Control method and system for intelligent gas model

Abstract

The invention discloses a control method and system for an intelligent gas model, and relates to the technical field of intelligent control of gas inflation structures, and the method comprises the steps: periodically obtaining environment data and gas model state data around a target gas model according to a preset time interval; the environment data and the gas model state data are preprocessed, and time sequence data are generated; inputting the time sequence data into the trained LSTM neural network model, and outputting a seam life attenuation coefficient at the seam of the target gas model; judging whether the joint life attenuation coefficient exceeds a preset threshold value or not; if so, generating a fault early warning signal; and according to the fault early warning signal, the air pressure adjustment amount is dynamically calculated through a PID control algorithm, and a control instruction is generated to drive an air pump to adjust the internal air pressure of the target air model. The gas model is controlled through the control method, the safety and reliability of the gas model can be effectively improved, and the service life of the gas model can be effectively prolonged.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control of inflatable structures, and more particularly to a control method and system for an intelligent airform. BACKGROUND

[0002] Airform structures, particularly those used in large-scale temporary constructions and advertising displays, are widely used due to their ease of installation, disassembly, and transportation. These structures are typically made of high-strength materials and rely on internal inflation to maintain their shape and stability. However, as the application scenarios continue to expand, higher requirements are placed on the safety and durability of airforms.

[0003] In the prior art, the monitoring and maintenance of airforms mainly rely on regular manual inspections and automatic alarm systems based on fixed thresholds. Although this method can ensure the safe operation of airforms to some extent, it still has the following shortcomings:

[0004] 1. Poor environmental adaptability: Traditional control systems often lack real-time response capabilities to external environmental changes (such as wind speed, temperature, humidity, etc.). For example, in extreme weather conditions, if the internal pressure of the airform cannot be adjusted in time, it may cause excessive stress on the joints, thereby accelerating aging or directly causing damage.

[0005] 2. Low data processing efficiency: Most existing monitoring systems use simple data analysis methods and cannot effectively utilize historical data for predictive maintenance. This means that potential problems may not be discovered until they occur, increasing maintenance costs and time.

[0006] 3. Lack of adaptive control mechanism: Although traditional PID control algorithms can adjust air pressure based on current conditions, their parameters are usually fixed and cannot adapt to different working conditions. Especially in complex and variable environments, this static adjustment method is particularly inadequate.

[0007] 4. Limitations of early warning systems: Current fault warning systems usually only provide basic alarm functions and lack specific guidance information, such as specific measures to adjust air pressure to avoid further damage. This makes it difficult for operators to take effective measures even if they receive warning signals due to a lack of professional knowledge.

[0008] Therefore, how to improve the safety, reliability, and service life of intelligent airforms through effective control is a problem that needs to be solved by those skilled in the art. SUMMARY

[0009] In view of the above problems, the present application provides a control method and system for an intelligent airform to at least solve some of the technical problems mentioned in the background.

[0010] To achieve the above object, the present application adopts the following technical solutions:

[0011] In one aspect, the present application provides a control method of an intelligent airbag, comprising the following steps:

[0012] Periodically acquiring environmental data and airbag state data around a target airbag at preset time intervals;

[0013] Preprocessing the environmental data and airbag state data to generate time series data;

[0014] Inputting the time series data into a trained LSTM neural network model to output a joint life attenuation coefficient at a joint of the target airbag;

[0015] Judging whether the joint life attenuation coefficient exceeds a preset threshold; if yes, generating a fault warning signal;

[0016] According to the fault warning signal, dynamically calculating a pressure adjustment amount by a PID control algorithm, and generating a control instruction to drive a gas pump to adjust the internal pressure of the target airbag.

[0017] Further, the environmental data includes wind speed, temperature, humidity, and airbag external air pressure;

[0018] The airbag state data includes airbag internal air pressure, airbag support point displacement, joint strain, and joint tension.

[0019] Further, the preset time interval is dynamically adjusted according to the current wind speed; and the preset time interval is inversely proportional to the wind speed.

[0020] Further, the preprocessing of the environmental data and airbag state data to generate time series data specifically includes:

[0021] Respectively cleaning the environmental data and airbag state data;

[0022] Performing timestamp alignment processing on the cleaned environmental data and airbag state data;

[0023] Respectively normalizing the aligned environmental data and airbag state data;

[0024] Taking the current time as the end time of the time window, and according to the preset time window length, a continuous data segment is cut from the preprocessed environmental data and airbag state data to generate time series data.

[0025] Further, the training steps of the LSTM neural network model include:

[0026] Obtain historical environment data and historical airbag state data corresponding to a large number of airbags;

[0027] Preprocess the historical environment data and historical airbag state data to generate a plurality of historical time series data;

[0028] Add a corresponding joint life attenuation coefficient label to each historical time series data;

[0029] The historical time series data is used as input, and the corresponding joint life attenuation coefficient label is used as output to train the LSTM neural network model.

[0030] Further, the weighted combination loss function of the LSTM neural network model is represented as:

[0031]

[0032] Wherein, L represents the weighted combination loss function; α represents the weight coefficient, and α∈[0,1]; represents the mean square error term; represents the smooth L1 loss term; y represents the true joint life attenuation coefficient label; represents the joint life attenuation coefficient predicted by the LSTM neural network model; N represents the total number of historical time series data in a training batch; i represents the i-th historical time series data; y i represents the true joint life attenuation coefficient label corresponding to the i-th historical time series data; represents the joint life attenuation coefficient predicted by the LSTM neural network model based on the i-th historical time series data.

[0033] Further, the LSTM neural network model includes an input layer, a first LSTM layer, a second LSTM layer, a Dropout layer, a fully connected layer and an output layer;

[0034] The time series data is received through the input layer;

[0035] The primary features are extracted from the time series data through the first LSTM layer;

[0036] High-order features are extracted from the primary features through the second LSTM layer;

[0037] The regularization features are obtained by regularizing the high-order features through the Dropout layer;

[0038] The abstract representation of the joint life is generated by spatial mapping the regularization features through the fully connected layer;

[0039] The output layer generates and outputs a seam life attenuation coefficient at the airbag joint.

[0040] Further, it also includes generating maintenance instructions according to the fault warning signal and sending to the user terminal.

[0041] Further, according to the fault warning signal, the fault warning signal is dynamically calculated by a PID control algorithm, and a control instruction is generated to drive the air pump to adjust the internal air pressure of the target airbag; specifically including:

[0042] (1) Obtain the current deviation of the internal air pressure of the airbag from the preset upper limit of the safety air pressure at the current time, and obtain the historical deviation of the internal air pressure of the airbag from the preset upper limit of the safety air pressure at the discrete time points in the current control period; denoted as:

[0043] e(t) = P max -P current

[0044] e(k) = P max -P current (k)

[0045] Where e(t) represents the current deviation; P max represents the upper limit of the preset safety air pressure; P current represents the internal air pressure of the airbag at the current time; e(k) represents the historical deviation at time k; P current (k) represents the internal air pressure of the airbag at time k.

[0046] (2) Calculate the proportional term, integral term and differential term according to the current deviation and the historical deviation sequence; denoted as:

[0047] P out = K p ·e(t)

[0048]

[0049]

[0050] Where P out represents the proportional term; I out represents the integral term; D out represents the differential term; K p represents the PID control parameter preset for the proportional term; K i represents the PID control parameter preset for the integral term; K d represents the PID control parameter preset for the differential term; k=0 represents the starting time of the current control period; t represents the current time; Δt represents the time difference between adjacent control times.

[0051] (3) Superimposing the proportional term, the integral term and the differential term to obtain a gas pressure adjustment amount; denoted as:

[0052] ΔP = P out + I out + D out

[0053] Wherein, ΔP represents the gas pressure adjustment amount;

[0054] (4) According to the fault early warning signal, triggering a compensation factor, correcting the gas pressure adjustment amount, and taking the corrected gas pressure adjustment amount as a control instruction; denoted as:

[0055] ΔP' = β · ΔP

[0056] Wherein, ΔP' represents the control instruction, used to drive the gas pump to adjust the internal gas pressure of the target air mold; β represents the compensation factor.

[0057] On the other hand, the present application provides a control system of intelligent air mold, applying the above method, the system comprises: data acquisition module, pretreatment module, attenuation coefficient prediction module, judgment module and control module;

[0058] The data acquisition module is used to periodically acquire the environment data and air mold state data around the target air mold according to a preset time interval;

[0059] The pretreatment module is used to pretreat the environment data and air mold state data to generate time series data;

[0060] The attenuation coefficient prediction module is used to input the time series data into the trained LSTM neural network model to output the joint life attenuation coefficient of the target air mold joint;

[0061] The judgment module is used to judge whether the joint life attenuation coefficient exceeds a preset threshold; if so, a fault early warning signal is generated;

[0062] The control module is used to dynamically calculate a gas pressure adjustment amount according to the fault early warning signal through a PID control algorithm, and generate a control instruction to drive the gas pump to adjust the internal gas pressure of the target air mold.

[0063] Through the above technical solution, compared with the prior art, the present application provides a control method and system of intelligent air mold, which has the following beneficial effects:

[0064] The application can identify potential failure risks in advance by periodically acquiring environmental data and air mold state data and using an LSTM neural network model to predict the joint life attenuation coefficient, thereby generating a failure warning signal; this enables users to take action before the problem becomes serious, greatly reducing the possibility of air mold sudden failure.

[0065] According to the failure warning signal, the internal air pressure of the air mold is adjusted by real-time calculation of the air pressure adjustment amount through a PID control algorithm; this flexibility ensures that the air mold can maintain the best state under different environments, improves the ability to respond to emergencies, and enhances the safety of use; this is particularly important in application scenarios such as emergency rescue tents.

[0066] The technical solutions of the embodiments of the application will be described in further detail below with reference to the drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the application, and those skilled in the art can obtain other drawings according to the provided drawings without creative labor.

[0068] Figure 1 The control method flowchart of the intelligent air mold provided for the embodiments of the application.

[0069] Figure 2 The control system framework schematic diagram of the intelligent air mold provided for the embodiments of the application. DETAILED DESCRIPTION

[0070] The technical solutions in the embodiments of the application will be described in detail below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only some embodiments of the application, not all embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0071] Embodiment 1

[0072] Embodiment 1 of the application discloses a control method of an intelligent air mold, as shown in Figure 1 The method comprises the following steps:

[0073] S1, periodically acquiring environmental data and air mold state data around the target air mold according to a preset time interval;

[0074] S2, preprocessing the environmental data and air mold state data to generate time series data;

[0075] S3, inputting the time series data into the trained LSTM neural network model, and outputting the seam life attenuation coefficient of the target inflatable mold seam;

[0076] S4. Determine whether the joint life attenuation coefficient exceeds a preset threshold; if so, generate a fault warning signal;

[0077] S5. According to the fault warning signal, the air pressure adjustment amount is dynamically calculated through the PID control algorithm, and a control instruction is generated to drive the air pump to adjust the internal air pressure of the target air model.

[0078] An embodiment of the present invention provides a control method for an intelligent inflatable mold. By periodically monitoring environmental data and inflatable mold status data and using an LSTM neural network model to predict the seam life attenuation coefficient, real-time monitoring of the seam health status can be achieved. When an abnormality is detected, a fault warning signal can be issued in a timely manner, and the internal air pressure can be dynamically adjusted according to the PID control algorithm, thereby effectively preventing safety accidents caused by seam damage.

[0079] The above-mentioned marks S1-S5 are only for the convenience of subsequent description and do not limit the execution order of each step. Next, each of the above-mentioned steps will be described in detail.

[0080] In the above step S1, the environmental data and the state data of the target inflatable model are periodically obtained according to a preset time interval;

[0081] The above environmental data include: wind speed, temperature, humidity and air pressure outside the inflatable model, which are obtained by ultrasonic anemometer, temperature sensor, humidity sensor and barometer arranged on the outer surface of the inflatable model respectively;

[0082] The above-mentioned inflatable model status data includes: air pressure inside the inflatable model, displacement of the inflatable model support points, joint strain and joint tension, which are respectively obtained by pressure sensors arranged inside the inflatable model, displacement sensors arranged at the inflatable model support points, and strain gauges and tension sensors arranged at the inflatable model joints;

[0083] The above-mentioned preset time interval can be dynamically adjusted according to the current wind speed, and the adjustment basis is that the preset time interval is inversely proportional to the wind speed; that is, the greater the wind speed, the shorter the preset time interval; by dynamically adjusting the preset time interval according to the current wind speed, the adaptability of the inflatable model control under different meteorological conditions is improved, especially under high wind speed conditions, the inflatable model status can be monitored more frequently, ensuring the safety of the inflatable model.

[0084] In the above step S2, the environmental data and the gas model state data are preprocessed to generate time series data; specifically, the process includes:

[0085] (1) The environmental data and the mold state data are respectively subjected to data cleaning; specifically, if the environmental data or the mold state data exceeds the range of the corresponding sensor, the data point is removed; for the removed data missing point, linear interpolation of the adjacent time point data is used for filling; if the continuous missing data point exceeds the preset value (for example, 5), it is marked as invalid data section and discarded;

[0086] (2) The cleaned environmental data and mold state data are subjected to timestamp alignment processing;

[0087] (3) The aligned environmental data and mold state data are respectively subjected to normalization processing;

[0088] (4) Taking the current time as the end time of the time window, a continuous data section is cut from the pre-processed environmental data and mold state data according to the preset time window length, to generate time series data;

[0089] Through the steps of cleaning, timestamp alignment, and normalization processing of the environmental data and mold state data, the data quality input into the LSTM neural network model is ensured, and the accuracy of the prediction result is improved.

[0090] In the above step S3, the time series data is input into the trained LSTM neural network model, and the joint life attenuation coefficient of the target mold joint is output; the joint life attenuation coefficient reflects the degradation degree of the joint strain and the joint tension;

[0091] The related content of the above LSTM neural network model will be described below:

[0092] (1) The training steps of the above LSTM neural network model include:

[0093] 1) Obtain a large amount of historical environmental data and historical mold state data corresponding to the mold;

[0094] 2) The historical environmental data and historical mold state data are pre-processed to generate a plurality of historical time series data; specifically:

[0095] The historical environmental data and historical mold state data are sequentially subjected to data cleaning, timestamp alignment, and normalization processing, and then a plurality of continuous data sections are cut from the normalized environmental data and mold state data according to the preset time window length, to generate a plurality of corresponding historical time series data;

[0096] 3) Add the corresponding joint life attenuation coefficient label to each historical time series data;

[0097] 4) The historical time series data is taken as input, and the corresponding joint life attenuation coefficient label is taken as output to realize the training of the LSTM neural network model.

[0098] (2) The weighted combination loss function of the LSTM neural network model is represented as:

[0099]

[0100] Wherein, L represents the weighted combination loss function; α represents the weight coefficient, and α ∈ [0, 1]; represents the mean square error term; represents the smooth L1 loss term; y represents the true joint life attenuation coefficient label; represents the joint life attenuation coefficient predicted by the LSTM neural network model; N represents that there are N historical time series data in a training batch; i represents the i-th historical time series data; y i represents the true joint life attenuation coefficient label corresponding to the i-th historical time series data; represents the joint life attenuation coefficient predicted by the LSTM neural network model based on the i-th historical time series data; The weighted combination loss function combines the mean square error and the smooth L1 loss term, which can reduce the influence of abnormal values on the model while ensuring the stability of the model training, and improve the generalization ability of the model;

[0101] (3) The composition of the above-mentioned LSTM neural network model:

[0102] The above-mentioned LSTM neural network model includes an input layer, a first LSTM layer, a second LSTM layer, a Dropout layer, a fully connected layer and an output layer; Specifically:

[0103] 1) The input layer serves as a data interface and receives preprocessed time series data;

[0104] 2) The first LSTM layer contains 128 memory cells, uses a tanh activation function, and extracts primary features from the original time series data through a gating mechanism (input gate, forget gate and output gate) to capture short period patterns (for example, instantaneous stress changes caused by gusts);

[0105] 3) The second LSTM layer also has 128 units, receives the output features of the first layer, learns long-term dependencies through a deep recurrent network structure, and extracts cross-time step high-order features (for example, material creep effect under the action of high temperature and high humidity);

[0106] 4) The Dropout layer randomly masks 20% of the neurons of the second LSTM layer output with a probability of 0.2, forces the network to establish redundant feature expression, suppresses overfitting, and outputs a regularized feature vector with strong generalization ability;

[0107] 5) The full connection layer performs a nonlinear transformation (ReLU activation) on the 128-dimensional regularized features, maps them to a 32-dimensional feature space through a weight matrix, filters out noise and compresses key information, and forms an abstract representation of the joint life;

[0108] 6) The output layer uses a linear activation function to convert the 32-dimensional feature vector into a scalar value, which quantitatively represents the degree of joint life decay and serves as a core criterion for subsequent warning and control.

[0109] In the above step S4, it is determined whether the joint life decay coefficient exceeds the preset threshold; if it does, a fault warning signal is generated; a maintenance instruction is generated according to the fault warning signal and is sent to the user terminal simultaneously.

[0110] In the above step S5, the PID control algorithm is used to dynamically calculate the air pressure adjustment amount according to the fault warning signal, and a control instruction is generated to drive the air pump to adjust the internal air pressure of the target air mold; specifically including:

[0111] (1) Obtain the current deviation of the internal air pressure of the air mold from the preset upper limit of the safe air pressure at the current time, and obtain the historical deviation of the internal air pressure of the air mold from the preset upper limit of the safe air pressure at discrete time points in the current control period; denoted as:

[0112] e(t) = P max -P current

[0113] e(k) = P max -P current (k)

[0114] Where e(t) represents the current deviation; P max represents the upper limit of the preset safe air pressure; P current represents the internal air pressure of the air mold at the current time; e(k) represents the historical deviation at time k; P current (k) represents the internal air pressure of the air mold at time k.

[0115] (2) Calculate the proportional term, integral term and derivative term according to the current deviation and historical deviation sequence; denoted as:

[0116] P out = K p ·e(t)

[0117]

[0118] Where P out represents the proportional term; I out represents the integral term; D out represents the derivative term; Kp Indicates the PID control parameters preset for the proportional term; K i Indicates the PID control parameters preset for the integral term; K d represents the PID control parameters preset for the differential term; k=0 represents the start time of this control cycle; t represents the current time; Δt represents the time difference between two adjacent control times;

[0119] (3) The proportional term, integral term, and differential term are superimposed to obtain the air pressure adjustment; it is expressed as:

[0120] ΔP=P out +I out +D out

[0121] Wherein, ΔP represents the air pressure adjustment;

[0122] (4) According to the fault warning signal, the compensation factor is triggered to correct the air pressure adjustment amount, and the corrected air pressure adjustment amount is used as the control instruction; it is expressed as:

[0123] ΔP'=β·ΔP

[0124] Wherein, ΔP' represents the control instruction, which is used to drive the air pump to adjust the internal air pressure of the target air model; β represents the compensation factor;

[0125] For example, the control method of the intelligent inflatable model provided in an embodiment of the present invention is applied to emergency rescue tents. When encountering strong winds, the volume of the tent is reduced by dynamically calculating and adjusting the control instructions of the internal air pressure, thereby reducing wind resistance and enhancing the safety and stability of the structure.

[0126] In summary, the embodiment of the present invention provides a control method for an intelligent inflatable model, which can not only periodically obtain environmental data and inflatable model status data around the target inflatable model, but also generate time series data through preprocessing, and input these data into a trained LSTM neural network model to predict the life attenuation coefficient at the joint. Once an abnormality is detected, a fault warning signal will be automatically generated and the air pressure adjustment amount will be dynamically calculated through an improved PID control algorithm, thereby achieving precise regulation of the air pressure inside the inflatable model. In addition, the present invention also takes into account the influence of external environmental factors, dynamically adjusts the time interval for data collection according to the current wind speed, and enhances the adaptability and flexibility of the inflatable model control. At the same time, through in-depth analysis of the fault warning signal, specific maintenance instructions can be provided to users to help users more effectively manage and maintain the inflatable model structure.

[0127] Example 2:

[0128] The embodiment of the present invention provides a control system for an intelligent inflatable model, which applies the control method provided in the above embodiment 1; seeFigure 2 As shown, the control system comprises a data acquisition module, a preprocessing module, an attenuation coefficient prediction module, a judgment module and a control module; wherein:

[0129] The data acquisition module is configured to periodically acquire environmental data and airform state data around the target airform at a preset time interval;

[0130] The preprocessing module is configured to preprocess the environmental data and the airform state data to generate time series data;

[0131] The attenuation coefficient prediction module is configured to input the time series data into the trained LSTM neural network model to output a joint life attenuation coefficient at a joint of the target airform;

[0132] The judgment module is configured to judge whether the joint life attenuation coefficient exceeds a preset threshold; if so, a fault warning signal is generated;

[0133] The control module is configured to dynamically calculate a gas pressure adjustment amount by a PID control algorithm according to the fault warning signal, and generate a control instruction to drive the air pump to adjust the internal gas pressure of the target airform.

[0134] The specific content of each module can be seen from the above embodiment 1.

[0135] Embodiment 3:

[0136] Embodiment 3 of the present application further comprises an LED array in the interior of the target airform, and the LED array dynamically changes the light pattern according to the cloud instruction;

[0137] For example, in the activity promotion scene, a 15m high inflatable arch airform is deployed on site, and the LED array (for example, 120 RGB lamp beads per square meter) is integrated on the inner surface of the airform; when the cloud issues the instruction: light mode = pulse wave, the control module adjusts the duty cycle through the PWM signal to realize the dynamic flowing effect (frequency 2Hz).

[0138] Each embodiment in the specification is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts of each embodiment can be referred to. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant part can be referred to the method part.

[0139] The foregoing description of the disclosed embodiments enables a person skilled in the art to make or use the application. Modifications of these embodiments will occur to persons of skill in the art, and that the appended claims are intended to cover all such modifications that do not depart from the true spirit and scope of the application. Therefore, the application is not limited to the embodiments shown but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A control method for an intelligent inflatable model, characterized in that: The steps include: Periodically acquiring environmental data and status data of the target inflatable model at preset time intervals; Preprocessing the environmental data and the gas model state data to generate time series data; Inputting the time series data into a trained LSTM neural network model, and outputting the seam life attenuation coefficient of the target inflatable mold seam; Determining whether the seam life attenuation coefficient exceeds a preset threshold; If it exceeds, a fault warning signal is generated; According to the fault warning signal, the air pressure adjustment amount is dynamically calculated through the PID control algorithm, and a control instruction is generated to drive the air pump to adjust the internal air pressure of the target air model.

2. The control method of an intelligent inflatable model according to claim 1, characterized in that: The environmental data include: wind speed, temperature, humidity and air pressure outside the inflatable model; The inflatable mold state data includes: air pressure inside the inflatable mold, displacement of the inflatable mold support points, joint strain and joint tension.

3. The control method of an intelligent inflatable model according to claim 1, characterized in that: The preset time interval is dynamically adjusted according to the current wind speed; and the preset time interval is inversely proportional to the wind speed.

4. The control method of an intelligent inflatable model according to claim 1, characterized in that: The preprocessing of the environmental data and the gas model state data to generate time series data specifically includes: performing data cleaning on the environmental data and the gas model state data respectively; Perform timestamp alignment on the environmental data after cleaning and the gas model status data; Normalize the aligned environmental data and gas model state data respectively; Taking the current moment as the end moment of the time window and according to the preset time window length, continuous data segments are intercepted from the preprocessed environmental data and gas model state data to generate time series data.

5. The control method of an intelligent inflatable model according to claim 1, characterized in that: The training steps of the LSTM neural network model include: Obtain a large amount of historical environmental data and historical gas model status data corresponding to the gas model; Preprocessing the historical environmental data and the historical gas model state data to generate a plurality of historical time series data; Adding a corresponding joint life attenuation coefficient label to each of the historical time series data; The historical time series data is used as input and the corresponding seam life attenuation coefficient label is used as output to implement the training of the LSTM neural network model.

6. The control method of an intelligent inflatable model according to claim 1, characterized in that: The weighted combination loss function of the LSTM neural network model is expressed as: Where L represents the weighted combination loss function; α represents the weight coefficient, and α∈[0,1]; represents the mean square error term; represents the smooth L1 loss term; y represents the true seam life attenuation coefficient label; represents the seam life attenuation coefficient predicted by the LSTM neural network model; N represents the total number of N historical time series data in a training batch; i represents the i-th historical time series data; y i Represents the true joint life attenuation coefficient label corresponding to the i-th historical time series data; Represents the seam life attenuation coefficient predicted by the LSTM neural network model based on the i-th historical time series data.

7. The control method of an intelligent inflatable model according to claim 1, characterized in that: The LSTM neural network model includes an input layer, a first LSTM layer, a second LSTM layer, a Dropout layer, a fully connected layer, and an output layer; Receiving the time series data through the input layer; Extracting primary features from the time series data through the first LSTM layer; Extracting high-order features from the primary features through the second LSTM layer; Regularizing the high-order features through the Dropout layer to obtain regularized features; Performing spatial mapping processing on the regularized features through the fully connected layer to generate an abstract representation of the seam life; The output layer generates and outputs a seam life attenuation coefficient at the inflatable mold seam.

8. The control method of an intelligent inflatable model according to claim 1, characterized in that: Also includes: A maintenance instruction is generated according to the fault warning signal and sent to the user terminal.

9. The control method of an intelligent inflatable model according to claim 1, characterized in that: The method comprises: dynamically calculating the air pressure adjustment amount by using a PID control algorithm according to the fault warning signal, and generating a control instruction to drive an air pump to adjust the internal air pressure of the target air model; specifically comprising: (1) Obtain the current deviation between the internal pressure of the inflatable model and the preset upper limit of the safety pressure at the current moment; and obtain the historical deviation between the internal pressure of the inflatable model and the preset upper limit of the safety pressure at discrete time points during this control cycle; expressed as: e(t)=P max -P current e(k)=P max -P current (k) Among them, e(t) represents the current deviation; P max Indicates the preset upper limit of safety pressure; P current represents the internal pressure of the inflatable model at the current moment; e(k) represents the historical deviation at time k; P current (k) represents the internal pressure of the inflatable model at time k; (2) Calculate the proportional term, integral term, and differential term based on the current deviation and the historical deviation sequence; expressed as: P out =K p ·e(t) Among them, P out Represents the proportional term; I out represents the integral term; D out represents the differential term; K p Indicates the PID control parameters preset for the proportional term; K i Indicates the PID control parameters preset for the integral term; K d represents the PID control parameters preset for the differential term; k=0 represents the start time of this control cycle; t represents the current time; Δt represents the time difference between two adjacent control times; (3) The proportional term, integral term, and differential term are superimposed to obtain the air pressure adjustment amount; expressed as: ΔP=P out +I out +D out Where ΔP represents the air pressure adjustment; (4) Triggering a compensation factor according to the fault warning signal, correcting the air pressure adjustment amount, and using the corrected air pressure adjustment amount as a control instruction; expressed as: ΔP'=β·ΔP Wherein, ΔP' represents a control instruction for driving an air pump to adjust the internal air pressure of the target air model; β represents a compensation factor.

10. A control system for an intelligent inflatable model, characterized in that: Applying the method described in any one of claims 1 to 9, the system comprises: a data acquisition module, a preprocessing module, an attenuation coefficient prediction module, a judgment module, and a control module; The data acquisition module is used to periodically acquire environmental data and inflatable model status data around the target inflatable model at preset time intervals; The preprocessing module is used to preprocess the environmental data and the gas model state data to generate time series data; The attenuation coefficient prediction module is used to input the time series data into the trained LSTM neural network model and output the seam life attenuation coefficient of the target inflatable mold seam; The judgment module is used to judge whether the joint life attenuation coefficient exceeds a preset threshold; if so, a fault warning signal is generated; The control module is used to dynamically calculate the air pressure adjustment amount through a PID control algorithm according to the fault warning signal, and generate a control instruction to drive the air pump to adjust the internal air pressure of the target air model.

Images