Health preserving cabin operation parameter adaptive generation method and system based on time series prediction

Abstract

The application discloses a health preserving cabin operation parameter adaptive generation method and system based on time sequence prediction, and relates to the technical field of artificial intelligence. The method comprises the following steps: collecting environment, equipment and user state data in the health preserving cabin to construct multivariate time sequence samples; inputting the multivariate time sequence samples into a pre-trained time sequence state prediction model to predict the cabin state change trend and generate candidate operation parameters accordingly; extracting historical data distribution and combining the prediction trend to dynamically correct the environment and equipment abnormality judgment threshold; using the dynamic threshold to perform safety constraint verification on the candidate operation parameters, filtering the out-of-limit parameters, and then issuing target operation parameters to the equipment end; obtaining an execution response result, generating an abnormality detection result when the dynamic abnormality threshold is triggered, and based on this, feeding the prediction error back to the time sequence state prediction model for network updating, and simultaneously switching to a conservative operation strategy. The method improves the forward-looking, abnormal response timeliness and operation control stability of the health preserving cabin operation parameter generation.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for adaptively generating operating parameters of a health cabin based on time-series prediction. Background Technology

[0002] With the development of the health industry, wellness cabins, as devices that regulate pressure, temperature, humidity, and oxygen levels within the cabin, have been widely adopted. During operation, wellness cabins require comprehensive regulation of multiple environmental parameters, including pressure, temperature, humidity, and oxygen levels, while simultaneously ensuring user comfort and the normal operation of the equipment.

[0003] Currently, the operation and control of health care cabins mostly adopt traditional feedback control strategies. These methods primarily rely on sensors to collect real-time data from within the cabin, triggering underlying equipment adjustments only when parameters deviate from preset target values. However, this passive response mechanism suffers from a certain control lag, making it difficult to cope with the complex dynamic changes caused by the coupling of multiple physical field parameters within the cabin. This can easily lead to fluctuations in the cabin environment, affecting the user experience. Secondly, existing control systems typically use fixed static alarms or anomaly detection thresholds to ensure safety. In practical applications, due to different operating stages, changes in the external environment, or natural wear and tear on equipment, fixed thresholds lack flexibility, often resulting in false alarms or missed alarms.

[0004] Furthermore, conventional systems often separate environmental monitoring, equipment control, and safety assessment, lacking in-depth analysis of the temporal dependencies between historical operational data and multidimensional variables. When faced with situations such as sensor drift, slow equipment response, or sudden load changes, the system cannot adaptively update control logic and safety boundaries, typically requiring manual intervention for troubleshooting and reconfiguration. This not only increases operation and maintenance costs but also limits the intelligence level of the health preservation chamber under complex operating conditions. Therefore, there is an urgent need for a health preservation chamber equipment control scheme that can predict changes in the chamber's state in advance, dynamically adjust safety constraints, and adaptively correct control parameters based on the chamber environment, equipment status, and user feedback. Summary of the Invention

[0005] This invention provides a method and system for adaptively generating operating parameters of a health cabin based on time-series prediction, which solves the problems of control lag, environmental fluctuation, threshold rigidity and lack of error correction capability in the prior art.

[0006] The first aspect of this invention proposes an adaptive generation method for the operating parameters of a health-preserving chamber based on time-series prediction, comprising the following steps:

[0007] Data on the in-cabin environment, equipment operation, and user status are collected, and multivariate time series samples are constructed using these data.

[0008] Multivariate time series samples are input into a pre-trained time series state prediction model to predict the trend of changes in the state of the health cabin over multiple future control cycles.

[0009] Based on the changing trends of the cabin's internal conditions, candidate operating parameters for the health preservation cabin are generated.

[0010] Extract the historical data distribution of multivariate time series samples, and combine it with the trend of changes in the cabin state to dynamically adjust the environmental anomaly judgment threshold for the cabin environment state and the equipment anomaly judgment threshold for the equipment operation state. The dynamic anomaly threshold for the current operation stage is obtained through the environmental anomaly judgment threshold and the equipment anomaly judgment threshold.

[0011] The candidate operating parameters are verified for safety constraints using a dynamic anomaly threshold. Candidate operating parameters that cause the change trend of the cabin state to exceed the dynamic anomaly threshold are filtered out to obtain the target operating parameters. The target operating parameters are then sent to the equipment configured in the health cabin for execution.

[0012] The system acquires the execution response results from the device based on the target operating parameters. When the execution response results trigger the dynamic anomaly threshold, it generates anomaly detection results. Based on the anomaly detection results, it feeds back the prediction error to the time-series state prediction model for network updates and controls the health cabin to automatically switch to a conservative operating strategy.

[0013] In one optional implementation, data on the cabin environment, equipment operation, and user status are collected, and a multivariate time-series sample is constructed using these data, including:

[0014] The sensor array deployed inside the health cabin acquires real-time data on cabin pressure, temperature, humidity, and oxygen levels to determine the cabin's environmental status.

[0015] The local controller of the health cabin collects the operating status of the fan, the status of the ventilation valve, the status of the cabin door opening and closing, the current operating mode, and the cumulative usage time of the equipment to serve as the equipment operating status.

[0016] It receives operation commands, historical parameter adjustment records, and user comfort feedback data based on user interaction terminal input, as user status data;

[0017] The data on the cabin environment status, equipment operation status, and user status are aligned in time and spliced ​​with multidimensional features according to a globally unified timestamp to construct a high-dimensional time series. A sliding time window of preset length is then used to truncate the high-dimensional time series to obtain a multivariate time series sample containing historical time series dependencies.

[0018] In one optional implementation, the time-series state prediction model includes an encoder and a decoder; multivariate time-series samples are input into a pre-trained time-series state prediction model to predict the trend of in-cabin state changes over multiple future control cycles, including:

[0019] Multivariate time series samples are input into the encoder, and multi-source time series latent features are extracted along the time step direction;

[0020] The attention weights of each dimension of features in multivariate time series samples are calculated using a multivariate attention mechanism. The attention weights are then used to perform weighted fusion of multi-source time series latent features to obtain the weighted fused multi-source time series latent features.

[0021] The weighted and fused multi-source temporal latent features are input into the decoder, and a multi-step autoregressive decoding operation is performed to output the future predicted values ​​corresponding to the cabin environment state and equipment operation state step by step.

[0022] The predicted state sequence is composed of future predictions from multiple consecutive time steps, and the predicted state sequence is used as the trend of changes in the cabin state.

[0023] In one optional implementation, candidate operating parameters for the wellness cabin are generated based on the changing trends of the cabin's internal state, including:

[0024] The overall objective function is to optimize the cabin environment to a preset comfort level and reduce equipment energy consumption.

[0025] Based on the trend of changes in the cabin state, an adaptive model predictive control algorithm is used to perform rolling optimization within the preset operating physical space at the equipment end, and solve for the global control action sequence that can minimize the comprehensive objective function.

[0026] Extract the immediate control action corresponding to the next immediately adjacent control cycle from the global control action sequence, and parse the immediate control action into a multi-dimensional control vector for the ventilation intensity, temperature and humidity adjustment range, and equipment start-up and shutdown commands of the health cabin;

[0027] Multidimensional control vectors are used as candidate operating parameters.

[0028] In one optional implementation, historical data distribution of multivariate time-series samples is extracted, and combined with the trend of changes in the cabin state, the environmental anomaly judgment threshold for the cabin environment state and the equipment anomaly judgment threshold for the equipment operating state are dynamically adjusted. The dynamic anomaly threshold for the current operating stage is obtained through the environmental anomaly judgment threshold and the equipment anomaly judgment threshold, including:

[0029] Statistical analysis was performed on the probability density distribution of multivariate time series samples to extract the basic anomaly thresholds of the health preservation chamber under the current operating stage. The basic anomaly thresholds include the basic environmental anomaly thresholds and the basic equipment anomaly thresholds.

[0030] The expected slope of future environmental changes in the wellness cabin is obtained by calculating the difference in predicted states between adjacent future control cycles in the trend of changes in the cabin's state.

[0031] When the absolute value of the expected slope change is greater than the preset slope fluctuation threshold, the upper and lower limits of the basic environment anomaly threshold and the basic equipment anomaly threshold are adaptively scaled according to the gain direction of the expected slope change and the prediction error variance confidence interval of the time series state prediction model.

[0032] The scaled-down basic environmental anomaly threshold and basic equipment anomaly threshold are used as the environmental anomaly judgment threshold and equipment anomaly judgment threshold for the current operating stage, respectively. The dynamic anomaly threshold for the current operating stage is obtained through the environmental anomaly judgment threshold and equipment anomaly judgment threshold.

[0033] In one optional implementation, a dynamic anomaly threshold is used to perform safety constraint verification on candidate operating parameters, filtering out candidate operating parameters that cause the cabin state change trend to exceed the dynamic anomaly threshold, to obtain the target operating parameters, including:

[0034] Candidate operating parameters are superimposed onto the current prediction node of the time series state prediction model to perform virtual forward physics extrapolation and obtain the extrapolation state curve after applying the candidate operating parameters.

[0035] Real-time comparison of the peak data of the simulation state curve in any future simulation control cycle to see if it touches or exceeds the safety boundary of the dynamic anomaly threshold.

[0036] If the peak data does not touch and does not exceed the safety boundary, the candidate operating parameter is determined to have passed the safety constraint verification, and the candidate operating parameter is directly locked as the target operating parameter;

[0037] If the peak data touches or exceeds the safety boundary, the adjustment step size of the candidate operating parameters is proportionally reduced to obtain the attenuation candidate parameters. The virtual forward physical simulation is then performed again using the attenuation candidate parameters until the simulation state curve corresponding to the attenuation candidate parameters is completely within the safety boundary. The attenuation candidate parameters that finally pass the safety constraint verification are used as the target operating parameters.

[0038] In one optional implementation, the execution response result from the device based on the target operating parameters is obtained. When the execution response result triggers a dynamic anomaly threshold, an anomaly detection result is generated. Based on the anomaly detection result, the prediction error is fed back to the time-series state prediction model for network updates, including:

[0039] The actual status feedback data after the device executes the target operating parameters is collected in real time by the local edge controller deployed in the health cabin, and the actual status feedback data is used as the execution response result.

[0040] The numerical deviation between the actual state feedback data and the corresponding periodic cabin state change trend is calculated as the prediction error.

[0041] When the actual status feedback data exceeds the dynamic anomaly threshold, the anomaly is confirmed and an anomaly detection result is generated;

[0042] By using prediction error as a supervision penalty signal, and based on the anomaly detection results, the network weight parameters of the time-series state prediction model are fine-tuned online through backpropagation in the local edge controller to complete the network update of the time-series state prediction model.

[0043] In one alternative implementation, the health preservation chamber is controlled to automatically switch to a conservative operating strategy, including performing the following tiered safety procedures when the execution response triggers a dynamic anomaly threshold:

[0044] When the execution response triggers the environmental anomaly judgment threshold in the dynamic anomaly threshold, the health cabin is controlled to execute the first-level conservative strategy. The first-level conservative strategy includes imposing a hard limit on the maximum allowable adjustment range of the target operating parameters and dynamically increasing the data acquisition frequency for the cabin environment.

[0045] When the execution response triggers the device anomaly judgment threshold in the dynamic anomaly threshold, the control health cabin will execute the second-level conservative strategy.

[0046] The second-level conservative strategy includes triggering the equipment hardware safety interlock mechanism of the health cabin, forcibly cutting off the parameter adjustment control authority, activating the physical backup ventilation valve for forced ventilation and depressurization, and simultaneously generating alarm reminder instructions.

[0047] In one alternative implementation, the temporal state prediction model is a composite neural network that combines a convolutional autoencoder and a long short-term memory network.

[0048] The encoder includes a convolutional autoencoder. The process of inputting multivariate time series samples into the encoder and extracting multi-source time series latent features along the time step direction includes: using a convolutional autoencoder to perform convolutional feature extraction and nonlinear dimensionality reduction on the multivariate time series samples in the spatial dimension, so as to remove high-frequency background noise in the multivariate time series samples and obtain a noise-free feature sequence after dimensionality reduction.

[0049] The decoder includes a long short-term memory network. The noiseless feature sequence is input into the long short-term memory network to perform recursive feature calculation in the time dimension, so as to obtain multi-source temporal implicit features that characterize the spatiotemporal coupling relationship of multi-physics parameters in the health cabin.

[0050] A second aspect of this invention proposes an adaptive generation system for the operating parameters of a health-preserving chamber based on time-series prediction. The system includes:

[0051] The time series sample construction module is used to collect data on the in-cabin environment, equipment operation, and user status of the health cabin, and to construct multivariate time series samples based on the in-cabin environment, equipment operation, and user status data.

[0052] The state change prediction module is used to input multivariate time series samples into a pre-trained time series state prediction model to predict the state change trend of the health cabin in multiple future control cycles.

[0053] The candidate parameter generation module is used to generate candidate operating parameters for the health preservation chamber based on the changing trends of the state inside the chamber.

[0054] The anomaly threshold correction module is used to extract the historical data distribution of multivariate time series samples and, in combination with the trend of changes in the cabin state, dynamically correct the environmental anomaly judgment threshold for the cabin environment state and the equipment anomaly judgment threshold for the equipment operating state. The dynamic anomaly threshold for the current operating stage is obtained through the environmental anomaly judgment threshold and the equipment anomaly judgment threshold.

[0055] The safety constraint filtering module is used to perform safety constraint verification on candidate operating parameters using dynamic anomaly thresholds, filter out candidate operating parameters that cause the change trend of the state inside the chamber to exceed the dynamic anomaly threshold, obtain the target operating parameters, and send the target operating parameters to the device configured in the health chamber for execution.

[0056] The feedback and safety interlock module is used to obtain the execution response results of the device based on the target operating parameters. When the execution response results trigger the dynamic abnormal threshold, an abnormal detection result is generated. Based on the abnormal detection result, the prediction error is fed back to the time-series state prediction model for network update, and the health preservation chamber is controlled to automatically switch to a conservative operation strategy.

[0057] A third aspect of this invention provides an electronic device, comprising: a memory and a processor;

[0058] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the adaptive generation method of health cabin operating parameters based on time-series prediction as proposed in the foregoing embodiments.

[0059] The fourth aspect of this invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the adaptive generation method for health cabin operating parameters based on time-series prediction as proposed in the foregoing embodiments.

[0060] Beneficial Effects: This invention provides a method and system for adaptively generating operating parameters of a health-preserving chamber based on time-series prediction. By constructing multivariate time-series samples and introducing an artificial intelligence-based time-series state prediction model, it transforms traditional passive response control into active feedforward regulation. Candidate operating parameters are generated in advance based on predicted trends in the chamber's state, overcoming the shortcomings of control lag and environmental volatility in existing technologies, and improving the smoothness of chamber regulation. Simultaneously, by integrating historical data distribution with predicted trends, the system dynamically corrects the anomaly detection thresholds for the environment and equipment, breaking the rigid limitations of static fixed thresholds and improving the accuracy of anomaly detection under complex and variable operating conditions. Furthermore, dynamic anomaly thresholds are used to pre-check the safety constraints of candidate parameters, filtering potential limit-crossing risks before parameter execution, reducing the risk of parameter limit exceeding during equipment operation. When the actual execution response triggers the anomaly threshold, the system can automatically extract the prediction error and feed it back to the prediction model for network parameter updates, while simultaneously switching to a conservative operating strategy. This closed-loop mechanism enables the control system to have continuous adaptive correction capabilities, effectively cope with external interference such as sensor drift or equipment aging, and improve the control stability and abnormal response capability of the health chamber during long-term operation. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention;

[0062] Figure 2 This is a flowchart illustrating the steps of an adaptive generation method for operating parameters of a health cabin based on time-series prediction, provided in an embodiment of the present invention.

[0063] Figure 3 This is a schematic diagram of the functional modules of an adaptive generation system for health cabin operating parameters based on time-series prediction, provided in an embodiment of the present invention. Detailed Implementation

[0064] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0065] The present invention will be further described below with reference to the accompanying drawings.

[0066] Reference Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention.

[0067] like Figure 1 As shown, the electronic device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (Wi-Fi). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0068] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0069] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a data storage module, a network communication module, a user interface module, and computer programs.

[0070] exist Figure 1 In the electronic device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of the present invention can be set in the electronic device. The electronic device calls the adaptive generation system of health cabin operation parameters based on time-series prediction stored in the memory 1005 through the processor 1001, and executes the adaptive generation method of health cabin operation parameters based on time-series prediction provided in the embodiment of the present invention.

[0071] The following is combined Figure 2This paper elaborates on the specific embodiments of the present invention. In the actual operation of health chambers (such as lightweight hyperbaric oxygen chambers and intelligent micro-hyperbaric chambers), the internal environment of the chamber is a complex nonlinear space with strong coupling of multiple physical fields (temperature, humidity, pressure, gas concentration, etc.), and the execution response of the chamber hardware (such as fans and solenoid valves) usually exhibits physical hysteresis. Traditional control methods that rely on passive feedback adjustment based on current sensor values ​​are ill-suited to this complexity and uncertainty, easily leading to overshoot or drastic fluctuations in the chamber's environmental parameters. Furthermore, traditional equipment often uses fixed static alarm thresholds, which cannot adapt to the zero-point drift of sensors over time or the personalized needs of different users at different operating stages, easily causing false alarms or missed alarms.

[0072] To overcome the technical shortcomings of existing technologies, such as control lag, environmental volatility, inflexible anomaly detection, and lack of self-correction capabilities, the first aspect of this invention proposes an adaptive generation method for the operating parameters of a health-preservation chamber based on time-series prediction. This method is executed by an edge computing node or main control chip configured locally in the health-preservation chamber and includes the following steps:

[0073] S101 collects data on the cabin environment, equipment operation, and user status, and constructs a multivariate time series sample based on these data.

[0074] In one optional implementation, in order to accurately capture transient changes in the cabin environment and the physical response characteristics of the equipment, step S101 specifically includes the following sub-steps:

[0075] S1011 utilizes a sensor array deployed within the health-preserving chamber to acquire real-time data on chamber pressure, temperature, humidity, and oxygen-related conditions, serving as the environmental status within the chamber. Specifically, the chamber is equipped with a distributed array of high-precision micro-pressure sensors, integrated temperature and humidity probes, and oxygen concentration sensors. These sensor arrays collect environmental data in real-time at a preset high-frequency sampling rate (e.g., 10 times per second). The acquired oxygen-related data includes the chamber's oxygen concentration and partial pressure, used to characterize the gaseous environment within the chamber. These physical parameters constitute the fundamental characteristics of the chamber's current microenvironment.

[0076] S1012 collects data on the fan operating status, ventilation valve status, cabin door opening / closing status, current operating mode, and cumulative equipment usage time through the local controller of the maintenance chamber, serving as the equipment operating status. To enable subsequent predictive models to understand the causal relationship between environmental changes and equipment actions, the local controller needs to synchronously record the operating conditions of the underlying actuators. Fan operating status includes the fan's current speed (revolutions per minute, RPM) and its real-time operating current; ventilation valve status includes the opening percentage of the inlet and outlet solenoid valves; cabin door opening / closing status is a discrete Boolean value (closed or open) representing the chamber's sealing performance; current operating modes include, for example, "pressure stabilization mode," "pressure boosting mode," "pressure deflating mode," or "sleep mode"; and cumulative equipment usage time implicitly characterizes the degree of mechanical wear or aging conditions such as filter clogging.

[0077] S1013 receives operation commands, historical parameter adjustment records, and user comfort feedback data based on user interaction terminal input, as user status data. User status data reflects the individual tolerance and subjective intentions of the cabin occupants. Operation commands include target pressure or temperature expectations set by the user via the touchscreen; historical parameter adjustment records characterize the user's adjustment preferences over a specific time period; user comfort feedback data can be user comfort feedback tags actively submitted by the user via terminal buttons, including perceived temperature and humidity, perceived pressure changes, and overall comfort rating.

[0078] S1014. The data on the cabin environment status, equipment operation status and user status are time-series aligned and multi-dimensional features are spliced ​​according to the globally unified timestamp to construct a high-dimensional time series. The high-dimensional time series is then truncated using a sliding time window of preset length to obtain a multivariate time series sample containing historical time series dependencies.

[0079] Furthermore, S1014 specifically includes: Since the sampling frequencies of various sensors and controllers may differ, this method first establishes a reference clock signal and uses linear interpolation or a zero-order hold algorithm to resample and fill in missing data, thereby achieving strict timestamp alignment of multimodal data. Subsequently, all numerical state data at a given moment and categorical state data processed by one-hot encoding are vectorized and concatenated to form the feature vector for that moment. As time progresses, these feature vectors are stacked in chronological order to form a high-dimensional time series matrix. Next, a sliding time window with a length equal to the observation history step size (e.g., the past 60 consecutive control cycles) is defined. This window slides forward along the time axis, extracting a two-dimensional matrix slice as the current multivariate time series sample each time it slides. This sample fully encompasses the dynamic evolution trajectory of the multivariate variables in the health cabin over a past period and their inter-variable coupling relationships.

[0080] S102 inputs multivariate time series samples into a pre-trained time series state prediction model to predict the trend of changes in the state of the health cabin over multiple future control cycles.

[0081] The training samples include historical cabin environment states, equipment operating states, user state data, and corresponding future state sequences; the input is multivariate time-series samples, and the output is the predicted values ​​of cabin environment states and equipment operating states for multiple future control cycles; the training loss includes at least the mean square error between the predicted values ​​and the actual feedback values. In one optional implementation, the time-series state prediction model includes an encoder and a decoder, and this time-series state prediction model is a composite neural network combining a convolutional autoencoder (CAE) and a long short-term memory (LSTM) network. This step S102 specifically includes the following sub-steps:

[0082] S1021, input the multivariate time-series samples into the encoder and extract multi-source temporal latent features along the time step direction. The encoder includes a convolutional autoencoder. This sub-step specifically includes: using the convolutional autoencoder to perform spatial convolutional feature extraction and nonlinear dimensionality reduction on the multivariate time-series samples to remove high-frequency background noise and obtain a noise-free feature sequence after dimensionality reduction. In practical engineering, Industrial Internet of Things (IIoT) data collected by sensors is often accompanied by a large amount of random electromagnetic interference noise. The convolutional autoencoder uses a series of one-dimensional convolutional kernels to slide along the feature dimension, capturing the local spatial interaction features between different physical quantities (such as temperature and pressure), and filtering out high-frequency disturbances through pooling operations. Subsequently, compression is performed through the autoencoder's bottleneck layer, forcing the model to learn a low-dimensional manifold representing the most essential state of the system, thereby outputting a smooth and robust noise-free feature sequence.

[0083] S1022 utilizes a multi-variable attention mechanism to calculate the attention weights of features in each dimension of the multivariate time-series samples. These attention weights are then used to weightedly fuse the multi-source time-series latent features, yielding a weighted fused multi-source time-series latent feature. Considering that the contribution of each environmental parameter to the future state dynamically changes under different operating modes (e.g., during the pressurization phase, the weights of fan speed and intake valve opening should be much greater than the current temperature), this method constructs an attention scoring network based on a deep multilayer perceptron (MLP). This network takes the aforementioned noiseless feature sequence as input, calculates dynamic attention scores for different feature channels, and transforms them into an attention weight distribution using a normalization function (Softmax). Subsequently, this attention weight matrix is ​​element-wise multiplied and weighted with the noiseless feature sequence, enabling the model to focus on the core features that play a decisive role in the system's evolution during subsequent predictions.

[0084] S1023: The weighted fused multi-source temporal latent features are input into the decoder to perform multi-step autoregressive decoding, outputting future predicted values ​​corresponding to the cabin environment state and equipment operating state at each time step. The decoder includes a Long Short-Term Memory (LSTM) network. Specifically, this sub-step involves inputting the weighted fused multi-source temporal latent features into the LSM network for recursive feature calculation along the time dimension. The LSM network contains forget gates, input gates, and output gates, effectively solving the gradient vanishing problem in long sequence propagation and accurately capturing long-distance spatiotemporal coupling relationships between variables. In the autoregressive decoding stage, the decoder receives the output of the previous time step as the input to the current time step, and, combined with the latent state vector, iteratively deduces the specific environmental values ​​(such as the pressure rise curve and temperature change curve within the next 10 seconds) for the first, second, and up to the Nth control cycles.

[0085] S1024: The predicted future values ​​of multiple consecutive time steps are combined into a predicted state sequence, which is then used as the trend of in-cabin state changes. This predicted state trend is essentially a set of multi-dimensional time series matrices, which not only depict the future trend of the macroscopic environment inside the cabin, but also implicitly contain the expected operating inertia of the equipment under these environmental changes, providing extremely important data support for subsequent feedforward control.

[0086] S103 generates candidate operating parameters for the health preservation chamber based on the changing trends of the internal state.

[0087] In one alternative implementation, to achieve optimal scheduling for a complex multivariable system, the nonlinear control problem is transformed into a constrained optimization problem. Step S103 specifically includes the following sub-steps:

[0088] S1031 uses optimizing the cabin environment to a preset comfort level and reducing equipment energy consumption as the comprehensive objective function. Specifically, the comprehensive objective function is a cost function, mainly composed of two penalty costs: the first is "state tracking error penalty," which calculates the sum of squares of deviations between the predicted cabin pressure, temperature, and other state values ​​and the user-set ideal comfort level; the larger the deviation, the higher the penalty value. The second is "control action energy consumption and wear penalty," which calculates the fan power change rate and energy loss caused by frequent valve opening and closing, aiming to suppress large-scale equipment vibration.

[0089] The comprehensive objective function can be mathematically modeled in the following way:

[0090]

[0091] In the formula, This represents the objective function value of the overall cost that needs to be minimized. This represents the forecast horizon, which is the total number of control periods predicted into the future. This represents the control horizon, which is the number of future cycles allowed to solve for changes in the action. Indicates the current moment Predicted future The cabin state vector for each cycle (i.e., the aforementioned cabin state change trend). Indicates the future number A preset comfort reference trajectory vector for each cycle; Indicates the future number The control action increment per cycle (such as the change in fan speed). This is the state error weight matrix, used to measure the relative importance of different state variables (e.g., pressure is more important than temperature). The incremental weight matrix is ​​used to control the intensity of equipment movements.

[0092] S1032, based on the changing trends of the cabin state, employs an adaptive model predictive control (MPC) algorithm to perform rolling optimization within a preset operational physical space at the equipment end, solving for a sequence of target control actions that satisfies the preset optimization conditions for the comprehensive objective function. The preset operational physical space is a feasible solution space determined by the hardware limits of the equipment (e.g., the maximum speed of the fan cannot exceed the rated parameters, and the valve opening can only be between 0% and 100%). The optimization solver will search for a set of future... The optimal control increment sequence within the step makes the above comprehensive objective function minimize.

[0093] S1033 extracts the immediate control action corresponding to the next adjacent control cycle from the global control action sequence, and parses the immediate control action into a multi-dimensional control vector for the ventilation intensity, temperature and humidity adjustment range, and equipment start-up and shutdown commands of the health cabin. The core mechanism of model predictive control lies in "rolling optimization," that is, although the actions for multiple future steps are calculated, only the first action in the sequence is extracted for actual application, and the remaining actions will be recalculated when the next control cycle arrives, thereby effectively compensating for the uncertainty disturbances of the system.

[0094] S1034 uses the multidimensional control vector as candidate operating parameters. These candidate operating parameters are theoretically optimal solutions, but have not yet undergone substantial filtering and verification by the dynamic safety boundary.

[0095] S104. Extract the historical data distribution of multivariate time series samples, and combine it with the trend of changes in the cabin state to dynamically correct the environmental anomaly judgment threshold for the cabin environment state and the equipment anomaly judgment threshold for the equipment operating state. The dynamic anomaly threshold for the current operating stage is obtained through the environmental anomaly judgment threshold and the equipment anomaly judgment threshold.

[0096] During the operation of the health maintenance chamber, fixed thresholds are the root cause of false alarms. For example, a rapid pressure rise is normal during the initial pressurization phase, but during the stabilization phase, the same pressure rise rate indicates a leak or venting failure. Therefore, this embodiment introduces an adaptive dynamic threshold determination mechanism based on the fusion of statistics and trend prediction. In one optional implementation, step S104 specifically includes the following sub-steps:

[0097] S1041, Statistical analysis is performed on the historical distributions of pressure, temperature, humidity, oxygen-related states, fan states, and valve states in the multivariate time-series samples to extract the basic anomaly thresholds for the health preservation chamber under the current operating stage. These basic anomaly thresholds include basic environmental anomaly thresholds and basic equipment anomaly thresholds. Specifically, nonparametric statistical methods such as kernel density estimation are used to fit the distribution of historical sample data, calculating the reasonable upper and lower limits of fluctuation for each parameter within a given normal confidence interval (e.g., 99% or 3-Sigma criterion), which serve as the static basic reference boundary for the current moment.

[0098] S1042: By calculating the predicted state difference between adjacent future control cycles in the trend of changes in the cabin state, the expected slope of future environmental changes in the health cabin is obtained. Using the future multi-step predicted values ​​output in step S102, the rate of change of future cabin pressure or temperature and humidity on the time axis is obtained through differential calculation. This slope directly reflects the intensity and direction of system evolution.

[0099] S1043, when the absolute value of the expected slope change is greater than the preset slope fluctuation threshold, the upper and lower limits of the basic environmental anomaly threshold and the basic equipment anomaly threshold are adaptively scaled according to the gain direction of the expected slope change and the prediction error variance confidence interval of the time-series state prediction model. Taking the "emergency depressurization" condition of the health cabin as an example, when a sharp drop in pressure is predicted in the future (the expected slope change is large and negative), if the narrow anomaly threshold of the stabilization period is still used, the system will inevitably generate a large number of "low pressure" false alarms. At this time, this step will automatically widen the lower limit of the basic environmental anomaly threshold significantly downward in the expected depressurization direction, and at the same time widen the upper limit of the equipment anomaly threshold of the fan speed to tolerate the strong airflow disturbance caused by the full opening of the pressure relief valve. This method of adjusting the upper and lower limits of the dynamic anomaly threshold in advance according to future trends ensures that the threshold envelope can always closely follow the actual physical evolution trajectory of the system.

[0100] S1044, the scaled basic environmental anomaly threshold and basic equipment anomaly threshold are used as the environmental anomaly judgment threshold and equipment anomaly judgment threshold for the current operating stage, respectively, and the dynamic anomaly threshold for the current operating stage is obtained through the environmental anomaly judgment threshold and equipment anomaly judgment threshold.

[0101] S105, using the dynamic anomaly threshold to perform safety constraint verification on the candidate operating parameters, filtering out candidate operating parameters that cause the change trend of the cabin state to exceed the dynamic anomaly threshold, obtaining the target operating parameters, and sending the target operating parameters to the equipment configured in the health cabin for execution.

[0102] In one alternative implementation, to prevent AI-generated candidate parameters from leading to unknown extreme situations, this step involves a simulation exercise before issuing the command, specifically including the following sub-steps:

[0103] S1051, the candidate operating parameters are superimposed onto the current prediction node of the time-series state prediction model to perform virtual forward physics extrapolation, and the extrapolated state curve after applying the candidate operating parameters is obtained. That is, assuming that the underlying equipment is executing completely according to the candidate parameters (such as increasing the fan speed to the limit), the powerful nonlinear mapping capability of the prediction model is used to prospectively extrapolate what kind of deformation of the environmental curve will occur in the next few seconds due to such extreme operation.

[0104] S1052, compare in real time whether the peak data of the simulation state curve in any future simulation control cycle touches or exceeds the safety boundary of the dynamic anomaly threshold.

[0105] S1053, if the peak data does not touch and does not exceed the safety boundary, the candidate operating parameter is determined to have passed the safety constraint verification, and the candidate operating parameter is directly locked as the target operating parameter.

[0106] S1054: If the peak data touches or exceeds the safety boundary, the adjustment step size of the candidate operating parameters is proportionally attenuated to obtain attenuation candidate parameters. These attenuation candidate parameters are then used to perform a virtual forward physical simulation again until the simulation state curve corresponding to the attenuation candidate parameter is completely within the safety boundary. The attenuation candidate parameter that finally passes the safety constraint verification is taken as the target operating parameter. For example, the original plan was to open the intake valve to 80%, but simulation showed that this would cause the local oxygen partial pressure to exceed the limit after 3 seconds (exceeding the dynamic safety upper limit). The system will then automatically attenuate the valve opening target to 60% and re-simulate. If it still exceeds the limit, it will continue to reduce it to 48% according to a preset ratio (e.g., an attenuation coefficient of 0.8), iterating until the simulated oxygen partial pressure curve falls completely within the safety envelope. Finally, the target operating parameter that passes the safety constraint verification is converted into a control signal and sent to the physical hardware layer composed of sensors and actuators via the local bus for execution.

[0107] S106, obtain the execution response result of the device based on the target operating parameters, generate the anomaly detection result when the execution response result triggers the dynamic anomaly threshold, feed back the prediction error to the time-series state prediction model based on the anomaly detection result for network update, and control the health cabin to automatically switch to the conservative operation strategy.

[0108] In the real physical world, even parameters that have undergone extremely rigorous simulation may fail during execution due to sudden hardware malfunctions (such as valve jamming), causing actual performance to deviate from expectations. Therefore, this method introduces a closed-loop mechanism based on edge-based prediction, anomaly detection, and online fine-tuning. In one optional implementation, step S106 specifically includes:

[0109] S1061 uses a local edge controller deployed in the health care cabin to collect real-time feedback data on the actual status of the device after executing the target operating parameters, and uses this feedback data as the execution response result. This is equivalent to immediately "reading back" the actual execution effect of the device after issuing the command.

[0110] S1062, calculate the numerical deviation between the actual state feedback data and the corresponding periodic cabin state change trend (i.e., the predicted value generated above), as the prediction error.

[0111] S1063, when the actual status feedback data exceeds the dynamic anomaly threshold, an anomaly is confirmed and an anomaly detection result is generated. Exceeding the threshold here means that not only has the actual performance of the device deviated from the model prediction, but the degree of deviation has exceeded the tolerance threshold after the system's adaptive relaxation, indicating a substantial fault or sudden environmental change.

[0112] In step S1064, using prediction error as a supervisory penalty signal, the weight parameters of the replica network of the time-series state prediction model are fine-tuned online in the local edge controller based on the anomaly detection results. After fine-tuning, the running model is updated after verification by preset validation samples or safety constraints. If the verification fails, the model parameters are rolled back to the previous state, thus completing the network update of the time-series state prediction model. In traditional industrial applications, once a model is deployed, it remains fixed. Once long-term sensor aging and drift occur, the prediction accuracy drops precipitously. In this embodiment, however, the edge controller immediately triggers an online lightweight learning mechanism after capturing severely biased anomaly slice data. It uses the calculated deviation as a loss function, calculates the partial derivatives along the backpropagation path of the neural network, and updates the weight matrices of the CNN convolutional kernels and LSTM gated units in the network. This continuous, lifelong adaptive learning capability allows the prediction model to continuously absorb the latest hardware wear characteristics, achieving synchronous matching between model evolution and physical entity aging.

[0113] Furthermore, in an alternative implementation, to ensure absolute personnel safety, the health cabin automatically switches to a conservative operating strategy, including performing the following tiered safety procedures when the response result triggers a dynamic anomaly threshold:

[0114] S1065, when the execution response triggers the environmental anomaly judgment threshold in the dynamic anomaly threshold (e.g., only the humidity inside the chamber is slightly exceeded), the health chamber is controlled to execute the first-level conservative strategy. The first-level conservative strategy includes imposing limiting constraints on the maximum adjustment range and unit time adjustment rate of the target operating parameters (e.g., locking the control step size of all devices to no more than 10% of the rated value), and dynamically increasing the data acquisition frequency for the state of the chamber environment (e.g., from 10 times per second to 50 times per second) in order to more closely monitor the environmental evolution and prevent the situation from deteriorating.

[0115] S1066 When the execution response result triggers the equipment anomaly judgment threshold in the dynamic anomaly threshold (for example, the fan speed feedback is zero, or the pressure increases abnormally, indicating that there is a risk that the equipment execution state or the cabin environment state exceeds the preset safety boundary, the health care cabin is controlled to execute the second level conservative strategy in an out-of-level manner).

[0116] S1067, the second-level conservative strategy includes triggering the equipment hardware safety interlock mechanism of the health cabin, forcibly cutting off the parameter adjustment control authority (that is, directly taking over the control authority to the underlying physical hardware protection circuit), activating the physical backup ventilation valve for forced ventilation and depressurization, and simultaneously generating alarm reminder instructions to push to the administrator terminal and the cabin voice alarm, and executing equipment protection and alarm prompts with preset safety priorities.

[0117] This invention also provides an adaptive generation system for the operating parameters of a health-preserving chamber based on time-series prediction, referring to... Figure 3 The diagram illustrates a functional block diagram of a time-series prediction-based adaptive generation system for health cabin operating parameters 300 according to the present invention. This system may include the following modules:

[0118] The time series sample construction module 301 is used to collect data on the cabin environment status, equipment operation status and user status of the health cabin, and to construct multivariate time series samples based on the cabin environment status, equipment operation status and user status data.

[0119] The state change prediction module 302 is used to input multivariate time series samples into a pre-trained time series state prediction model to predict the state change trend of the health cabin in multiple future control cycles.

[0120] The candidate parameter generation module 303 is used to generate candidate operating parameters for the health preservation chamber based on the trend of changes in the state inside the chamber.

[0121] The abnormal threshold correction module 304 is used to extract the historical data distribution of multivariate time series samples and, in combination with the trend of changes in the cabin state, dynamically correct the environmental abnormality judgment threshold for the cabin environment state and the equipment abnormality judgment threshold for the equipment operating state. The dynamic abnormality threshold for the current operating stage is obtained through the environmental abnormality judgment threshold and the equipment abnormality judgment threshold.

[0122] The safety constraint filtering module 305 is used to perform safety constraint verification on candidate operating parameters using dynamic anomaly thresholds, filter out candidate operating parameters that cause the trend of changes in the state inside the chamber to exceed the dynamic anomaly thresholds, obtain target operating parameters, and send the target operating parameters to the device configured in the health chamber for execution.

[0123] The feedback and safety interlock module 306 is used to obtain the execution response results of the device based on the target operating parameters. When the execution response results trigger the dynamic abnormal threshold, an abnormal detection result is generated. Based on the abnormal detection result, the prediction error is fed back to the time-series state prediction model for network update, and the health preservation chamber is controlled to automatically switch to a conservative operation strategy.

[0124] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus.

[0125] Memory, used to store computer programs;

[0126] When the processor executes the program stored in the memory, it implements the adaptive generation method for health cabin operating parameters based on time-series prediction of the present invention.

[0127] The communication bus mentioned in the aforementioned electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the figure, but this does not mean that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned terminal and other devices. The memory can include RAM, or it can include NVM, such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor.

[0128] The processors mentioned above can be general-purpose processors, including CPUs, network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0129] Furthermore, to achieve the above objectives, embodiments of the present invention also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the adaptive generation method for health cabin operating parameters based on time-series prediction, as described in the embodiments of the present invention.

[0130] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products embodied on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0131] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, electronic devices, apparatuses, and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0132] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for adaptively generating operating parameters of a health-preserving chamber based on time-series prediction, characterized in that, Includes the following steps: Data on the in-cabin environment, equipment operation, and user status are collected, and multivariate time-series samples are constructed using the in-cabin environment, equipment operation, and user status data. The multivariate time series samples are input into a pre-trained time series state prediction model to predict the trend of the cabin state change in the future multiple control cycles. Based on the trend of changes in the cabin's internal state, candidate operating parameters for the health-preserving cabin are generated; Extract the historical data distribution of the multivariate time series sample, and combine it with the trend of the change in the cabin state to dynamically correct the environmental anomaly judgment threshold for the cabin environment state and the equipment anomaly judgment threshold for the equipment operating state. The dynamic anomaly threshold for the current operating stage is obtained through the environmental anomaly judgment threshold and the equipment anomaly judgment threshold. The candidate operating parameters are verified for safety constraints using the dynamic anomaly threshold. Candidate operating parameters that cause the change trend of the cabin state to exceed the dynamic anomaly threshold are filtered out to obtain the target operating parameters. The target operating parameters are then sent to the device configured in the health cabin for execution. The device acquires the execution response result based on the target operating parameters. When the execution response result triggers the dynamic anomaly threshold, an anomaly detection result is generated. Based on the anomaly detection result, the prediction error is fed back to the time-series state prediction model for network update, and the health cabin is controlled to automatically switch to a conservative operating strategy.

2. The adaptive generation method for health cabin operating parameters based on time-series prediction according to claim 1, characterized in that, The system collects data on the cabin environment, equipment operation, and user status, and constructs multivariate time-series samples using these data, including: The sensor array deployed inside the health cabin acquires real-time data on cabin pressure, temperature, humidity, and oxygen levels to serve as the environmental status of the cabin. The local controller of the health cabin collects the fan operating status, air exchange valve status, cabin door opening and closing status, current operating mode, and cumulative equipment usage time as the equipment operating status. The system receives operation commands, historical parameter adjustment records, and user comfort feedback data based on user interaction terminal input, as the user status data. The data on the cabin environment status, equipment operation status, and user status are time-series aligned and multidimensional features are stitched together according to a globally unified timestamp to construct a high-dimensional time series. The high-dimensional time series is then truncated using a sliding time window of a preset length to obtain the multivariate time series sample containing historical time series dependencies.

3. The adaptive generation method for health cabin operating parameters based on time-series prediction according to claim 1, characterized in that, The time-series state prediction model includes an encoder and a decoder; the step of inputting the multivariate time-series samples into the pre-trained time-series state prediction model to predict the trend of the cabin's state changes over multiple future control cycles includes: The multivariate time series samples are input into the encoder, and multi-source time series latent features are extracted along the time step direction; The attention weights of each dimension of features in the multivariate time series samples are calculated using a multivariate attention mechanism. The attention weights are then used to perform weighted fusion of the multi-source time series latent features to obtain the weighted fused multi-source time series latent features. The weighted and fused multi-source temporal latent features are input into the decoder, and a multi-step autoregressive decoding operation is performed to output the future predicted values ​​corresponding to the cabin environment state and the equipment operating state step by step. The predicted future values ​​from multiple consecutive time steps are combined to form a predicted state sequence, and the predicted state sequence is used as the trend of the cabin state change.

4. The adaptive generation method for health cabin operating parameters based on time-series prediction according to claim 1, characterized in that, The step of generating candidate operating parameters for the health-preserving cabin based on the changing trend of the cabin's internal state includes: The overall objective function is to optimize the cabin environment to a preset comfort level and reduce equipment energy consumption. Based on the trend of changes in the cabin state, an adaptive model predictive control algorithm is used to perform rolling optimization within the preset operating physical space at the equipment end, and solve for the global control action sequence that can minimize the comprehensive objective function. Extract the immediate control action corresponding to the next adjacent control cycle from the global control action sequence, and parse the immediate control action into a multi-dimensional control vector for the ventilation intensity, temperature and humidity adjustment range, and equipment start-stop commands of the health cabin; The multidimensional control vector is used as the candidate operating parameter.

5. The adaptive generation method for health cabin operating parameters based on time-series prediction according to claim 1, characterized in that, The process involves extracting the historical data distribution of the multivariate time-series samples and, in conjunction with the trend of changes in the cabin state, dynamically adjusting the environmental anomaly detection threshold for the cabin environment state and the equipment anomaly detection threshold for the equipment operating state. The dynamic anomaly threshold for the current operating phase is obtained through the environmental anomaly detection threshold and the equipment anomaly detection threshold, including: Statistical analysis is performed on the probability density distribution of the multivariate time series samples to extract the basic anomaly threshold of the health cabin under the current operating stage. The basic anomaly threshold includes the basic environmental anomaly threshold and the basic equipment anomaly threshold. The expected slope of the future environment of the health cabin is obtained by calculating the predicted state difference between adjacent future control cycles in the trend of the cabin's state change. When the absolute value of the expected slope change is greater than the preset slope fluctuation threshold, the upper and lower limits of the basic environment anomaly threshold and the basic equipment anomaly threshold are adaptively scaled according to the gain direction of the expected slope change and the prediction error variance confidence interval of the time series state prediction model. The scaled-down basic environment anomaly threshold and the basic equipment anomaly threshold are used as the environment anomaly determination threshold and the equipment anomaly determination threshold for the current operating stage, respectively, and the dynamic anomaly threshold for the current operating stage is obtained through the environment anomaly determination threshold and the equipment anomaly determination threshold.

6. The adaptive generation method for health cabin operating parameters based on time-series prediction according to claim 1, characterized in that, The step of using the dynamic anomaly threshold to perform safety constraint verification on the candidate operating parameters, filtering out candidate operating parameters that cause the trend of the cabin state change to exceed the dynamic anomaly threshold, and obtaining the target operating parameters includes: The candidate operating parameters are superimposed onto the current prediction node of the time-series state prediction model to perform virtual forward physical deduction, and the deduction state curve after applying the candidate operating parameters is obtained. Real-time comparison of whether the peak data of the simulation state curve in any future simulation control cycle touches or exceeds the safety boundary of the dynamic anomaly threshold; If the peak data does not touch or exceed the safety boundary, then the candidate operating parameter is determined to have passed the safety constraint verification, and the candidate operating parameter is directly locked as the target operating parameter. If the peak data touches or exceeds the safety boundary, the adjustment step size of the candidate operating parameter is proportionally attenuated to obtain attenuation candidate parameters. The virtual forward physical simulation is then performed again using the attenuation candidate parameters until the simulation state curve corresponding to the attenuation candidate parameters is completely within the safety boundary. The attenuation candidate parameters that finally pass the safety constraint verification are taken as the target operating parameters.

7. The adaptive generation method for operating parameters of a health-preserving chamber based on time-series prediction according to claim 1, characterized in that, The process of obtaining the execution response result of the device based on the target operating parameters, generating an anomaly detection result when the execution response result triggers the dynamic anomaly threshold, and feeding back the prediction error to the time-series state prediction model based on the anomaly detection result for network update includes: The actual status feedback data after the device executes the target operating parameters is collected in real time by a local edge controller deployed in the health cabin, and the actual status feedback data is used as the execution response result. The numerical deviation between the actual state feedback data and the corresponding periodic cabin state change trend is calculated as the prediction error; When the actual state feedback data exceeds the dynamic anomaly threshold, an anomaly is confirmed and the anomaly detection result is generated; Using the prediction error as a supervision penalty signal, and based on the anomaly detection results, the network weight parameters of the time-series state prediction model are fine-tuned online via backpropagation in the local edge controller to complete the network update of the time-series state prediction model.

8. The adaptive generation method for operating parameters of a health-preserving chamber based on time-series prediction according to claim 1, characterized in that, The automatic switching of the health cabin to a conservative operation strategy includes performing the following graded safety handling steps when the execution response result triggers the dynamic anomaly threshold: When the execution response result triggers the environmental anomaly judgment threshold in the dynamic anomaly threshold, the health cabin is controlled to execute the first-level conservative strategy. The first-level conservative strategy includes imposing a hard limit on the maximum allowable adjustment range of the target operating parameters and dynamically increasing the data acquisition frequency for the cabin environment status. When the execution response result triggers the device anomaly determination threshold in the dynamic anomaly threshold, the health cabin is controlled to execute the second-level conservative strategy in an out-of-level manner. The second-level conservative strategy includes triggering the hardware safety interlock mechanism of the health cabin, forcibly cutting off the parameter adjustment control authority, activating the physical backup ventilation valve for forced ventilation and depressurization, and simultaneously generating an alarm reminder command.

9. The adaptive generation method for operating parameters of a health-preserving chamber based on time-series prediction according to claim 3, characterized in that, The temporal state prediction model is a composite neural network that combines a convolutional autoencoder and a long short-term memory network. The encoder includes the convolutional autoencoder. The process of inputting the multivariate time series samples into the encoder and extracting multi-source time series latent features along the time step direction includes: using the convolutional autoencoder to perform convolutional feature extraction and nonlinear dimensionality reduction on the multivariate time series samples in the spatial dimension, so as to remove high-frequency background noise in the multivariate time series samples and obtain a noise-free feature sequence after dimensionality reduction. The decoder includes the long short-term memory network, into which the noiseless feature sequence is input for recursive feature calculation in the time dimension, in order to obtain the multi-source temporal implicit features that characterize the spatiotemporal coupling relationship of multi-physics parameters in the health cabin.

10. An adaptive generation system for operating parameters of a health-preserving cabin based on time-series prediction, characterized in that, The system is applied to the adaptive generation method for operating parameters of a health cabin based on time-series prediction as described in any one of claims 1 to 9, and the system comprises: The time-series sample construction module is used to collect data on the in-cabin environment status, equipment operation status, and user status of the health cabin, and to construct multivariate time-series samples using the in-cabin environment status, equipment operation status, and user status data. The state change prediction module is used to input the multivariate time series samples into a pre-trained time series state prediction model to predict the state change trend of the health cabin in multiple future control cycles. The candidate parameter generation module is used to generate candidate operating parameters for the health cabin based on the trend of changes in the cabin's state. An anomaly threshold correction module is used to extract the historical data distribution of the multivariate time series sample and, in combination with the trend of changes in the cabin state, dynamically correct the environmental anomaly judgment threshold for the cabin environment state and the equipment anomaly judgment threshold for the equipment operating state. The dynamic anomaly threshold for the current operating stage is obtained through the environmental anomaly judgment threshold and the equipment anomaly judgment threshold. The safety constraint filtering module is used to perform safety constraint verification on the candidate operating parameters using the dynamic anomaly threshold, filter out candidate operating parameters that cause the change trend of the cabin state to exceed the dynamic anomaly threshold, obtain the target operating parameters, and send the target operating parameters to the device configured in the health cabin for execution. The feedback and safety interlock module is used to obtain the execution response result of the device based on the target operating parameters, generate an anomaly detection result when the execution response result triggers the dynamic anomaly threshold, feed back the prediction error to the time-series state prediction model based on the anomaly detection result for network update, and control the health cabin to automatically switch to a conservative operation strategy.

Images