Dynamic adjustment method for multi-mode intelligent controller of sauna room

Abstract

The invention provides a dynamic adjustment method for a multi-mode intelligent controller of a sauna room, and belongs to the technical field of sauna room controllers, and the method comprises the steps: predicting a future temperature track sequence through forward rolling of a thermal prediction control model in a submerged space, and outputting an optimal heating power instruction in combination with an entropy production minimization feedforward power reference value; the feedback adjusting quantity of the gain scheduling proportional integral differential controller is superposed to obtain a final heating power instruction to drive the electric heating module, an overheat protection strategy is triggered according to the temperature of the controller, a low-temperature compensation strategy is triggered according to the external temperature of the sauna room, and an artificial intelligence model is continuously optimized through online parameter updating. And finally, monitoring data are uploaded through the wireless communication module, and an abnormal alarm is pushed, so that the technical problem that the heating power adjustment is lagged or overshoot due to the fact that the sauna room temperature control system cannot perform prospective prediction on the future trend of the temperature rise process in a dynamic thermal environment is solved.

Description

Technical Field

[0001] This invention belongs to the technical field of sauna room controllers, and more specifically, relates to a dynamic adjustment method for a multi-mode intelligent controller for sauna rooms. Background Technology

[0002] Sauna room temperature control systems fall under the category of industrial thermal closed-loop control. Traditional implementations rely on proportional-integral-derivative (PID) controllers to provide feedback adjustment based on the deviation between the measured and target temperatures. This is combined with solid-state relays or pulse-width modulation (PWM) circuits to drive electric heating elements. These systems are widely deployed in commercial sauna equipment, home infrared saunas, and steam rooms. Existing commercial controllers generally employ fixed-parameter PID control strategies, while some high-end products introduce fuzzy control or lookup table-based gain switching to handle different operating conditions.

[0003] However, the aforementioned traditional solutions all belong to the pure feedback control paradigm, where control decisions rely solely on the temperature error at the current sampling moment, making it impossible to model or predict future heating trajectories. In sauna heating systems, the heating elements have high thermal inertia, the cavity's heat capacity varies significantly with the sealing state, and fluctuations in external ambient temperature cause boundary condition drift. These factors collectively lead to pure feedback control continuously outputting high power in the initial stage of heating due to the large temperature difference. However, as the temperature approaches the target temperature, severe overshoot occurs due to accumulated thermal inertia, and integral saturation further deteriorates the regulation performance.

[0004] In current intelligent control systems for sauna rooms, due to the complex dynamic characteristics of the thermal system and the continuous drift of system parameters as heating elements age and the cavity sealing changes, traditional proportional-integral-derivative (PID) controllers lack the ability to proactively sense the heating trend. This prevents them from incorporating predictions of future temperature trajectories into control decisions, resulting in power regulation remaining in a passive response state and failing to achieve proactive power optimization. In other words, existing technologies suffer from the technical problem that sauna room temperature control systems cannot proactively predict future heating trends under dynamic thermal environments, leading to lag or overshoot in heating power regulation. Summary of the Invention

[0005] In view of this, the present invention provides a dynamic adjustment method for a multi-mode intelligent controller for sauna rooms, which can solve the technical problem in the prior art that the sauna room temperature control system cannot predict the future trend of the heating process in a forward-looking manner under dynamic thermal environment, resulting in lag or overshoot in heating power adjustment.

[0006] This invention is implemented as follows: This invention provides a dynamic adjustment method for a multi-mode intelligent controller for sauna rooms, comprising the following steps:

[0007] The main control module collects the temperature inside the sauna, the temperature of the controller itself, and the temperature outside the sauna through an ambient temperature sensor, a controller temperature sensor, and an external temperature sensor, respectively. It continuously acquires the three temperature data at a sampling interval of 100ms, and stores the three temperature data in the external serial peripheral interface flash memory chip after compressing them through an adaptive sampling compression algorithm.

[0008] The main control module reads the sauna configuration mode selected by the user through the human-machine interaction module, and activates the corresponding relay output interface and pulse width modulation dimming interface according to the mode adaptive matching logic to complete the equipment linkage configuration.

[0009] The main control module inputs the temperature inside the sauna, the temperature of the controller itself, the temperature outside the sauna, and the target temperature threshold set by the user into the thermal prediction control model. The thermal prediction control model then predicts the future temperature trajectory sequence by rolling forward within the latent space. Based on the deviation between the future temperature trajectory sequence and the target temperature threshold, and combined with the entropy production minimization feedforward power reference value, the optimal heating power command for the current moment is output.

[0010] The main control module is based on the optimal heating power command, and superimposed with the feedback adjustment output of the proportional integral derivative controller based on the gain scheduling mechanism according to the difference between the internal temperature of the sauna room and the target temperature threshold to obtain the final heating power command. The electric heating module and sauna furnace module are driven by the pulse width modulation power adjustment circuit.

[0011] The main control module calculates the controller temperature deviation index based on the controller's own temperature and the controller's safe temperature threshold. It triggers the corresponding overheat protection strategy based on the range to which the controller temperature deviation index belongs, and triggers the low temperature compensation strategy based on the comparison result between the sauna room's external temperature and the low temperature compensation threshold. It also inputs the measured three-channel temperature data and the final heating power command into the thermal prediction control model to perform online parameter updates.

[0012] The main control module uploads the sauna room's internal temperature, the controller's own temperature, the final heating power command, and the pulse width modulation dimming interface status to the mobile application via the wireless communication module. When abnormal sensor data occurs, heating is immediately stopped and an alarm frame is pushed.

[0013] Specifically, the adaptive sampling compression algorithm uses whether the rate of change of the three temperature data exceeds a set slope threshold as the criterion. It only records the current data point when the rate of change exceeds the slope threshold, and skips storage at other times, compressing the storage amount to 3% to 8% of the original data. Differential encoding and compression algorithms are used to reduce bandwidth usage during uploading.

[0014] Specifically, the mode adaptive matching logic involves the main control module pre-storing a table of equipment control strategies corresponding to each sauna configuration mode. After the user selects a sauna configuration mode, the main control module looks up the table to obtain the corresponding relay output interface power supply logic and pulse width modulation dimming interface preset parameters, and only sends enable signals to the equipment interfaces required by the current sauna configuration mode, while keeping the other interfaces in a power-off state.

[0015] The thermal predictive control model adopts a cyclic state-space model architecture, which includes four sub-modules: a deterministic gated cyclic unit, a random latent variable encoder, a latent space forward rolling predictor, and an optimal power sequence solver. The input of the deterministic gated cyclic unit is four values: the current temperature inside the sauna, the controller's own temperature, the external temperature of the sauna, and the final heating power command. The output is the deterministic latent state vector at the current moment.

[0016] The random latent variable encoder takes the deterministic latent state vector as input and outputs the mean vector and variance vector of the random latent variables that characterize the uncertainty of the system. The two together constitute the latent space state representation. The latent space forward rolling predictor takes the latent space state representation as the initial state and predicts the temperature trajectory sequence of the next 30 steps in an autoregressive forward rolling within the latent space.

[0017] The optimal power sequence solver uses the weighted sum of squared deviations between the temperature trajectory sequence of the next 30 steps and the target temperature threshold as the objective function to search for the optimal heating power sequence that minimizes the objective function in the latent space, and takes the value of the first step of the sequence as the optimal heating power command at the current moment.

[0018] The training of the thermal predictive control model uses time-series sample pairs as input. The reconstruction loss is the square of the difference between the sauna room temperature predicted by the deterministic gated cyclic unit and the measured sauna room temperature. The regularization loss is the KL divergence between the mean vector and variance vector of the random latent variables output by the random latent variable encoder. The weighted sum of the two is the total loss function. All parameters are updated using the backpropagation algorithm.

[0019] The calculation of the feedforward power benchmark for minimizing entropy production treats the sauna heating process as an irreversible thermodynamic process, with the optimization objective of minimizing the total entropy production during the heating process. Under given target temperature thresholds and heating time constraints, a variational problem is established, and the Euler-Lagrange equation is solved to obtain the exponential asymptotic power curve of the optimal heating power over time. The calculation formula is as follows: .

[0020] The gain scheduling mechanism loads the corresponding proportional coefficient, integral coefficient, and derivative coefficient from the pre-calibrated parameter set table based on the range of the difference between the internal temperature of the sauna room and the target temperature threshold, as well as the range of the external temperature of the sauna room. In conjunction with the integral limiting strategy, the accumulated amount of the integral term is limited to between the set upper and lower bounds. When the difference between the internal temperature of the sauna room and the target temperature threshold is greater than 0, the conditional integral strategy activates the accumulation of the integral term. When the difference is less than or equal to 0, the accumulation of the integral term stops.

[0021] The formula for calculating the controller temperature deviation index is as follows: ;when When, maintain the current final heating power command unchanged; when When, the final heating power command will be reduced to 50% of the current value; when When this occurs, shutdown protection is triggered and an overheat alarm frame is pushed.

[0022] Among them, the low temperature compensation strategy refers to loading the low temperature compensation parameter group into the pre-calibrated parameter group table of the gain scheduling mechanism when the external temperature of the sauna room is lower than the low temperature compensation threshold of 5℃, and extending the heating time to compensate for heat loss.

[0023] Online parameter update refers to the continuous calculation of reconstruction loss and gradient descent based on measured time series samples during the operation phase, thereby completing the iterative update of parameters of the thermal predictive control model.

[0024] The pulse width modulation power adjustment circuit is used to convert the final heating power command output by the main control module into a pulse width modulation signal with a corresponding duty cycle, which drives a solid-state relay or an insulated gate bipolar transistor to achieve stepless adjustment of the heating power within the range of 0% to 100%.

[0025] Among them, the alarm frame is an abnormal status data packet that has been compressed and encoded, containing a fault type identifier, the current internal temperature of the sauna room, the controller's own temperature, the external temperature of the sauna room, and the final heating power command, and is sent with priority over the monitoring data frame; the wireless communication module uses the message queue telemetry transmission protocol with a service quality of level 1 for data transmission, and the security protection logic is executed locally on the main control module.

[0026] The training dataset for the thermal prediction control model is established by collecting historical heating process data under different load power (500W to 3000W), different sauna room external temperatures, and different sauna configuration modes. Each sample records the current sauna room internal temperature, controller temperature, sauna room external temperature, final heating power command, and sauna room internal temperature at the next moment, forming time-series sample pairs. After being classified according to sauna configuration mode, the samples are stored in the external serial peripheral interface flash memory chip.

[0027] This invention employs a cyclic state-space model architecture to construct a thermal predictive control model. The thermodynamic process of the sauna room is encoded as a cyclic latent space state. Within each control cycle, it predicts the temperature trajectory sequence for the next 30 steps, and searches within the latent space for the optimal heating power sequence that simultaneously minimizes temperature deviation and power variation. This allows the control decision to have a forward-looking perception of future temperature trends at every moment, thus avoiding passive lag in power regulation. The thermal predictive control model of this invention continuously adapts to system parameter drifts such as heating element aging and changes in sauna room sealing through an online parameter update mechanism, maintaining long-term prediction accuracy. This overcomes the feedback regulation inaccuracy problem caused by model mismatch in traditional proportional-integral-derivative control. Simultaneously, entropy production minimizes the feedforward power reference value, constraining the temperature rise path based on the second law of thermodynamics, causing power to decay prematurely as it approaches the target temperature, eliminating overshoot caused by thermal inertia. In summary, this invention solves the technical problem mentioned in the background art: sauna room temperature control systems cannot make forward-looking predictions of future temperature trends in dynamic thermodynamic environments, leading to lag or overshoot in heating power regulation. Attached Figure Description

[0028] Figure 1 This is a flowchart of the method of the present invention.

[0029] Figure 2 This is a comparison chart of the predicted temperature trajectory sequence for the next 30 steps in the latent space by the thermal predictive control model and the measured temperature.

[0030] Figure 3 The temperature rise curve of the final heating power command is obtained by superimposing the feedforward power reference value for minimizing entropy production with the proportional-integral-derivative feedback adjustment.

[0031] Figure 4 This is a graph showing the switching of proportional, integral, and derivative parameters and the changes in the cumulative integral term for different temperature deviation ranges under the gain scheduling mechanism. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below.

[0033] like Figure 1 The diagram shown is a flowchart of a dynamic adjustment method for a multi-mode intelligent controller for sauna rooms provided by the present invention. This method includes the following steps:

[0034] S01. The main control module collects the temperature inside the sauna, the temperature of the controller itself, and the temperature outside the sauna through the ambient temperature sensor, the controller temperature sensor, and the external temperature sensor, respectively. It continuously acquires the three temperature data at a sampling interval of 100ms, and stores the three temperature data in the external serial peripheral interface flash memory chip after compressing them through an adaptive sampling compression algorithm.

[0035] S02. The main control module reads the sauna configuration mode selected by the user through the human-machine interaction module, and activates the corresponding relay output interface and pulse width modulation dimming interface according to the mode adaptive matching logic to complete the equipment linkage configuration.

[0036] S03. The main control module inputs the internal temperature of the sauna room, the temperature of the controller itself, the external temperature of the sauna room, and the target temperature threshold set by the user into the thermal prediction control model. The thermal prediction control model forward rolls within the latent space to predict the future temperature trajectory sequence. Based on the deviation between the future temperature trajectory sequence and the target temperature threshold, and combined with the entropy production minimization feedforward power reference value, it outputs the optimal heating power command for the current moment.

[0037] S04. The main control module uses the optimal heating power command as a basis, and superimposes the feedback adjustment amount output by the proportional-integral-derivative controller based on the gain scheduling mechanism according to the difference between the internal temperature of the sauna room and the target temperature threshold to obtain the final heating power command. The electric heating module and sauna furnace module are driven by the pulse width modulation power adjustment circuit.

[0038] S05. The main control module calculates the controller temperature deviation index based on the controller's own temperature and the controller's safe temperature threshold. It triggers the corresponding overheat protection strategy based on the range to which the controller temperature deviation index belongs, and triggers the low temperature compensation strategy based on the comparison result between the sauna room's external temperature and the low temperature compensation threshold. It also inputs the measured three-channel temperature data and the final heating power command into the thermal prediction control model to perform online parameter updates.

[0039] S06. The main control module uploads the sauna room's internal temperature, the controller's own temperature, the final heating power command, and the pulse width modulation dimming interface status to the mobile application via the wireless communication module. When abnormal sensor data occurs, heating is immediately stopped and an alarm frame is pushed.

[0040] The principle of the adaptive sampling compression algorithm is as follows: the change rate of the three temperature data exceeds the set slope threshold as the judgment criterion. The current data point is recorded only when the change rate exceeds the slope threshold. The storage is skipped at other times, and the storage volume is compressed to 3% to 8% of the original data. The compressed data is stored in the flash memory chip of the external serial peripheral interface. Differential encoding and compression algorithm are used to reduce bandwidth usage when uploading.

[0041] The principle of the adaptive matching logic is as follows: The main control module pre-stores the equipment control strategy table corresponding to each sauna configuration mode. After the user selects a sauna configuration mode, the main control module looks up the table to obtain the corresponding relay output interface power supply logic and pulse width modulation dimming interface preset parameters, and only sends enable signals to the equipment interfaces required by the current sauna configuration mode, while the other interfaces remain powered off.

[0042] The specific structure of the thermal predictive control model is as follows: The thermal predictive control model adopts a cyclic state-space model architecture, comprising four sub-modules: a deterministic gated cyclic unit, a random latent variable encoder, a latent space forward rolling predictor, and an optimal power sequence solver. The deterministic gated cyclic unit receives four inputs: the current sauna room internal temperature, the controller's own temperature, the sauna room external temperature, and the final heating power command. Its output is the deterministic latent state vector at the current moment. The random latent variable encoder takes the deterministic latent state vector as input and outputs the mean vector and variance vector of the random latent variables, representing the system's uncertainty. These two vectors together constitute the latent space state representation. The latent space forward rolling predictor... The rolling predictor uses the latent space state representation as the initial state and regressively predicts the temperature trajectory sequence for the next 30 steps within the latent space. Each prediction is updated by a deterministic gated loop unit and the next latent space state representation is generated by a random latent variable encoder. After decoding, the predicted value of the sauna room's internal temperature at the corresponding time is obtained. The optimal power sequence solver uses the sum of squared deviations between the temperature trajectory sequence for the next 30 steps and the target temperature threshold, weighted by the change in the final heating power command, as the objective function. It searches for the optimal heating power sequence that minimizes the objective function within the latent space and takes the value of the first step of the sequence as the optimal heating power command for the current time and outputs it to step S03.

[0043] The steps for establishing the training dataset of the thermal prediction and control model specifically include: collecting historical heating process data under different load power (500W to 3000W), different sauna room external temperatures, and different sauna configuration modes. Each sample records the current sauna room internal temperature, controller temperature, sauna room external temperature, final heating power command, and sauna room internal temperature at the next moment, forming time-series sample pairs. After being classified according to sauna configuration mode, the samples are stored in the external serial peripheral interface flash memory chip and uploaded to the training terminal for offline training.

[0044] The specific steps for training the thermal prediction and control model include: using time-series sample pairs as input, the square of the difference between the sauna room's internal temperature predicted by the deterministic gated recurrent unit and the measured sauna room internal temperature at the next time step is used as the reconstruction loss, and the KL divergence between the mean vector and variance vector of the random latent variables output by the random latent variable encoder is used as the regularization loss. The weighted sum of the two is the total loss function. The backpropagation algorithm is used to update all parameters of the deterministic gated recurrent unit, the random latent variable encoder, the latent space forward rolling predictor, and the optimal power sequence solver. After training, the model parameters are fixed to the main control module. During the running phase, the reconstruction loss is continuously calculated using measured time-series sample pairs, and gradient descent is performed to complete the online parameter update.

[0045] The aforementioned thermal predictive control model brings the following technical advantages to the entire scheme: Traditional proportional-integral-derivative (PID) controllers rely solely on current errors for feedback adjustment, failing to predict future trends in the heating process, leading to power response lag or overshoot. The thermal predictive control model encodes the sauna's thermodynamic process as a cyclic latent space state, predicting multiple temperature trajectory sequences forward within each control cycle, and searching within the latent space for the optimal heating power sequence that simultaneously minimizes temperature deviation and power variation, thus enabling forward-looking control decisions. The online parameter update mechanism allows the model to continuously adapt to system drift such as heating element aging and changes in sauna sealing, maintaining long-term control accuracy. Working in conjunction with the PID feedback layer, it combines theoretical optimality with real-time robustness.

[0046] The calculation principle of the entropy production minimization feedforward power benchmark value is as follows: The sauna heating process is considered an irreversible thermodynamic process. The irreversibility stems from finite temperature difference heat transfer and internal dissipation of the power regulator. With minimizing the total entropy production during the heating process as the optimization objective, a variational problem is established under a given target temperature threshold and heating time constraint. The Euler-Lagrange equation is solved to obtain the theoretical trajectory of the optimal heating power changing with time. This theoretical trajectory is an exponential asymptotic power curve. After discretizing the exponential asymptotic power curve, the entropy production minimization feedforward power benchmark value at each time point is obtained. The calculation formula for the entropy production minimization feedforward power benchmark value is expressed as follows:

[0047] ;

[0048] in for The time-entropy production minimizes the feedforward power reference value (W). This is the initial rated power (W). The target temperature threshold (°C) is used. for Constant temperature inside the sauna (°C) The temperature inside the sauna room (°C) at the start of the heating process. is the thermal time constant (s). The time (s) from the start of the heating to the current time is given; both sides of the equation are dimensionless, with the left side being the power ratio and the right side being the product of the temperature difference ratio and the exponential decay factor. The equation holds true after the dimensions cancel each other out.

[0049] The technical effects of minimizing the entropy production feedforward power reference value on the entire scheme are as follows: Traditional proportional-integral-derivative (PID) control continuously outputs high power in the early stage of heating due to the large temperature difference, resulting in a large amount of irreversible heat loss. When approaching the target temperature threshold, it causes overshoot due to inertia. The entropy production minimization algorithm starts from the second law of thermodynamics, constraining the heating path to the theoretically optimal trajectory with the minimum total entropy production. In the early stage of heating, the power gradually changes to adapt to the heat dissipation loss. When approaching the target temperature threshold, the power is reduced in advance, fundamentally eliminating overshoot and reducing irreversible heat loss. This makes the entire heating process approach the optimal energy efficiency in a thermodynamic sense. The PID feedback layer is responsible for correcting the actual deviation. The two layers work together to make the control both theoretically optimal and real-time robust.

[0050] The gain scheduling mechanism operates as follows: Based on the range of the difference between the sauna room's internal temperature and the target temperature threshold, and the range of the sauna room's external temperature, corresponding proportional coefficients, integral coefficients, and derivative coefficients are loaded from the pre-calibrated parameter set table to prevent overshoot or response lag of a single fixed parameter under different operating conditions. Simultaneously, an integral limiting strategy is employed to restrict the accumulated integral term between a set upper and lower bound to prevent integral saturation. When the difference between the sauna room's internal temperature and the target temperature threshold is greater than 0, the conditional integral strategy activates the accumulation of the integral term. When the difference between the sauna room's internal temperature and the target temperature threshold is less than or equal to 0, the integral term stops accumulating to prevent reverse accumulation of the integral term during the overshoot phase.

[0051] The formula for calculating the controller temperature deviation index is as follows:

[0052] ;

[0053] in The controller temperature deviation index (dimensionless). The temperature is the controller's own temperature (°C). The controller's safe temperature threshold (°C); when When, maintain the current final heating power command unchanged; when When, the final heating power command will be reduced to 50% of the current value; when When the temperature outside the sauna room is 5°C below the low temperature compensation threshold, the low temperature compensation parameter group is loaded into the pre-calibrated parameter group table of the gain scheduling mechanism to extend the heating time to compensate for heat loss.

[0054] The pulse width modulation power adjustment circuit is used to convert the final heating power command output by the main control module into a pulse width modulation signal with a corresponding duty cycle, drive a solid-state relay or an insulated gate bipolar transistor, and realize stepless adjustment of the heating power within the range of 0% to 100%.

[0055] The alarm frame is a compressed and encoded abnormal status data packet containing a fault type identifier, the current internal temperature of the sauna, the controller's own temperature, the external temperature of the sauna, and the final heating power command. It is sent before the monitoring data frame to ensure the real-time transmission of abnormal information. The wireless communication module uses the message queue telemetry transmission protocol with a service quality of 1 level for data transmission. The security protection logic is executed locally on the main control module. The wireless communication module only transmits monitoring and configuration data to ensure that signal attenuation in high humidity environments does not affect the timeliness of protection response.

[0056] The KL divergence is a measure of the difference between the probability distribution represented by the mean vector and variance vector of the random latent variables output by the random latent variable encoder and the standard normal distribution. It is used to constrain the output distribution of the random latent variable encoder, prevent the latent space state representation from being overly dispersed, and improve the generalization ability of the thermal predictive control model.

[0057] The thermal time constant is a time-domain parameter characterizing the temperature response rate of the sauna room's thermodynamic system during the heating process. It is determined by the product of the sauna room's heat capacity and equivalent thermal resistance. It is obtained through offline identification during the training phase of the thermal prediction and control model and stored in the main control module for use in calculating the feedforward power reference value for entropy production minimization.

[0058] The specific implementation of step S01 is as follows: The main control module polls and collects the internal temperature of the sauna, the controller's own temperature, and the external temperature of the sauna through three hardware sensor interfaces at a fixed sampling period of 100ms, forming a time-aligned three-channel digital temperature sequence. To address the constraints of storage capacity and write lifespan of the external serial peripheral interface flash memory chip, the main control module executes an adaptive sampling compression algorithm in parallel on the three original sequences. The core criterion is the comparison between the absolute value of the first-order difference between adjacent sampling points, i.e., the rate of temperature change, and a preset slope threshold (reference value is 0.05℃ / 100ms). Only when the rate of change exceeds the slope threshold is the current data point written to the external serial peripheral interface flash memory chip; otherwise, the write operation is skipped, thereby compressing the actual storage volume to 3% to 8% of the original data. During the data upload stage, the compressed sparse sequence is further processed by differential encoding, recording only the incremental values ​​between adjacent storage points, and then packaged by the compression algorithm to reduce the wireless communication bandwidth usage.

[0059] The specific implementation of step S02 is as follows: The main control module continuously polls the input register of the human-machine interaction module to obtain the sauna configuration mode number currently selected by the user. Then, using this number as an index, it performs a table lookup operation in the device control strategy table pre-stored within the main control module to read the power supply logic (high-level enable or low-level enable) of each relay output interface and the preset duty cycle parameters of each pulse width modulation dimming interface under the corresponding sauna configuration mode. The main control module then only sends enable signals to the interfaces required by the current sauna configuration mode, keeping the interfaces marked as inactive in the strategy table powered off to prevent unexpected device operation and complete the device linkage configuration under multiple modes.

[0060] The specific implementation of step S03 is as follows: The main control module sends four data points—the current sampling time's sauna room internal temperature, the controller's own temperature, the sauna room external temperature, and the target temperature threshold set by the user through the human-machine interaction module—to the deterministic gated loop unit of the thermal predictive control model. After the gated loop unit updates the deterministic latent state vector, the random latent variable encoder outputs the random latent variable mean vector and variance vector, which together constitute the latent space state representation at the current moment. The latent space forward rolling predictor uses this as the initial state and autoregressively predicts the sauna room internal temperature trajectory sequence for the next 30 steps (corresponding to 3000ms) within the latent space. Simultaneously, the main control module calculates the entropy-minimizing feedforward power reference value based on the current sauna room internal temperature, the target temperature threshold, the sauna room internal temperature at the start of heating, and the thermal time constant (reference value is 300s). This reference value follows an exponential asymptotic power curve, maintaining a high power in the early stages of heating to overcome heat loss and prematurely decaying the power to suppress overshoot when approaching the target temperature threshold. The optimal power sequence solver uses the sum of squared deviations between the temperature trajectory sequence of the next 30 steps and the target temperature threshold as the weighted change in the final heating power command as the objective function. It searches for the optimal heating power sequence that minimizes the objective function in the latent space, takes the first step value of the sequence and the entropy production minimized feedforward power benchmark value as a weighted fusion, and outputs the optimal heating power command at the current moment.

[0061] The specific implementation of step S04 is as follows: The main control module uses the optimal heating power command output in step S03 as the feedforward reference, and superimposes the feedback adjustment amount output by the proportional-integral-derivative controller of the gain scheduling mechanism to obtain the final heating power command. The gain scheduling mechanism loads the corresponding proportional coefficient, integral coefficient, and derivative coefficient from the pre-calibrated parameter group table based on the difference between the internal temperature of the sauna room and the target temperature threshold (reference zones: difference greater than 20℃, 5~20℃, 0~5℃) and the external temperature of the sauna room (reference zones: below 5℃, 5~25℃, above 25℃), to avoid overshoot or response hysteresis of a single fixed parameter in a wide operating range. The integral limiting strategy constrains the accumulated amount of the integral term between the set upper limit (reference value is 30% of the rated power) and the lower limit (reference value is -10% of the rated power). The conditional integral strategy allows the accumulation of the integral term when the internal temperature of the sauna room is lower than the target temperature threshold; otherwise, it freezes the integral term to prevent reverse accumulation during the overshoot phase. The final heating power command is converted into a pulse width modulation signal with a corresponding duty cycle by the pulse width modulation power adjustment circuit, which drives the solid-state relay or insulated gate bipolar transistor to realize stepless adjustment of the heating power of the electric heating module and sauna module.

[0062] The specific implementation of step S05 is as follows: the main control module calculates the controller temperature deviation index in each control cycle. The controller's safe temperature threshold reference value is 70℃. When... No protection action is triggered at times; when When the final heating power command is forcibly reduced to 50% of the current value and maintained, The system triggers shutdown protection, cutting off all heating interface enable signals and pushing overheat alarm frames via the wireless communication module. The low-temperature compensation strategy uses the difference between the current external sauna temperature and the low-temperature compensation threshold (reference value 5℃) as the criterion. When the external sauna temperature is 5℃ lower than the low-temperature compensation threshold, the low-temperature compensation parameter set is loaded from the pre-calibrated parameter set table, extending the heating time to compensate for heat loss caused by increased environmental heat dissipation. Online parameter updates use the measured three-channel temperature data from the current control cycle and the final heating power command to form a time-series sample pair. The square of the difference between the predicted next-time sauna internal temperature and the measured value by the deterministic gated loop unit is calculated as the reconstruction loss. A one-step gradient descent is then performed to update all parameters of the thermal predictive control model, enabling the model to continuously adapt to system dynamic drift.

[0063] The specific implementation of step S06 is as follows: In each upload cycle, the main control module encapsulates the sauna room internal temperature, the controller's own temperature, the final heating power command, and the pulse width modulation dimming interface status into a monitoring data frame, and uploads it to the mobile application through the wireless communication module with message queue telemetry transmission protocol service quality level 1. Service quality level 1 ensures that each data frame is transmitted at least once and confirmed by the receiving end, and can still maintain data integrity when the signal attenuates in a high humidity environment. Sensor data anomaly detection is based on three temperature data points exceeding the physically reasonable range (reference range: sauna internal temperature 0~150℃, controller own temperature 0~85℃, sauna external temperature -40~60℃) or adjacent sampling point abrupt changes exceeding the abrupt change threshold (reference value is 5℃ / 100ms). Once an anomaly is triggered, the main control module immediately cuts off all heating interface enable signals and constructs an alarm frame. The alarm frame contains a fault type identifier, the current three temperature data points, and the final heating power command. After compression and encoding, it is sent before the monitoring data frame. The safety protection logic is executed locally on the main control module and decoupled from the wireless communication link status to ensure that signal attenuation in high humidity environments does not affect the protection response time.

[0064] It should be noted that the first key technical idea of ​​this invention is thermal predictive control based on a cyclic state-space model. Traditional proportional-integral-derivative controllers can only sense the temperature error at the current moment, and the control decision is essentially a post-event compensation for the deviation that has already occurred. However, the thermal predictive control model encodes the sauna room thermodynamics into a latent state vector through a deterministic gated cyclic unit, and predicts the temperature trajectory of the future multiple steps in an autoregressive rolling manner within the latent space. This ensures that the power command at each moment contains a forward-looking judgment of the future trend, fundamentally changing the control paradigm from passive feedback to active predictive optimization.

[0065] The second key technical approach is to minimize the entropy production feedforward power reference value. Traditional control strategies do not impose any constraints on the thermodynamic optimality of the heating path, resulting in a large amount of irreversible heat loss in the early stages of heating. The entropy production minimization method starts from the second law of thermodynamics and uses the exponential asymptotic power curve obtained by solving the Euler-Lagrange variational equation as the feedforward reference, so that the power change trajectory approaches the optimality in a thermodynamic sense, thus avoiding overshoot caused by thermal inertia in advance.

[0066] The third key technical approach is the synergy between gain scheduling, proportional-integral-derivative (PID) feedback, and online parameter updates. The gain scheduling mechanism dynamically switches PID parameters based on multi-dimensional operating condition partitions, combining conditional integration and integral limiting strategies to suppress integral saturation, thus accurately correcting residual deviations in the predictive control layer at the real-time level. Online parameter updates enable the thermodynamic predictive control model to continuously iterate as the system ages and the environment changes, maintaining prediction accuracy. These three mechanisms complement each other, giving the overall control system thermodynamic optimality, predictive foresight, and real-time robustness—qualities that no single mechanism can independently achieve.

[0067] It should be noted that this invention also solves the following technical problems: In existing technologies, sauna room controllers typically write all temperature data collected into the storage medium at a fixed frequency. Long-term operation with a 100ms sampling interval generates a massive amount of raw data, severely depleting the storage capacity and write lifespan of the flash memory chip for external serial peripheral interfaces. Simultaneously, it consumes a large amount of bandwidth resources during the wireless upload phase. This invention implements sparse storage using an adaptive sampling compression algorithm with a change rate threshold as the gating condition. Data points are recorded only when there is a meaningful temperature change, compressing the storage volume to 3% to 8% of the original data. Furthermore, differential coding and compression algorithms are superimposed during the upload phase, significantly reducing both storage pressure and transmission bandwidth consumption. This solves the technical problems of redundant temperature data storage and wasted transmission bandwidth in high-frequency sampling scenarios.

[0068] Furthermore, this invention solves the technical problem that the response time of safety protection in sauna room control systems under harsh high-humidity environments depends on the reliability of the wireless communication link. In existing technologies, some intelligent controllers rely on instructions from the cloud or application side to trigger core safety logic such as overheat protection and over-temperature power-off. When high humidity causes wireless signal attenuation or communication interruption, the protection response may be delayed or even fail. This invention executes all safety protection logic (including controller temperature deviation index exceeding limits, sensor data anomaly detection, and shutdown protection triggering) locally on the main control module. The wireless communication module only undertakes the responsibility of transmitting monitoring data and alarm frames, making the safety protection response completely independent of the communication link status, thus eliminating the risk of protection failure due to channel attenuation.

[0069] Specifically, the principle of this invention is as follows: The fundamental reason why this invention can solve the aforementioned core technical problems is that the thermal predictive control model encodes the internal dynamics of the sauna room's thermodynamic system into a latent space state representation, and predicts multiple temperature trajectories in an autoregressive manner within this compressed state space. This allows the controller to grasp the future temperature evolution trend before issuing power commands, rather than relying solely on the instantaneous feedback of the current error. A random latent variable encoder introduces probabilistic modeling of system uncertainties, enabling the latent space state representation to simultaneously contain both deterministic dynamics and random disturbances. Based on this, the optimal power sequence solver solves a multi-objective optimization problem that balances temperature tracking accuracy and power stability, outputting the optimal heating power command for the current moment. The entropy production minimization feedforward power reference value pre-plans the optimal heating power trajectory based on the thermodynamic variational principle, forming a feedforward-feedback dual-layer collaborative structure with the predictive control layer. The gain-scheduled proportional-integral-derivative controller provides real-time error correction on this basis. These three mechanisms jointly ensure that the control decision possesses theoretical optimality and real-time robustness at every moment. The online parameter update mechanism enables the model parameters to continuously iterate with the measured data, ensuring that the prediction accuracy of the thermal predictive control model does not degrade due to system parameter drift. This mechanism guarantees the effectiveness of the forward-looking prediction capability in long-term operation.

[0070] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0071] The specific implementation of step S01 is as follows: the main control module continuously collects the internal temperature of the sauna room at a sampling interval of 100ms through the ambient temperature sensor, the controller temperature sensor, and the external temperature sensor. (°C), controller's own temperature (°C) and the external temperature of the sauna (°C) Three temperature data streams. The adaptive sampling compression algorithm determines whether the rate of change of each temperature data stream exceeds a slope threshold. The judgment is made based on (℃ / s), for the first... The road temperature signal is at the first The rate of change at each sampling time is calculated using the following formula:

[0072] ;

[0073] In the formula, For the first The road temperature signal is at the first Rate of change at each sampling time (°C / s); For the first Road temperature sensor in the first Temperature values ​​(°C) at each sampling time; For the first Road temperature sensor in the first Temperature values ​​(°C) at each sampling time; The sampling interval (s) is 0.1s. The time normalization reference value (s) is set to 1s by default, so that the denominator... Dimensionless The dimensionless unit is ℃ / s, and Dimensions consistent; when The system records the current data point in real time; otherwise, it skips storage and compresses the storage volume to 3% to 8% of the original data. The compressed data is stored in the flash memory chip of the external serial peripheral interface. Differential encoding and compression algorithms are used to reduce bandwidth usage during uploading.

[0074] The specific implementation of step S02 is as follows: The main control module reads the sauna configuration mode selected by the user through the human-machine interaction module, queries the pre-stored device control strategy table according to the mode adaptive matching logic, obtains the corresponding relay output interface power supply logic and pulse width modulation dimming interface preset parameters, sends enable signals only to the device interfaces required for the current mode, and keeps the other interfaces in a power-off state to complete the device linkage configuration.

[0075] The specific implementation of step S03 is as follows: the main control module will... , , and user-defined target temperature threshold (°C) Input the thermal predictive control model, which then forward-rolls within the latent space to predict the temperature trajectory sequence for the next 30 steps, and then, based on the temperature trajectory sequence and... The deviation is considered, and combined with the minimum entropy production feedforward power reference value, the optimal heating power command for the current moment is output. (W). The feedforward power benchmark for minimizing entropy production constrains the heating path to the theoretically optimal trajectory that minimizes total entropy production. In the initial stage of heating, the power gradually changes to match the heat dissipation loss, approaching... The power is attenuated in advance to fundamentally eliminate overshoot. The calculation formula is as follows:

[0076] ;

[0077] In the formula, for The time-entropy production minimizes the feedforward power reference value (W); Initial rated power (W); The target temperature threshold (°C); for The temperature inside the sauna room at any given time (°C) is The time-varying form; The temperature inside the sauna room (°C) at the moment the temperature rises; The thermal time constant (s) is determined by the heat capacity of the sauna room. (J / ℃) and equivalent thermal resistance The product of (℃ / W) is determined, that is The dimensionless quantity is s; The time (in seconds) from the start of the heating process to the current time; the left side of the formula... The power ratio is shown on the right, and the temperature difference ratio is shown on the right. With exponential decay factor The product of the product has dimensionless dimensions on both sides. The thermal prediction and control model was identified through an offline system during the training phase. The specific steps are as follows: Step 1, apply a step heating power to the sauna room under controlled conditions. Record at 100ms intervals Step 2: Perform exponential fitting on the heating curve using the least squares fitting method to extract the curve over time. The estimated value is stored in the main control module for subsequent calculations.

[0078] The specific implementation of step S04 is as follows: the main control module commands the optimal heating power. As a feedforward reference, the feedback adjustment output of the gain-dispatch proportional-integral-derivative controller is superimposed. (W), to obtain the final heating power command (W), the formula is expressed as follows:

[0079] ;

[0080] In the formula, The final heating power command (W); The feedback adjustment (W) is the output of the gain-dispatch proportional-integral-derivative controller. The initial rated power (W) has the same dimensions for all three terms; after division, both sides of the formula are dimensionless. Temperature deviation. The calculation formula is expressed as follows:

[0081] ;

[0082] In the formula, for The difference (°C) between the internal temperature of the sauna and the target temperature threshold at any given time. The normalized temperature reference value (°C) is assumed to be 100°C; both sides of the formula are dimensionless temperature deviation ratios. Feedback adjustment amount. The calculation formula is expressed as follows:

[0083] ;

[0084] In the formula, , , These are the proportional coefficient, integral coefficient, and differential coefficient, respectively, all obtained from the pre-calibration parameter set table, and are dimensionless. For integration dummy variable (s); proportional term Dimensionless; integral term middle, The dimension is ℃·s, divided by (°C) and The product of (s) is dimensionless; the differential term middle, The dimension is ℃ / s, divided by (℃ / s) is dimensionless; all three terms are dimensionless, as shown on the left. Consistent. Gain scheduling mechanism based on The interval and The corresponding interval is loaded from the pre-calibration parameter group table. , , This prevents overshoot or hysteresis from occurring under different operating conditions with a single fixed parameter. The integral limiting strategy restricts the cumulative amount of the integral term to a set upper limit. and the lower world To prevent integral saturation, the conditional integration strategy stipulates that integration only occurs when... When the integral term accumulates, The integral term stops accumulating during the time step to prevent reverse accumulation of the integral term during the overshoot phase. Ultimately... The pulse width modulation (PWM) power adjustment circuit converts the signal into a PWM signal with a corresponding duty cycle, which drives a solid-state relay or an insulated gate bipolar transistor to achieve stepless adjustment of the heating power within the range of 0% to 100%.

[0085] The specific implementation of step S05 is as follows: the main control module determines the temperature based on the controller's own temperature. With the controller's safe temperature threshold (°C) Calculate the controller temperature deviation index The formula is expressed as follows:

[0086] ;

[0087] In the formula, The temperature deviation index of the controller is dimensionless. The controller's own temperature (°C); The controller's safe temperature threshold (°C); both the numerator and denominator are in the dimension of °C, and after division... It is dimensionless. When At that time, maintain the current state. Unchanged; when At that time, Reduce to 50% of the current value; when When this occurs, shutdown protection is triggered, and an overheat alarm frame is pushed through the wireless communication module. When the temperature drops below the low-temperature compensation threshold by 5°C, a low-temperature compensation parameter set is loaded into the pre-calibrated parameter set table of the gain scheduling mechanism to extend the heating time and compensate for heat loss. The main control module then compares the measured temperature data from the three channels with... Input the thermal predictive control model, continuously calculate the reconstruction loss and perform gradient descent to complete online parameter updates, so that the model can adapt to system drift such as heating element aging and changes in sauna room sealing.

[0088] The specific implementation of step S06 is as follows: the main control module will... , , The status of the pulse width modulation dimming interface is uploaded to the mobile application via the wireless communication module using a message queue telemetry transmission protocol with a service quality of 1 level. When abnormal sensor data occurs, heating is immediately stopped and an alarm frame is pushed. The alarm frame includes a fault type identifier, , , and Prioritizing the transmission of monitoring data frames, the security protection logic is executed locally on the main control module, ensuring that signal attenuation in high humidity environments does not affect the timeliness of protection response.

[0089] To better understand and implement this invention, the following is a specific application scenario of the invention, Example 2: To verify the effectiveness of the invention, technicians set up a test environment, using a heating machine with a rated heating power of 2000W and a cavity volume of 2... A home-use wooden sauna room was used as the test object. The main control module uses a 32-bit microcontroller, and the external serial peripheral interface flash memory chip has a capacity of 16... The wireless communication module supports Message Queuing Telemetry Transmission Protocol (MQTP) service quality level 1 transmission. The test scenario was set with an external sauna temperature of 8℃ (below the low-temperature compensation threshold of 5℃, triggering the low-temperature compensation strategy). The user selected the dry sauna mode via the human-computer interaction module, with a target temperature threshold set to 85℃. The internal sauna temperature at the start of heating was 12℃, and the offline thermal time constant identification result was 320. The controller's safe temperature threshold is set to 70℃, and the low temperature compensation threshold is set to 10℃.

[0090] During the execution of step S01, the main control module continuously collects three channels of temperature data at a sampling interval of 100ms. In the initial stage of heating (0~120℃)... During the initial stage, the temperature inside the sauna rose rapidly at a rate of approximately 0.18℃ / 100ms, exceeding the slope threshold of 0.05℃ / 100ms. An adaptive sampling compression algorithm continuously recorded data points. In the later stages of the heating process (600℃...),... Afterwards, the temperature change rate dropped to approximately 0.02℃ / 100ms, below the slope threshold, and a large number of sampling points were skipped from storage. Statistically, the entire heating process (approximately 1200...) A total of 12,000 raw sampling points were generated, of which 476 data points were actually written to the external serial peripheral interface flash memory chip, resulting in a storage compression rate of approximately 3.97%. After differential coding and compression algorithms during the upload phase, the data packet size was reduced to approximately 4.2% of the original data, effectively reducing the wireless communication bandwidth usage. Technicians recorded the sampling and storage status of the three temperature channels, as shown in Table 1.

[0091] Table 1. Statistics of Three-Way Temperature Adaptive Sampling Compression

[0092]

[0093] During step S02, the main control module reads the dry steam mode number from the input register of the human-machine interface module and looks up the corresponding equipment control strategy: relay output interface 1 (electric heating module) and relay output interface 2 (sauna oven module) are enabled; pulse width modulation dimming interface 1 (ambient light) has a preset duty cycle of 60%; and relay output interface 3 (steam generator) and pulse width modulation dimming interface 2 (ultraviolet lamp) remain de-energized. The main control module only sends valid signals to the above-mentioned enabled interfaces, while the other interfaces remain de-energized, thus completing the equipment linkage configuration for the dry steam mode.

[0094] During step S03, the deterministic gated loop unit of the thermal predictive control model receives four numerical values ​​in each control cycle: the sauna's internal temperature, the controller's own temperature, the sauna's external temperature, and the final heating power command, updating the deterministic latent state vector (64 dimensions). The random latent variable encoder outputs the random latent variable mean vector and variance vector (each with 16 dimensions), forming the latent space state representation. The latent space forward rolling predictor uses this as the initial state to predict the temperature trajectory for the next 30 steps, such as... Figure 2 As shown, the average absolute error between the predicted trajectory and the measured temperature trajectory is approximately 0.8℃. Meanwhile, with the sauna's internal temperature at the start of the heating process at 12℃, the target temperature threshold at 85℃, and the thermal time constant at 320... Calculate the feedforward power reference value to minimize entropy production at the 120th temperature rise. hour, The corresponding feedforward power reference value is approximately 1106. The optimal power sequence solver searches the latent space and obtains the first-step value of the optimal heating power sequence, which is then weighted and fused with the feedforward power reference value (weight ratio 7:3). The output optimal heating power command for the current moment is approximately 1180. .

[0095] During step S04, if the external temperature of the sauna room is 8℃, which is lower than the low-temperature compensation threshold of 10℃, the gain scheduling mechanism loads the low-temperature compensation parameter set. When the difference between the internal temperature of the sauna room and the target temperature threshold is 55℃ (greater than the 20℃ setting), the corresponding proportional coefficient, integral coefficient, and derivative coefficient are loaded (reference values ​​are 0.8, 0.02, and 0.02, respectively). 0.5 The feedback adjustment is superimposed on the optimal heating power command to obtain a final heating power command of approximately 1240. The signal is converted into a pulse width modulation signal with a 62% duty cycle by the pulse width modulation power regulation circuit to drive the electric heating module and the sauna module. The changes in the final heating power command at different stages are shown in Table 2:

[0096] Table 2. Record of Final Heating Power Commands at Typical Moments During the Heating Process

[0097]

[0098] During step S05, the controller's own temperature reached a maximum of 58°C during the heating process, and the controller temperature deviation index... The overheat protection was not triggered, and the final heating power command maintained normal output. Figure 3 The figure shows the temperature rise curve of the final heating power command obtained by superimposing the entropy production minimization feedforward power reference value and the proportional-integral-derivative feedback adjustment. Since the external temperature of the sauna room is 8℃, which is lower than the low-temperature compensation threshold of 10℃ (the difference is 2℃, not exceeding the 5℃ trigger condition; the low-temperature compensation parameter group has been activated in step S04 according to the actual configuration), the online parameter update is based on the 120th... The reconstruction loss is calculated using time-series samples collected at each step, and a gradient descent step is performed. The parameters of the thermal predictive control model are continuously iterated with the measured data, and the mean absolute error of the prediction is within 1200. The temperature was reduced to approximately 0.5℃ at the end of the heating process, verifying the effectiveness of the online parameter update mechanism in maintaining model accuracy. Figure 4 This is a graph showing the switching of proportional, integral, and derivative parameters and the changes in the cumulative integral term for different temperature deviation ranges under the gain scheduling mechanism.

[0099] During step S06, the wireless communication module uploads a frame of monitoring data every 500ms using the Message Queuing Telemetry Transmission Protocol (MQTP) with Quality of Service (QoS) Level 1. This data includes the current sauna room temperature, the controller's own temperature, the final heating power command, and the pulse width modulation (PWM) dimming interface status. Throughout the test, the sensor data remained within a reasonable range, and no abnormal alarms were triggered. After the test, technicians simulated an over-temperature injection into the controller (…). In abnormal scenarios, the main control module immediately triggers shutdown protection and constructs an alarm frame locally. The alarm frame is sent before the monitoring data frame. The local response delay from the triggering of the abnormality to the issuance of the alarm frame is about 15ms. The delay of the wireless communication link does not affect the triggering timing of the protection action, which verifies the effectiveness of the design of decoupling the security protection logic from the communication link.

[0100] Compared to traditional pure proportional-integral-derivative (PID) control methods, the advancements of this invention are reflected in the following aspects: Traditional control decisions only include error information at the current moment and have no ability to perceive the inertial characteristics of the thermodynamic system. In contrast, this invention compresses the thermodynamic dynamics into a latent space state through a thermodynamic predictive control model. The control decisions include explicit predictions of future multi-step temperature trajectories, transforming power regulation from ex-post compensation to ex-ante planning. The entropy production minimization feedforward mechanism incorporates the constraints of the second law of thermodynamics into the control objective, making the entire heating power trajectory approach optimal in a thermodynamic sense, rather than simply pursuing the minimization of the current error. The online parameter update mechanism enables the model to have adaptive capabilities as the system state drifts, overcoming the performance degradation problem that occurs in traditional fixed parameter control after heating element aging or changes in cavity sealing.

[0101] It should be noted that the variables involved in this invention are explained in detail in Table 3.

[0102] Table 3. Variable Explanation Table

[0103]

[0104] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A dynamic adjustment method for a multi-mode intelligent controller of a sauna room, characterized in that, Includes the following steps: The main control module collects the temperature inside the sauna, the temperature of the controller itself, and the temperature outside the sauna through an ambient temperature sensor, a controller temperature sensor, and an external temperature sensor, respectively. It continuously acquires the three temperature data at a sampling interval of 100ms, and stores the three temperature data in the external serial peripheral interface flash memory chip after compressing them through an adaptive sampling compression algorithm. The main control module reads the sauna configuration mode selected by the user through the human-machine interaction module, and activates the corresponding relay output interface and pulse width modulation dimming interface according to the mode adaptive matching logic to complete the equipment linkage configuration. The main control module inputs the temperature inside the sauna, the temperature of the controller itself, the temperature outside the sauna, and the target temperature threshold set by the user into the thermal prediction control model. The thermal prediction control model then predicts the future temperature trajectory sequence by rolling forward within the latent space. Based on the deviation between the future temperature trajectory sequence and the target temperature threshold, and combined with the entropy production minimization feedforward power reference value, the optimal heating power command for the current moment is output. The main control module is based on the optimal heating power command, and superimposed with the feedback adjustment output of the proportional integral derivative controller based on the gain scheduling mechanism according to the difference between the internal temperature of the sauna room and the target temperature threshold to obtain the final heating power command. The electric heating module and sauna furnace module are driven by the pulse width modulation power adjustment circuit. The main control module calculates the controller temperature deviation index based on the controller's own temperature and the controller's safe temperature threshold. It triggers the corresponding overheat protection strategy based on the range to which the controller temperature deviation index belongs, and triggers the low temperature compensation strategy based on the comparison result between the sauna room's external temperature and the low temperature compensation threshold. It also inputs the measured three-channel temperature data and the final heating power command into the thermal prediction control model to perform online parameter updates. The main control module uploads the sauna room's internal temperature, the controller's own temperature, the final heating power command, and the pulse width modulation dimming interface status to the mobile application via the wireless communication module. When abnormal sensor data occurs, heating is immediately stopped and an alarm frame is pushed.

2. The dynamic adjustment method of the multi-mode intelligent controller for sauna rooms according to claim 1, characterized in that, The adaptive sampling compression algorithm specifically uses whether the rate of change of the three temperature data exceeds a set slope threshold as the judgment criterion. It only records the current data point when the rate of change exceeds the slope threshold, and skips storage at other times, compressing the storage amount to 3% to 8% of the original data. Differential encoding and compression algorithms are used to reduce bandwidth usage during uploading.

3. The dynamic adjustment method of the multi-mode intelligent controller for sauna rooms according to claim 2, characterized in that, The adaptive matching logic of the mode is specifically that the main control module pre-stores the equipment control strategy table corresponding to each sauna configuration mode. After the user selects a sauna configuration mode, the main control module looks up the table to obtain the corresponding relay output interface power supply logic and pulse width modulation dimming interface preset parameters, and only sends enable signals to the equipment interfaces required by the current sauna configuration mode, while the other interfaces remain in a power-off state.

4. The dynamic adjustment method of the multi-mode intelligent controller for sauna rooms according to claim 3, characterized in that, The thermal predictive control model adopts a cyclic state-space model architecture, which includes four sub-modules: a deterministic gated cyclic unit, a random latent variable encoder, a latent space forward rolling predictor, and an optimal power sequence solver. The input of the deterministic gated cyclic unit is four values: the current temperature inside the sauna, the controller's own temperature, the external temperature of the sauna, and the final heating power command. The output is the deterministic latent state vector at the current moment.

5. The dynamic adjustment method of the multi-mode intelligent controller for sauna rooms according to claim 4, characterized in that, The random latent variable encoder takes the deterministic latent state vector as input and outputs the mean vector and variance vector of the random latent variables that characterize the uncertainty of the system. The two together constitute the latent space state representation. The latent space forward rolling predictor takes the latent space state representation as the initial state and predicts the temperature trajectory sequence of the next 30 steps in an autoregressive forward rolling within the latent space.

6. The dynamic adjustment method of the multi-mode intelligent controller for sauna rooms according to claim 5, characterized in that, The optimal power sequence solver uses the weighted sum of squared deviations between the temperature trajectory sequence of the next 30 steps and the target temperature threshold as the objective function to search for the optimal heating power sequence that minimizes the objective function in the latent space, and takes the value of the first step of the sequence as the optimal heating power command at the current moment.

7. The dynamic adjustment method of the multi-mode intelligent controller for sauna rooms according to claim 6, characterized in that, The training of the thermal predictive control model uses time-series sample pairs as input. The reconstruction loss is the square of the difference between the sauna room temperature predicted by the deterministic gated cyclic unit and the measured sauna room temperature. The regularization loss is the KL divergence between the mean vector and variance vector of the random latent variables output by the random latent variable encoder. The weighted sum of the two is the total loss function. All parameters are updated using the backpropagation algorithm.

8. The dynamic adjustment method of the multi-mode intelligent controller for sauna rooms according to claim 7, characterized in that, The calculation of the feedforward power benchmark value for minimizing entropy production treats the heating process of the sauna as an irreversible thermodynamic process. With the minimization of the total entropy production of the heating process as the optimization objective, a variational problem is established under the given target temperature threshold and heating time constraints. The Euler-Lagrange equation is solved to obtain the exponential asymptotic power curve of the optimal heating power changing with time.

9. The dynamic adjustment method of the multi-mode intelligent controller for sauna rooms according to claim 8, characterized in that, The gain scheduling mechanism loads the corresponding proportional coefficient, integral coefficient, and derivative coefficient from the pre-calibrated parameter set table based on the range of the difference between the temperature inside the sauna and the target temperature threshold, as well as the range of the temperature outside the sauna. The integral term accumulation is limited to between the set upper and lower bounds by the integral limit strategy; when the difference between the temperature inside the sauna and the target temperature threshold is greater than 0, the conditional integral strategy activates the integral term accumulation. When the difference is less than or equal to 0, the integral term stops accumulating.

10. The dynamic adjustment method of the multi-mode intelligent controller for sauna rooms according to claim 9, characterized in that, The low-temperature compensation strategy refers to loading a low-temperature compensation parameter set into the pre-calibrated parameter set table of the gain scheduling mechanism when the external temperature of the sauna room is 5°C lower than the low-temperature compensation threshold, thereby extending the heating time to compensate for heat loss.

Images