Water dispenser temperature control method based on AI adaptive algorithm

Abstract

The invention discloses a water dispenser temperature control method based on an AI adaptive algorithm, and relates to the technical field of intelligent household electrical appliance control, and the method comprises the steps: obtaining an environment monitoring data flow and an equipment operation state parameter flow in real time; time sequence alignment and feature fusion are carried out on the data to generate a user water consumption behavior feature vector and an equipment working condition feature vector; inputting the feature vector into a pre-trained adaptive temperature control decision model, and outputting a target heating temperature set value, a preheating start advance and a step power adjustment strategy; calculating a real-time power compensation amount according to the target temperature and the thermal inertia parameters of the equipment; and dynamically correcting the stepped power strategy based on the compensation amount, generating a power control instruction sequence and sending the power control instruction sequence to a heating controller for execution. According to the method, intelligent prediction and self-adaptive decision making of the water demand are achieved by fusing multi-source data, dynamic power compensation is conducted in combination with equipment thermal inertia, and therefore the instant hot water demand of a user is accurately met, and meanwhile energy consumption is effectively reduced.

Description

Technical Field

[0001] This invention belongs to the field of intelligent home appliance control technology, specifically a water dispenser temperature control method based on AI adaptive algorithms. Background Technology

[0002] Currently, most water dispenser temperature control solutions on the market employ either on / off control based on preset temperatures or simple timed heating strategies. These methods primarily rely on temperature sensors within the heating chamber; heating begins when the water temperature is detected to be below a set threshold and stops once the threshold is reached. Some improved solutions introduce a timer function, maintaining heating for a preset period to accommodate more regular water usage patterns. These technologies constitute the mainstream practice in current water dispenser temperature control.

[0003] Existing technical solutions have shortcomings. Their control logic is relatively isolated and static, focusing only on the device's instantaneous water temperature while completely ignoring the dynamic impact of external environmental changes and users' actual water usage patterns. Current control methods cannot perceive these factors, leading to frequent energy waste, such as ineffective heat preservation during unoccupied periods or in low-temperature environments, or failure to prepare sufficient hot water before peak water usage times, impacting user experience. Furthermore, conventional methods do not adequately consider the device's thermal inertia characteristics, and their fixed power output mode makes it difficult to achieve rapid and accurate temperature control response during reheating after frequent water extraction, easily causing temperature overshoot or heating delays.

[0004] There is a need for an intelligent temperature control method that can comprehensively sense environmental and user behavior and adaptively and dynamically adjust control strategies. This method needs to address how to effectively extract features from multi-source heterogeneous time-series data and predict demand, and how to finely adjust power output based on the real-time thermal dynamic characteristics of the equipment, thereby optimizing overall energy consumption while ensuring the demand for instant hot water. Summary of the Invention

[0005] This invention aims to solve at least one of the technical problems existing in the prior art; Therefore, this invention proposes a water dispenser temperature control method based on an AI adaptive algorithm, comprising: The system acquires real-time environmental monitoring data streams and equipment operating status parameter streams output from the water dispenser's sensor network. The environmental monitoring data streams include ambient temperature, ambient humidity, and the intensity of human movement signals in the surrounding area. The equipment operating status parameter streams include real-time heating chamber temperature, inlet water flow rate, historical heating power data, and historical water usage data for the current time period. The environmental monitoring data stream and the equipment operating status parameter stream are subjected to time-series alignment and feature fusion processing to generate user water use behavior feature vectors and equipment operating condition feature vectors. The user water usage behavior feature vector and the equipment operating condition feature vector are input into the pre-trained adaptive temperature control decision model for calculation, and the target heating temperature setpoint, preheating start advance amount and stepped power adjustment strategy are output. The real-time power compensation is calculated based on the target heating temperature setpoint and the thermal inertia parameters in the equipment operating condition feature vector. The stepped power adjustment strategy is dynamically modified based on the real-time power compensation amount, a power control command sequence is generated, and the power control command sequence is sent to the water dispenser's heating controller for execution.

[0006] Further, the step of performing time-series alignment and feature fusion processing on the environmental monitoring data stream and the equipment operating status parameter stream to generate user water usage behavior feature vectors and equipment operating condition feature vectors includes: The user water usage behavior feature vector includes predicted water usage time, predicted water temperature preference, and estimated water consumption; the equipment operating condition feature vector includes thermal inertia parameters, heat loss rate, and heating efficiency attenuation coefficient. A unified timeline based on the water dispenser system clock is established, and each data point in the environmental monitoring data stream and each data point in the equipment operating status parameter stream are mapped onto the unified timeline. Based on the fluctuation pattern of the surrounding human movement signal intensity, the time sequence of the user's approach to the water dispenser is identified; The sequence of behavioral time points is matched and associated with the actual water collection time in the current water usage history, and the ambient temperature, real-time temperature change curve of the heating chamber, and water consumption before and after each water usage event are extracted. Using time series analysis algorithms, data from multiple consecutive water usage events are analyzed to extract the user's preference patterns for water temperature under different ambient temperatures and at different times, as well as the average interval between two adjacent water usages, and to generate the predicted water usage time, predicted water temperature preference, and estimated water consumption. Based on the historical heating power record and the real-time temperature change curve of the heating cavity, the thermal inertia parameters, heat loss rate, and heating efficiency attenuation coefficient are calculated by fitting a thermodynamic model.

[0007] Furthermore, the time series analysis algorithm is used to analyze data from multiple consecutive water usage events to extract user preferences for water temperature under different ambient temperatures and at different times, including: Construct a hidden Markov model that takes ambient temperature and time of day as inputs and the actual water temperature finally selected by the user as output. The hidden Markov model is trained using the ambient temperature, the real-time temperature change curve of the heating chamber, and the water usage time data from historical water usage events. Input the current ambient temperature and time point into the trained Hidden Markov Model, and the model outputs the corresponding water temperature state probability distribution. The water temperature value with the highest probability in the water temperature probability distribution is selected as the core value for predicting the water temperature preference, and the variance of the probability distribution is calculated as the confidence level of the predicted water temperature preference.

[0008] Further, the step of inputting the user water usage behavior feature vector and the equipment operating condition feature vector into a pre-trained adaptive temperature control decision model for calculation, and outputting the target heating temperature setpoint, preheating start-up advance, and stepped power adjustment strategy, includes: The stepped power adjustment strategy defines the heating power curves used in different heating stages; The predicted water usage time and predicted water temperature preference are input into the demand prediction sub-network of the adaptive temperature control decision model, and the output of the demand prediction sub-network is the optimal water storage temperature range that satisfies the predicted water temperature preference. The thermal inertia parameters and heat loss rate are input into the device dynamic sub-network of the decision model. The device dynamic sub-network calculates the theoretical shortest time and the corresponding theoretical average power required to heat water from the current temperature to the lower limit of the optimal water storage temperature range. The theoretical shortest time is compared with the predicted water usage time, and the time required to start heating in advance is calculated to ensure that users can obtain water with the preferred water temperature within the predicted water usage time, i.e., the preheating start advance amount. Combining the heating efficiency decay coefficient and safety constraints, the theoretical average power is optimized in segments to form a stepped power adjustment strategy consisting of high power rapid heating in the initial stage, smooth transition in the middle stage, and low power heat preservation in the final stage. Set the center value of the optimal water storage temperature range as the target heating temperature setting value.

[0009] Further, the step of calculating the real-time power compensation amount based on the target heating temperature setpoint and the thermal inertia parameter in the equipment operating condition feature vector includes: At preset time intervals, the difference between the real-time temperature of the heating chamber and the target heating temperature setting is continuously monitored and recorded as the real-time temperature deviation. The real-time temperature deviation and the thermal inertia parameter are input together into the proportional-integral-derivative controller; The proportional-integral-derivative controller adjusts its control parameters according to the thermal inertia parameter. The larger the thermal inertia parameter, the slower the system temperature response. Therefore, the weight of the integral term is increased to reduce the steady-state error. The proportional-integral-derivative controller calculates based on the adjusted control parameters and the real-time temperature deviation, and outputs the real-time power compensation amount used to dynamically eliminate the real-time temperature deviation.

[0010] Furthermore, the step of dynamically correcting the stepped power adjustment strategy based on the real-time power compensation amount to generate a power control command sequence includes: Obtain the preset reference power value corresponding to the current heating stage in the stepped power adjustment strategy; The real-time power compensation amount is superimposed with the preset reference power value to obtain the actual execution power at the current moment; Determine whether the actual power output exceeds the safe operating upper or lower limit of the water dispenser's heating element; If the power exceeds the upper limit, the actual operating power will be limited to the safe operating upper limit value; if it is below the lower limit, the actual operating power will be limited to the safe operating lower limit value or the heating will be turned off. The actual power values ​​calculated and verified at each moment are arranged in chronological order to form the power control command sequence of the power adjustment unit of the direct-drive heating controller.

[0011] Furthermore, the method also includes an online learning and updating step for the adaptive temperature control decision model: After executing the power control command sequence and completing one heating-water extraction cycle, the actual execution data of the heating-water extraction cycle is collected; the actual execution data includes the complete real-time temperature change curve of the heating chamber, the actual total power consumption, and the user's final actual water temperature. The actual execution data is compared with the predicted water temperature preference and water consumption estimate in the user water use behavior feature vector to generate model prediction error feedback data. The model prediction error feedback data, together with the corresponding environmental monitoring data stream, equipment operating status parameter stream, and decision parameters output by the previous model, constitute a training sample. The training samples are used to incrementally train the pre-trained adaptive temperature control decision model to fine-tune the model parameters, so that the model's subsequent predictions and decisions can better adapt to the actual working conditions of a specific water dispenser and the actual water usage habits of a specific user.

[0012] Further, the incremental training of the pre-trained adaptive temperature control decision model using the training samples includes: The environmental monitoring data stream and equipment operating status parameter stream in the training samples are used as inputs, and the model prediction error feedback data is used as a supervision signal. The backpropagation algorithm is used to calculate the gradient of the loss function between the current output of the adaptive temperature control decision model and the ideal output based on feedback from actual execution data; The connection weights of each layer of the neural network inside the model are updated along the direction of gradient descent of the loss function with a preset incremental learning rate. After completing incremental learning of a predetermined number of training samples, the performance of the updated model is evaluated using a cross-validation set. If the performance improvement is less than a threshold or a decline occurs, the model is reverted to the version before the update.

[0013] Furthermore, the pre-trained adaptive temperature control decision model is obtained through the following steps: Construct a neural network model whose input layer corresponds to the user water use behavior feature vector and the equipment operating condition feature vector, and whose output layer corresponds to the target heating temperature setpoint, preheating start advance, and stepped power adjustment strategy parameters. Multiple sets of training samples are obtained. Each set of training samples includes: historical environmental monitoring data stream and equipment operating status parameter stream, historical feature vectors extracted from them, and optimal decision parameters that have been optimized and verified under the historical scenarios. The neural network model is trained under supervision using the multiple sets of training samples, with the loss function being the loss function being the difference between the decision parameters output by the model and the optimal decision parameters. When the model's prediction accuracy on the validation set reaches a preset threshold, training is complete, and the pre-trained adaptive temperature control decision model is obtained.

[0014] Furthermore, the generation of the power control command sequence also takes into account energy efficiency optimization objectives: Establish a function with the total heating energy consumption as the optimization objective, wherein the variables of the function are the power value and duration of each stage in the stepped power adjustment strategy; Under the constraints of heating water to the target heating temperature setpoint and ensuring the preheating start advance, a set of theoretically optimal power allocation schemes are obtained by solving the extreme values ​​of the function. When generating the actual execution power, the theoretically optimal power allocation scheme is used as a reference, so that while dynamically compensating for temperature deviations, the overall energy consumption approaches the theoretically optimal value.

[0015] Compared with the prior art, the beneficial effects of the present invention are: By acquiring and integrating environmental monitoring data and equipment operating status parameters in real time, feature vectors representing user habits and equipment conditions are generated and input into a pre-trained adaptive decision-making model for calculation. This enables proactive prediction and precise adaptation to user water demand. Ambient temperature, humidity, and human movement signals provide external evidence for judging water activity and heat demand, while water usage history and power records depict internal patterns. This allows the system to move beyond simply responding to water temperature thresholds, instead predicting peak water usage in advance and intelligently deciding on target temperature and preheating timing. This ensures that hot water at the appropriate temperature is provided precisely when the user needs it, reducing unnecessary waiting time and redundant heating energy consumption during low-demand periods, thus improving convenience and energy efficiency.

[0016] By calculating the power compensation amount based on the target temperature and real-time thermal inertia parameters, and dynamically correcting the stepped power strategy output by the model, real-time matching between the control strategy and the physical characteristics of the equipment is achieved. The thermal inertia parameters reflect the current thermal dynamic response characteristics of the heating system, and the calculation of the compensation amount ensures that the power output can be adjusted promptly and appropriately when the equipment operating conditions change. This closed-loop correction mechanism enables the heating process to approach the set temperature more smoothly and quickly, effectively avoiding the problems of large temperature fluctuations, slow response, or energy waste caused by traditional on / off control or fixed power curves, thus improving the stability and dynamic performance of temperature control. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of the water dispenser temperature control method based on AI adaptive algorithm described in this invention. Figure 2 This is a flowchart for extracting water temperature preference patterns based on a hidden Markov model; Figure 3 The curve showing the coordinated change between the real-time temperature and temperature deviation of the heating chamber; Figure 4 The dynamic correction curve for PID power compensation and temperature deviation; Figure 5 The response curve is the linkage between PID power compensation and energy efficiency-temperature control dynamic weighting coefficient. Detailed Implementation

[0018] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] See Figure 1The system acquires real-time environmental monitoring data streams and equipment operating status parameter streams from the water dispenser's sensor network. The environmental monitoring data streams include ambient temperature, humidity, and the intensity of surrounding human movement signals. The equipment operating status parameter streams include real-time heating chamber temperature, inlet water flow rate, historical heating power data, and historical water usage data for the current period. These data streams are then time-aligned and feature-fused to generate user water usage behavior feature vectors and equipment operating condition feature vectors. The user water usage behavior feature vector includes predicted water usage time, predicted water temperature preference, and estimated water consumption. The equipment operating condition feature vector includes thermal inertia parameters, heat loss rate, and heating efficiency decay coefficient. These vectors are then input into a pre-trained adaptive temperature control decision model for calculation, outputting the target heating temperature setpoint, preheating start advance, and a stepped power adjustment strategy. The stepped power adjustment strategy defines the heating power curves used at different heating stages. Based on the target heating temperature setpoint and the thermal inertia parameters in the equipment operating condition feature vectors, real-time power compensation is calculated. The stepped power adjustment strategy is dynamically corrected based on the real-time power compensation, a power control command sequence is generated, and the power control command sequence is sent to the water dispenser's heating controller for execution.

[0020] In one embodiment of the present invention, see [reference] Figure 2 A unified timeline based on the water dispenser system clock was established, mapping each data point in the environmental monitoring data stream and each data point in the equipment operating status parameter stream onto the unified timeline. This time alignment operation ensures that ambient temperature, ambient humidity, surrounding human movement signal strength, real-time heating chamber temperature, water flow rate, heating power historical records, and current period water usage historical records have consistent reference in the time dimension. Based on the fluctuation pattern of surrounding human movement signal strength, the time sequence of user approaching the water dispenser is identified. The time sequence of the behavior time point is represented by a continuous period of time during which the signal strength continuously rises and exceeds a preset threshold. The behavior time sequence is matched and correlated with the actual water filling time in the current period water usage historical records to extract the ambient temperature, real-time heating chamber temperature change curves, and water usage data within a specific time window before and after each water usage event.

[0021] In some embodiments, time series analysis algorithms are used to analyze data from multiple consecutive water usage events to extract user preferences for water temperature under different ambient temperatures and at different times, as well as the average interval between two consecutive water usages. The analysis process involves pattern mining of ambient temperature, real-time temperature change curves of the heating chamber, and water usage time points from historical water usage events, thereby generating a user water usage behavior feature vector containing predicted water usage time, predicted water temperature preferences, and estimated water consumption. Based on historical heating power records and real-time temperature change curves of the heating chamber, thermodynamic models are used to fit and calculate the thermal inertia parameters, heat loss rate, and heating efficiency decay coefficient included in the equipment operating condition feature vector.

[0022] In practice, a Hidden Markov Model (HMM) is constructed, taking ambient temperature and time of day as inputs and the user's final selected water temperature as output. The HMM contains a set of hidden states, each corresponding to a different user water temperature preference pattern. The HMM is trained using historical water usage data, including ambient temperature, real-time temperature change curves of the heating chamber, and water usage time points. The training process involves using the Baum-Welch algorithm to estimate the initial state distribution, state transition probability matrix, and observation probability matrix parameters of the HMM. The current ambient temperature and time point are then input into the trained HMM, which outputs the corresponding water temperature state probability distribution.

[0023] Optionally, the probability distribution of water temperature state can be calculated as follows: Where: symbol Indicates at time step The hidden state at that time is the water temperature preference mode, symbol Indicates the time step from the start time. All observation sequences, i.e., ambient temperature and time point sequences, symbol The posterior probability distribution of the current hidden state given the observation sequence is represented by the symbol. Represents the probability of observation, symbol Represents the state transition probability, symbol The posterior probability of the hidden state at the previous time step is represented by the symbol. It is a normalized constant. The water temperature value with the highest probability in the water temperature state probability distribution is selected as the core value for predicting water temperature preference, and the variance of the water temperature state probability distribution is calculated as the confidence level for predicting water temperature preference.

[0024] In practical implementation, the predicted water usage time in the user water usage behavior feature vector is obtained by applying a moving average or exponential smoothing to the interval sequence between two consecutive historical water usages. The estimated water consumption is obtained by integrating the inflow rate data of historical water usage events over the duration of water usage. The thermal inertia parameter in the equipment operating condition feature vector is characterized by a time constant obtained by fitting a first-order system model to the response relationship between historical heating power records and the real-time temperature change curve of the heating chamber during the heating phase. The heat loss rate is calculated by analyzing the natural cooling curve of the heating chamber's real-time temperature during the static phase without heating input. The heating efficiency decay coefficient is obtained through trend regression analysis of long-term historical heating power records and effective temperature rise data, reflecting the change in heating element performance over time.

[0025] In one embodiment of the invention, the predicted water usage time and predicted water temperature preference are input into the demand prediction sub-network of the adaptive temperature control decision model. The output of the demand prediction sub-network is the optimal water storage temperature range that satisfies the predicted water temperature preference. Thermal inertia parameters and heat loss rate are input into the equipment dynamics sub-network of the decision model. The equipment dynamics sub-network calculates the theoretical shortest time and corresponding theoretical average power required to heat water from the current temperature to the lower limit of the optimal water storage temperature range. The theoretical shortest time is compared with the predicted water usage time to calculate the preheating start time required to ensure that users receive water at their preferred temperature during the predicted water usage time. Combining the heating efficiency decay coefficient and safety constraints, the theoretical average power is optimized in segments, forming a stepped power adjustment strategy consisting of high-power rapid heating in the initial stage, smooth transition in the middle stage, and low-power heat preservation in the final stage. The center value of the optimal water storage temperature range is set as the target heating temperature setpoint.

[0026] In practical implementation, the process of inputting user water usage behavior feature vectors and equipment operating condition feature vectors into a pre-trained adaptive temperature control decision model for calculation is detailed below. The predicted water usage time and predicted water temperature preference from the user water usage behavior feature vectors are input into the demand prediction sub-network of the adaptive temperature control decision model. This demand prediction sub-network is a trained feedforward neural network whose output is the optimal water storage temperature range to satisfy the predicted water temperature preference. The thermal inertia parameters and heat loss rate from the equipment operating condition feature vectors are input into the equipment dynamics sub-network of the adaptive temperature control decision model. This equipment dynamics sub-network is a simulator based on a physical model or another neural network module. Based on the input thermal inertia parameters, heat loss rate, and the initial water temperature and target lower limit of the heating chamber, the equipment dynamics sub-network calculates the theoretical shortest time and corresponding theoretical average power required to heat the water from the current temperature to the lower limit of the optimal water storage temperature range.

[0027] In some embodiments, the process of calculating the theoretical shortest time in the device dynamic subnetwork involves solving a first-order differential equation describing the dynamic changes in the heating chamber water temperature. The solution for the theoretical shortest time and theoretical average power can be expressed as an optimization problem satisfying constraints. The calculated theoretical shortest time is compared with the predicted water usage time in the user's water usage behavior feature vector. Specifically, the predicted water usage time is subtracted from the current system time to obtain a time difference, which is then compared with the theoretical shortest time. If the theoretical shortest time is less than or equal to the time difference, the preheating start-up advance is zero; if the theoretical shortest time is greater than the time difference, the preheating start-up advance is equal to the absolute value of the theoretical shortest time minus the time difference.

[0028] Understandably, after calculating the preheating start-up advance, it is necessary to perform piecewise optimization design of the theoretical average power by combining the heating efficiency attenuation coefficient in the equipment operating condition characteristic vector with preset safety constraints. The goal of piecewise optimization design is to balance heating rate, component safety, and energy efficiency while ensuring that water is heated to the target temperature and available at the predicted water usage time. Safety constraints include the allowable instantaneous maximum power of the heating element, the continuous operating power limit, and the maximum rate of change limit of adjacent power adjustment step sizes. The optimization process forms a stepped power adjustment strategy consisting of high-power rapid heating in the initial stage, smooth transition in the middle stage, and low-power heat preservation in the final stage. The stepped power adjustment strategy is defined in the form of a set of time-power setpoint sequences.

[0029] In one embodiment of the present invention, the difference between the real-time temperature of the heating cavity and the target heating temperature setpoint is continuously monitored at preset time intervals and recorded as the real-time temperature deviation. The real-time temperature deviation and the thermal inertia parameter are input into a proportional-integral-derivative (PID) controller. The PID controller adjusts its control parameters according to the thermal inertia parameter. The larger the thermal inertia parameter, the slower the system temperature response, so the weight of the integral term is increased to reduce steady-state error. The PID controller calculates and outputs a real-time power compensation amount to dynamically eliminate the real-time temperature deviation based on the adjusted control parameters and the real-time temperature deviation. The preset reference power value corresponding to the current heating stage in the stepped power adjustment strategy is obtained, and the real-time power compensation amount is superimposed with the preset reference power value to obtain the actual execution power at the current moment. It is determined whether the actual execution power exceeds the safe operating upper or lower limit of the water dispenser heating element. If it exceeds the upper limit, the actual execution power is limited to the safe operating upper limit value; if it is lower than the lower limit, the actual execution power is limited to the safe operating lower limit value or the heating is turned off. The actual execution power values ​​calculated and verified at each moment are arranged in chronological order to form a power control command sequence that directly drives the power adjustment unit of the heating controller.

[0030] In practical implementation, the process of calculating the real-time power compensation based on the target heating temperature setpoint and the thermal inertia parameter in the equipment operating condition feature vector operates in a closed-loop control manner. The difference between the real-time temperature of the heating chamber and the target heating temperature setpoint is continuously monitored at preset fixed time intervals; this difference is recorded as the real-time temperature deviation. The real-time temperature deviation and the thermal inertia parameter in the equipment operating condition feature vector are input together into a proportional-integral-derivative (PID) controller. The PID controller dynamically adjusts its internal proportional, integral, and derivative coefficients based on the input thermal inertia parameter values. Specifically, a larger thermal inertia parameter indicates a slower response of the heating chamber water temperature system to heating power. The adjustment strategy of the PID controller is to increase the weighting coefficient of the integral term. This adjustment aims to strengthen the cumulative correction effect of historical deviations for systems with slow response, thereby reducing steady-state errors.

[0031] In some embodiments, the proportional-integral-derivative (PID) controller calculates a real-time power compensation amount based on adjusted control parameters and real-time temperature deviation, outputting this compensation to dynamically eliminate the real-time temperature deviation. The calculation process for the real-time power compensation amount is embodied in the discretized recursive formula of the PID controller, which incorporates the gain parameter adjusted according to the thermal inertia parameter. The output of the PID controller can be expressed as: Where: symbol Indicates the first The real-time power compensation amount calculated at each sampling time, with the symbol... The basic gain coefficient of the proportional term, symbol [symbol missing]. Indicates the first Real-time temperature deviation at each sampling time, sign Represents thermal inertia parameters The relevant integral term gain function, the value of which varies with Increase and increase, symbol Indicates the preset sampling time interval, symbol Represents from the initial moment to the present moment. The sum of all real-time temperature deviations at any given time, sign Represents thermal inertia parameters The relevant differential gain function, sign Indicates the first The real-time temperature deviation at each sampling moment is calculated. After obtaining the real-time power compensation amount, the stepped power adjustment strategy is dynamically modified based on the real-time power compensation amount to generate a power control command sequence. This process first obtains the preset reference power value corresponding to the current heating stage in the stepped power adjustment strategy.

[0032] It can be understood that the calculated real-time power compensation is algebraically superimposed with the preset baseline power value corresponding to the current moment. The result of the algebraic superposition is the actual execution power planned for the current moment. Immediately after the superposition process is completed, a safety check is performed on the actual execution power to determine whether the value of the actual execution power exceeds the instantaneous safe operating upper limit or falls below the safe operating lower limit defined by the technical specifications of the water dispenser heating element. If the actual execution power exceeds the safe operating upper limit, the value of the actual execution power is limited to the safe operating upper limit value; if the actual execution power is lower than the safe operating lower limit, the value of the actual execution power is limited to the safe operating lower limit value or a command to turn off heating is directly generated. The actual execution power value calculated and safety checked at each sampling moment is arranged and cached in strict chronological order to form a series of digital or analog power control command sequences that directly drive the internal power adjustment unit of the heating controller. The transmission frequency of the power control command sequence is consistent with the sampling time interval.

[0033] See Figure 3 This figure demonstrates the complete dynamic characteristics of real-time temperature monitoring and temperature deviation changes in the heating cavity. The horizontal axis represents sampling time (minutes), the left vertical axis represents the real-time temperature of the heating cavity (°C), and the right vertical axis represents the real-time temperature deviation (°C). The solid red line represents the real-time temperature curve of the heating cavity, which gradually rises from an initial temperature of approximately 25°C and stabilizes around 40°C, reflecting the synergistic effect of the stepped power adjustment strategy and the real-time power compensation mechanism. The dashed red line represents the target heating temperature setpoint, which is maintained at approximately 95°C. The solid blue line represents the real-time temperature deviation curve, which starts at approximately 70°C and gradually decreases to approximately 55°C as the heating process progresses, demonstrating the continuous correction effect of the temperature deviation after the PID controller is dynamically adjusted in conjunction with thermal inertia parameters. The dashed blue line represents the temperature deviation baseline (0°C).

[0034] In one embodiment of the present invention, after executing the power control command sequence and completing a complete heating-water extraction cycle, the system collects the actual execution data of this heating-water extraction cycle. The actual execution data includes the complete real-time temperature change curve of the heating chamber monitored throughout the entire process from preheating start to the end of water extraction by the user, the actual total power consumption recorded by the power metering module, and the actual water temperature obtained by the temperature sensor when the user finally extracts water. The collected actual execution data is compared with the predicted water temperature preference and water consumption estimate in the user water use behavior feature vector on which the power control command sequence was generated. The comparison operation specifically includes calculating the absolute difference between the actual water temperature and the core value of the predicted water temperature preference, calculating the ratio of the actual total power consumption to the theoretical power consumption estimated based on the predicted water consumption estimate and the heating strategy, and generating model prediction error feedback data including indicators such as temperature deviation and energy efficiency deviation.

[0035] In some embodiments, the model prediction error feedback data, along with the original environmental monitoring data stream that triggered the heating decision, the equipment operating status parameter stream, and the decision parameters previously output by the adaptive temperature control decision model, constitute a new training sample. This training sample's structured representation includes an input portion and a expected output portion. The input portion consists of the historical environmental monitoring data stream and the equipment operating status parameter stream, while the expected output portion consists of the decision parameters validated and corrected by actual execution data. The newly generated training sample is used to incrementally train the pre-trained adaptive temperature control decision model. Incremental training aims to fine-tune the model's internal parameters, enabling the adaptive temperature control decision model's subsequent predictions and decisions to better adapt to the actual operating conditions of a specific water dispenser and the actual water usage habits of a specific user. Specifically, the incremental training process involves using the environmental monitoring data stream and the equipment operating status parameter stream from the training sample as input to the neural network, and transforming the model prediction error feedback data to serve as a supervisory signal to guide the adjustment of the model weights. The transformation method can be to treat the actual execution data as a better decision result, thereby calculating the difference between the actual data and the original model output as the training objective. Refer to Table 1 for the data organization format of the training sample.

[0036] Table 1: Incremental Training Sample Data Structure Table Sample field group Field Name Data type / description Input data Environmental monitoring data stream A time-series array containing ambient temperature, humidity, and human movement signals. Equipment operating status parameter stream The timing array includes heating chamber temperature, inlet water flow rate, historical power, etc. raw output Model prior decision parameters The structure includes the originally predicted target temperature, preheating advance, etc. Supervision signals Actual execution data The structure includes actual temperature profiles, actual power consumption, and actual water temperature. Calculated error feedback The structure includes quantitative error indicators such as temperature deviation and energy consumption deviation. In practice, the backpropagation algorithm is used to calculate the gradient of the loss function between the current output of the adaptive temperature control decision model and the ideal output based on actual execution data feedback. The loss function typically uses mean squared error or cross-entropy to measure the difference between the model's original decision parameters and the ideal parameters corrected based on actual feedback. The connection weights of each layer of the neural network are updated along the direction of the loss function gradient descent with a preset incremental learning rate. The incremental learning rate is a positive decimal smaller than the initial training learning rate, designed to fine-tune the parameters rather than drastically change them. An example of the mathematical expression for weight updates is as follows: Where: symbol This indicates the neural network after incremental update. Layer weight parameters, symbol This indicates the neural network before the update. Layer weight parameters, symbol This represents the preset incremental learning rate, denoted by [symbol]. Represents the loss function For the The partial derivatives of the layer weight parameters are called gradients. After incremental learning with a predetermined number of training samples, the performance of the updated model is evaluated using an independently collected and saved cross-validation set, which includes historical data that did not participate in this incremental training. If the evaluation results show that the performance improvement of the updated model is less than a preset threshold or that the performance has decreased, the weight parameters updated in this incremental step are discarded, and the model is reverted to the version before the update.

[0037] It is understandable that the pre-trained adaptive temperature control decision model is obtained through supervised training. A deep neural network model is constructed, with its input layer dimension corresponding to the total dimension of the user water usage behavior feature vector and the equipment operating condition feature vector after feature fusion, and its output layer dimension corresponding to the dimensions of the target heating temperature setpoint, preheating start-up advance, and stepped power adjustment strategy parameters. Multiple sets of training samples are obtained for initial training. Each set of training samples includes: historical environmental monitoring data streams and equipment operating status parameter streams, historical feature vectors extracted from the historical data streams, and optimal decision parameters determined by optimization algorithms or expert rules in historical scenarios. The loss function is to minimize the difference between the model's output decision parameters and the optimal decision parameters. The neural network model is trained under supervised conditions using multiple sets of training samples, employing stochastic gradient descent or its variants as the training algorithm. When the model's prediction accuracy on an independent validation set reaches a preset threshold, training is complete, and the weights are solidified, resulting in the pre-trained adaptive temperature control decision model.

[0038] See Figure 4 In the real-time control optimization stage of the water dispenser based on AI adaptive algorithms, the dynamic response relationship between PID power compensation and temperature deviation can be systematically analyzed using this curve. Specifically, the curve is plotted with time on the horizontal axis, temperature deviation (the difference between the actual temperature and the target temperature) on the left vertical axis, and PID power compensation on the right vertical axis. In the initial under-temperature stage (light blue area), the temperature deviation converges rapidly from -70℃, and the PID power compensation is maintained at a high power output of 200W to achieve rapid heating. When the temperature deviation approaches 0 and enters the over-temperature stage (light red area), the PID power compensation immediately switches to a negative compensation state of -100W to suppress heating power and avoid temperature overshoot. In the heat preservation compensation stage (light green area), the temperature deviation fluctuates around 0℃, and the PID power compensation maintains temperature stability through dynamic fluctuations, adjusting the compensation amount in real time between -100W and 50W to offset the effects of system thermal inertia and environmental heat loss. Entering the subsequent under-temperature stage (light blue area on the right), the temperature deviation increases negatively again, and the PID power compensation increases synchronously and returns to a high power output state to drive the temperature to converge back to the target value. The curve clearly shows the dynamic correction process of temperature deviation by PID power compensation. Through power compensation strategies at different stages, a balance between heating efficiency and temperature stability is achieved.

[0039] In one embodiment of the present invention, the generation process of the power control command sequence also considers the energy efficiency optimization objective, establishing a function with the total electrical energy consumption required to complete one heating task as the optimization objective. This function is called the total heating energy consumption function, and the variables of the total heating energy consumption function are the power values ​​of each heating stage defined in the stepped power adjustment strategy and the duration of that stage. The specific form of the total heating energy consumption function depends on the electrical characteristics of the water dispenser's heating element, and is usually expressed as the sum of the products of the power values ​​and durations of each stage. Under the constraints of heating the water from the initial temperature to the target heating temperature setpoint and ensuring the preheating start advance requirement, the extreme values ​​of the total heating energy consumption function are solved by numerical optimization methods to obtain a set of power allocation schemes with the lowest energy consumption in theory. This scheme defines the optimal power value and optimal duration for each stage.

[0040] In some embodiments, the process of solving the theoretically optimal power allocation scheme can be formalized as a constrained nonlinear programming problem. The constraints include the thermodynamic equations that the heating process must satisfy, the upper time limit specified by the preheating start-up advance, and the allowable power range of the heating element. This nonlinear programming problem can be solved using optimization algorithms such as gradient descent, interior-point methods, or sequential quadratic programming. After obtaining the theoretically optimal power allocation scheme, this scheme will serve as a reference benchmark to guide the final generation of the real-time power control command sequence. In the process of dynamically correcting the stepped power adjustment strategy based on real-time power compensation to generate the actual executed power, the generation of the actual executed power not only relies on feedback compensation of real-time temperature deviation but also needs to consider its closeness to the theoretically optimal power allocation scheme.

[0041] It is understandable that generating the actual power output is a multi-objective decision-making process. Objective one is to quickly eliminate real-time temperature deviations to maintain water temperature stability; objective two is to bring the overall energy consumption close to the theoretical optimal value to achieve energy efficiency optimization. These two objectives are coordinated through a weighted adjustment mechanism. When generating the actual power output value at the current moment, in addition to adding the real-time power compensation from the proportional-integral-derivative controller, the reference power value corresponding to the theoretically optimal power allocation scheme at the current moment is also calculated. The final power decision formula can be expressed as: Where: symbol Indicates at time The final determined actual power output, after safety verification, will be sent to the heating controller. (Symbol) Indicates at time The preset reference power value obtained from the original stepped power adjustment strategy, symbol Indicates at time The real-time power compensation calculated by the proportional-integral-derivative controller, symbol Indicates at time The reference power value given in the theoretically optimal power allocation scheme, with the sign... It is a dynamically adjusted weighting coefficient, with a value between 0 and 1. Weighting coefficient The logic for its value is that when the real-time temperature deviation is large, Approaching 1, emphasizing the speed of temperature tracking; when the real-time temperature deviation is very small, Approaching 0, emphasizing the tracking of the theoretically optimal energy efficiency curve.

[0042] Optionally, the calculation of the theoretically optimal power allocation scheme can be performed offline and stored as a lookup table, or it can be solved quickly online, depending on the computing power of the controller. At each step of generating the power control command sequence, the system executes the aforementioned weighted decision process, ensuring that while dynamically compensating for temperature deviations, the power-time curve of the overall heating process is guided towards the theoretically optimal power allocation scheme, thereby bringing the overall energy consumption close to the theoretically optimal value.

[0043] See Figure 5 In the dynamic decision-making stage of water dispenser temperature control based on AI adaptive algorithms, the linkage curve between PID compensation and energy efficiency-temperature control weight coefficient intuitively presents the dynamic process of multi-objective power decision-making. Specifically, the horizontal axis represents heating time (minutes), the left vertical axis represents the real-time PID power compensation (W), and the right vertical axis represents the dynamic weight coefficient λ (0~1). The red curve represents the real-time PID power compensation, reflecting the power adjustment range generated to eliminate real-time temperature deviation; the purple curve represents the dynamic weight coefficient λ, used to coordinate the two objectives of "temperature control accuracy" and "energy efficiency optimization". The yellow-marked area in the figure is the "large temperature deviation → λ≈1 (prioritize temperature control)" stage. At this time, the dynamic weight coefficient λ approaches 1, the system prioritizes the speed of temperature tracking, the real-time PID power compensation fluctuates greatly, and the temperature deviation is quickly reduced through high-frequency power adjustment. The light green area in the figure represents the "small temperature deviation → λ≈0 (prioritizing energy efficiency)" stage. At this time, the dynamic weight coefficient λ approaches 0, the system prioritizes tracking the theoretically optimal energy efficiency curve, the fluctuation range of the PID real-time power compensation is significantly narrowed, and the overall power curve is guided towards the theoretically optimal power allocation scheme to reduce the total heating energy consumption.

[0044] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A water dispenser temperature control method based on AI adaptive algorithm, characterized in that, Includes the following steps: The system acquires real-time environmental monitoring data streams and equipment operating status parameter streams output from the water dispenser's sensor network. The environmental monitoring data streams include ambient temperature, ambient humidity, and the intensity of human movement signals in the surrounding area. The equipment operating status parameter streams include real-time heating chamber temperature, inlet water flow rate, historical heating power data, and historical water usage data for the current time period. The environmental monitoring data stream and the equipment operating status parameter stream are subjected to time-series alignment and feature fusion processing to generate user water use behavior feature vectors and equipment operating condition feature vectors. The user water usage behavior feature vector and the equipment operating condition feature vector are input into the pre-trained adaptive temperature control decision model for calculation, and the target heating temperature setpoint, preheating start advance amount and stepped power adjustment strategy are output. The real-time power compensation is calculated based on the target heating temperature setpoint and the thermal inertia parameters in the equipment operating condition feature vector. The stepped power adjustment strategy is dynamically modified based on the real-time power compensation amount, a power control command sequence is generated, and the power control command sequence is sent to the water dispenser's heating controller for execution.

2. The water dispenser temperature control method based on AI adaptive algorithm according to claim 1, characterized in that, The step of performing time-series alignment and feature fusion processing on the environmental monitoring data stream and the equipment operating status parameter stream to generate user water usage behavior feature vectors and equipment operating condition feature vectors includes: The user water usage behavior feature vector includes predicted water usage time, predicted water temperature preference, and estimated water consumption; the equipment operating condition feature vector includes thermal inertia parameters, heat loss rate, and heating efficiency attenuation coefficient. A unified timeline based on the water dispenser system clock is established, and each data point in the environmental monitoring data stream and each data point in the equipment operating status parameter stream are mapped onto the unified timeline. Based on the fluctuation pattern of the surrounding human movement signal intensity, the time sequence of the user's approach to the water dispenser is identified; The sequence of behavioral time points is matched and associated with the actual water collection time in the current water usage history, and the ambient temperature, real-time temperature change curve of the heating chamber, and water consumption before and after each water usage event are extracted. Using time series analysis algorithms, data from multiple consecutive water usage events are analyzed to extract the user's preference patterns for water temperature under different ambient temperatures and at different times, as well as the average interval between two adjacent water usages, and to generate the predicted water usage time, predicted water temperature preference, and estimated water consumption. Based on the historical heating power record and the real-time temperature change curve of the heating cavity, the thermal inertia parameters, heat loss rate, and heating efficiency attenuation coefficient are calculated by fitting a thermodynamic model.

3. The water dispenser temperature control method based on AI adaptive algorithm according to claim 2, characterized in that, The method utilizes time series analysis algorithms to analyze data from multiple consecutive water usage events, extracting patterns in user preferences for water temperature under different ambient temperatures and at different times, including: Construct a hidden Markov model that takes ambient temperature and time of day as inputs and the actual water temperature finally selected by the user as output. The hidden Markov model is trained using the ambient temperature, the real-time temperature change curve of the heating chamber, and the water usage time data from historical water usage events. Input the current ambient temperature and time point into the trained Hidden Markov Model, and the model outputs the corresponding water temperature state probability distribution. The water temperature value with the highest probability in the water temperature probability distribution is selected as the core value for predicting the water temperature preference, and the variance of the probability distribution is calculated as the confidence level of the predicted water temperature preference.

4. The water dispenser temperature control method based on AI adaptive algorithm according to claim 2, characterized in that, The step of inputting the user water usage behavior feature vector and the equipment operating condition feature vector into a pre-trained adaptive temperature control decision model for calculation, and outputting the target heating temperature setpoint, preheating start advance, and stepped power adjustment strategy, includes: The stepped power adjustment strategy defines the heating power curves used in different heating stages; The predicted water usage time and predicted water temperature preference are input into the demand prediction sub-network of the adaptive temperature control decision model, and the output of the demand prediction sub-network is the optimal water storage temperature range that satisfies the predicted water temperature preference. The thermal inertia parameters and heat loss rate are input into the device dynamic sub-network of the decision model. The device dynamic sub-network calculates the theoretical shortest time and the corresponding theoretical average power required to heat water from the current temperature to the lower limit of the optimal water storage temperature range. The theoretical shortest time is compared with the predicted water usage time, and the time required to start heating in advance is calculated to ensure that users can obtain water with the preferred water temperature within the predicted water usage time, i.e., the preheating start advance amount. Combining the heating efficiency decay coefficient and safety constraints, the theoretical average power is optimized in segments to form a stepped power adjustment strategy consisting of high power rapid heating in the initial stage, smooth transition in the middle stage, and low power heat preservation in the final stage. Set the center value of the optimal water storage temperature range as the target heating temperature setting value.

5. The water dispenser temperature control method based on AI adaptive algorithm according to claim 4, characterized in that, The step of calculating the real-time power compensation based on the target heating temperature setpoint and the thermal inertia parameters in the equipment operating condition feature vector includes: At preset time intervals, the difference between the real-time temperature of the heating chamber and the target heating temperature setting is continuously monitored and recorded as the real-time temperature deviation. The real-time temperature deviation and the thermal inertia parameter are input together into the proportional-integral-derivative controller; The proportional-integral-derivative controller adjusts its control parameters according to the thermal inertia parameter. The larger the thermal inertia parameter, the slower the system temperature response. Therefore, the weight of the integral term is increased to reduce the steady-state error. The proportional-integral-derivative controller calculates based on the adjusted control parameters and the real-time temperature deviation, and outputs the real-time power compensation amount used to dynamically eliminate the real-time temperature deviation.

6. The water dispenser temperature control method based on AI adaptive algorithm according to claim 5, characterized in that, The step-wise power adjustment strategy is dynamically modified based on the real-time power compensation amount to generate a power control command sequence, including: Obtain the preset reference power value corresponding to the current heating stage in the stepped power adjustment strategy; The real-time power compensation amount is superimposed with the preset reference power value to obtain the actual execution power at the current moment; Determine whether the actual power output exceeds the safe operating upper or lower limit of the water dispenser's heating element; If the power exceeds the upper limit, the actual operating power will be limited to the safe operating upper limit value; if it is below the lower limit, the actual operating power will be limited to the safe operating lower limit value or the heating will be turned off. The actual power values ​​calculated and verified at each moment are arranged in chronological order to form the power control command sequence of the power adjustment unit of the direct-drive heating controller.

7. The water dispenser temperature control method based on AI adaptive algorithm according to claim 1, characterized in that, The method also includes an online learning and updating step for the adaptive temperature control decision model: After executing the power control command sequence and completing one heating-water extraction cycle, the actual execution data of the heating-water extraction cycle is collected; the actual execution data includes the complete real-time temperature change curve of the heating chamber, the actual total power consumption, and the user's final actual water temperature. The actual execution data is compared with the predicted water temperature preference and water consumption estimate in the user water use behavior feature vector to generate model prediction error feedback data. The model prediction error feedback data, together with the corresponding environmental monitoring data stream, equipment operating status parameter stream, and decision parameters output by the previous model, constitute a training sample. The training samples are used to incrementally train the pre-trained adaptive temperature control decision model to fine-tune the model parameters, so that the model's subsequent predictions and decisions can better adapt to the actual working conditions of a specific water dispenser and the actual water usage habits of a specific user.

8. The water dispenser temperature control method based on AI adaptive algorithm according to claim 7, characterized in that, The incremental training of the pre-trained adaptive temperature control decision model using the training samples includes: The environmental monitoring data stream and equipment operating status parameter stream in the training samples are used as inputs, and the model prediction error feedback data is used as a supervision signal. The backpropagation algorithm is used to calculate the gradient of the loss function between the current output of the adaptive temperature control decision model and the ideal output based on feedback from actual execution data; The connection weights of each layer of the neural network inside the model are updated along the direction of gradient descent of the loss function with a preset incremental learning rate. After completing incremental learning of a predetermined number of training samples, the performance of the updated model is evaluated using a cross-validation set. If the performance improvement is less than a threshold or a decline occurs, the model is reverted to the version before the update.

9. The water dispenser temperature control method based on AI adaptive algorithm according to claim 1, characterized in that, The pre-trained adaptive temperature control decision model is obtained through the following steps: Construct a neural network model whose input layer corresponds to the user water use behavior feature vector and the equipment operating condition feature vector, and whose output layer corresponds to the target heating temperature setpoint, preheating start advance, and stepped power adjustment strategy parameters. Multiple sets of training samples are obtained. Each set of training samples includes: historical environmental monitoring data stream and equipment operating status parameter stream, historical feature vectors extracted from them, and optimal decision parameters that have been optimized and verified under the historical scenarios. The neural network model is trained under supervision using the multiple sets of training samples, with the loss function being the loss function being the difference between the decision parameters output by the model and the optimal decision parameters. When the model's prediction accuracy on the validation set reaches a preset threshold, training is complete, and the pre-trained adaptive temperature control decision model is obtained.

10. The water dispenser temperature control method based on AI adaptive algorithm according to claim 6, characterized in that, The generation of the power control command sequence also takes into account energy efficiency optimization objectives: Establish a function with the total heating energy consumption as the optimization objective, wherein the variables of the function are the power value and duration of each stage in the stepped power adjustment strategy; Under the constraints of heating water to the target heating temperature setpoint and ensuring the preheating start advance, a set of theoretically optimal power allocation schemes are obtained by solving the extreme values ​​of the function. When generating the actual execution power, the theoretically optimal power allocation scheme is used as a reference, so that while dynamically compensating for temperature deviations, the overall energy consumption approaches the theoretically optimal value.

Images