second probability distribution of the small models in the small model library for scheduling; Based on the second probability distribution, determine the label order of the target small model for scheduling; Obtain the weight coefficient corresponding to the target small model, and determine the label order of the target small model and the weight coefficient of the target small model as the scheduling model set.

[0073] In an embodiment of the present application, the prediction module 204 is specifically configured to: Sequentially schedule the target small models in the scheduling model set, and obtain first mine pressure prediction results of each target small model based on the multi-modal spatio-temporal features; Based on the weight coefficients of each target small model and the first mine pressure prediction results, perform weighted summation to obtain a second mine pressure prediction result; Based on the target scene type, determine the corresponding historical scheduling model set and input semantics; Based on the historical scheduling model set and the input semantics, generate natural language explanation content through the large language model; Determine the explanation content and the second mine pressure prediction result as the target mine pressure prediction result.

[0074] In an embodiment of the present application, the device is further configured to: Obtain the real acquisition value corresponding to the moment of the target mine pressure prediction result; Based on the target mine pressure prediction result and the real acquisition value, determine a prediction error; Based on the prediction error, determine whether to adjust the large model and the target small models in the scheduling model set; If it is determined to adjust the large model and the target small models in the scheduling model set, adjust the large model and the target small models in the scheduling model set based on the target mine pressure prediction result and the real acquisition value.

[0075] In an embodiment of the present application, the device is further configured to: If the prediction error exceeds a preset threshold, it is determined to adjust the large model and the target small models in the scheduling model set.

[0076] In an embodiment of the present application, the device is further configured to: Based on the target mine pressure prediction result and the real collected value, a loss value is determined through a loss function; Based on the loss value, network parameters in the target small model are adjusted to obtain an adjusted small model. Based on the target mine pressure prediction result and the real collected value, the large model is fine-tuned to obtain a fine-tuned large model.

[0077] The mine pressure prediction device based on the large and small model mixing can perform semantic analysis and identify the current target scene type through the large language model, and call the target small model in the dispatching model set matched with the target scene type to perform mine pressure prediction, thereby improving the adaptability and accuracy of the mine pressure prediction result, enhancing the explainability and engineering usability, providing clear reference basis for the underground dispatcher, and improving the trust degree and decision efficiency of the mine pressure prediction result.

[0078] In the description of the present specification, the description of the terms “one embodiment”, “some embodiments”, “an example”, “a specific example”, or “some examples” means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the present specification and the features of the different embodiments or examples without contradiction.

[0079] In addition, the terms “first”, “second” are only used for description purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features limited by “first”, “second” can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of “a plurality of” is at least two, for example, two, three, etc., unless otherwise specifically limited.

Claims

1. A mine pressure prediction method based on size model mixing, characterized in that, The method comprises: acquiring raw data collected by at least one sensor underground and processing the raw data to obtain multi-modal spatio-temporal features; determining a target scene type through a large language model based on the multi-modal spatio-temporal features; determining a corresponding scheduling model set from a small model library based on the target scene type and the multi-modal spatio-temporal features; obtaining a target mine pressure prediction result through the scheduling model set and the large language model based on the multi-modal spatio-temporal features and the target scene type.

2. The method of claim 1, wherein, The processing of the raw data to obtain multi-modal spatio-temporal features comprises: resampling the raw data using a sliding window mechanism to obtain resampled data; aligning the resampled data to obtain aligned data; embedding text information in the aligned data into a first feature vector through an encoder and determining numerical information in the aligned data as a second feature vector; splicing the first feature vector and the second feature vector to obtain multi-modal spatio-temporal features.

3. The method of claim 1, wherein, The determination of a target scene type through a large language model based on the multi-modal spatio-temporal features comprises: determining a semantic prompt word corresponding to the multi-modal spatio-temporal features; inputting the multi-modal spatio-temporal features and the semantic prompt word into a large language model to obtain a first probability distribution corresponding to a scene type; determining a target scene type based on the first probability distribution.

4. The method of claim 1, wherein, The determination of a corresponding scheduling model set from a small model library based on the target scene type and the multi-modal spatio-temporal features comprises: inputting the target scene type and the multi-modal spatio-temporal features into a target model selector to obtain a second probability distribution of a small model in the small model library for scheduling; determining a label order of a target small model for scheduling based on the second probability distribution; obtaining a weight coefficient corresponding to the target small model and determining the label order of the target small model and the weight coefficient of the target small model as a scheduling model set.

5. The method of claim 3, wherein, The obtaining of a target mine pressure prediction result through the scheduling model set and the large language model based on the multi-modal spatio-temporal features and the target scene type comprises: sequentially scheduling target small models in the scheduling model set to obtain first mine pressure prediction results of the target small models based on the multi-modal spatio-temporal features; performing weighted summation on the weight coefficients of the target small models and the first mine pressure prediction results to obtain second mine pressure prediction results; determining a corresponding historical scheduling model set and input semantics based on the target scene type; generating natural language explanation content through the large language model based on the historical scheduling model set and the input semantics; determining the explanation content and the second mine pressure prediction results as a target mine pressure prediction result.

6. The method of claim 1, wherein, The method further comprises: acquiring a real collection value at a time corresponding to the target mine pressure prediction result; determining a prediction error based on the target mine pressure prediction result and the real collection value; determining whether to adjust the large model and a target small model in the scheduling model set based on the prediction error; If it is determined to adjust the large model and the target small model in the scheduling model set, the large model and the target small model in the scheduling model set are adjusted based on the target mine pressure prediction result and the real collected value.

7. The method of claim 6, wherein, The determining whether to adjust the large model and the target small model in the scheduling model set based on the prediction error comprises: If the prediction error exceeds a preset threshold, it is determined to adjust the large model and the target small model in the scheduling model set.

8. The method of claim 6, wherein, The adjusting the large model and the target small model in the scheduling model set based on the target mine pressure prediction result and the real collected value comprises: Based on the target mine pressure prediction result and the real collected value, a loss value is determined through a loss function; Based on the loss value, network parameters in the target small model are adjusted to obtain an adjusted small model; Based on the target mine pressure prediction result and the real collected value, the large model is fine-tuned to obtain a fine-tuned large model.

9. A size model hybrid based mine pressure prediction device, characterized in that, Comprise: A processing module configured to obtain original data collected by at least one sensor underground and process the original data to obtain multi-modal spatio-temporal features; A first determining module configured to determine a target scene type through a large language model based on the multi-modal spatio-temporal features; A second determining module configured to determine a corresponding scheduling model set from a small model library based on the target scene type and the multi-modal spatio-temporal features; A prediction module configured to obtain a target mine pressure prediction result through the scheduling model set and the large language model based on the multi-modal spatio-temporal features and the target scene type.

10. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the method of any one of claims 1-8.

Images