Dynamic energy efficiency optimization method and system for multi-temperature-zone on-demand water supply of water dispenser

Abstract

The invention discloses a dynamic energy efficiency optimization method and system for multi-temperature-zone on-demand water supply of a water dispenser, and belongs to the technical field of water dispenser control systems. The method comprises the following steps: acquiring user group water consumption historical data and environmental parameters to generate group demand trend prediction, solving to obtain a reference power scheduling scheme, combining user instant intention analysis, generating and executing a fine tuning instruction, collecting energy consumption data and user water taking behavior data in an actual water supply process, and performing power scheduling. And forming a closed-loop feedback data set to update a group demand trend prediction process and a heat conduction coefficient in the dynamic thermal coupling relation graph. And by adopting a control strategy of dynamic physical modeling, behavior prediction analysis and closed-loop feedback learning, power scheduling and instant fine adjustment can be performed on the basis of predicting the water demand, so that the operation energy consumption is reduced, and meanwhile, the quick response to the user demand is ensured.

Description

Technical Field

[0001] This invention relates to the field of water dispenser control system technology, and in particular to a dynamic energy efficiency optimization method and system for multi-temperature zone on-demand water supply in a water dispenser. Background Technology

[0002] Multi-zone water dispensers are common appliances in modern homes and offices, providing hot, warm, and cold water at different temperatures simultaneously, bringing convenience to users. The core of this type of equipment lies in its internal heating, cooling, and insulation control system. This system monitors and adjusts the temperature of each zone's water tank to ensure the water temperature remains within a preset range. The sophistication of its control logic directly determines the equipment's energy efficiency and user experience, falling under the category of process control systems.

[0003] In existing technologies, most water dispensers employ a constant temperature control strategy based on fixed thresholds. Specifically, the control system deploys temperature sensors in the water tanks of each temperature zone. When the sensor detects that the water temperature is below a preset lower threshold, the corresponding heating or cooling module is activated; when the water temperature reaches a preset upper threshold, the power output is shut off. Some products may offer a simple energy-saving mode, such as reducing the frequency of heating or cooling activation by widening the upper and lower limits of temperature control during nighttime or preset non-working periods.

[0004] However, the aforementioned existing technical solutions have significant technical flaws. First, this passive control logic, based purely on internal states, completely ignores actual external water demand, causing the equipment to frequently start up to maintain a precise water temperature even during periods of prolonged inactivity, resulting in substantial waste of standby energy. Second, its response mechanism is lagging; it only starts working when the water temperature deviates from the set value due to water intake or natural cooling. Users often face the predicament of substandard water temperature or having to wait when taking water for the first time after continuous large-volume water intake or after a long period of standby. Finally, its control parameters are relatively fixed once set, failing to dynamically adapt to changes in user habits, seasonal environmental temperature variations, and the aging of the equipment's insulation performance, causing its long-term energy efficiency and performance to gradually deviate from optimal levels. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a dynamic energy efficiency optimization method and system for multi-temperature zone on-demand water supply in water dispensers. Employing a control strategy that combines dynamic physical modeling, behavioral predictive analysis, and closed-loop feedback learning, the system can perform power scheduling and real-time fine-tuning based on predicted water demand, thereby reducing operating energy consumption while ensuring rapid response to user needs.

[0006] The above objectives can be achieved through the following approach: A dynamic energy efficiency optimization method for multi-temperature zone on-demand water supply in a water dispenser includes: constructing the physical topology of the internal heat exchange network of the water dispenser; identifying the thermal coupling relationships between nodes in the physical topology; generating a dynamic thermal coupling relationship diagram, which includes real-time thermal state parameters of each node and thermal conductivity coefficients of the connecting edges; acquiring historical water usage data and environmental parameters of the user group, performing multi-temperature zone water usage habit analysis, and generating a group demand trend prediction; establishing a joint optimization objective function of energy efficiency and response speed based on the dynamic thermal coupling relationship diagram and the group demand trend prediction, and solving for a baseline power scheduling scheme for maintaining the temperature of each temperature zone; monitoring the interaction preparation signal between the user and the water dispenser, and triggering an instant intent parsing process when the interaction preparation signal is detected, generating a fine-tuning instruction for the baseline power scheduling scheme based on the current state of the dynamic thermal coupling relationship diagram; executing the fine-tuning instruction, and collecting energy consumption data and user water consumption behavior data during the actual water supply process to form a closed-loop feedback dataset; and using the closed-loop feedback dataset to synchronously update the group demand trend prediction process and the thermal conductivity coefficients in the dynamic thermal coupling relationship diagram.

[0007] Based on the same inventive concept, this invention also provides a dynamic energy efficiency optimization system for multi-temperature zone on-demand water supply in a water dispenser. The system comprises: a heat network modeling module, used to construct the physical topology of the internal heat exchange network of the water dispenser, identify the thermal coupling relationships between nodes in the physical topology, and generate a dynamic thermal coupling relationship diagram, wherein the dynamic thermal coupling relationship diagram includes real-time thermal state parameters of each node and the thermal conductivity coefficient of the connecting edges; a behavior prediction module, used to acquire historical water usage data and environmental parameters of the user group, perform multi-temperature zone water usage habit analysis, and generate a group demand trend prediction; and a solution solving module, used to establish an energy efficiency and response optimization system based on the dynamic thermal coupling relationship diagram and the group demand trend prediction. The system employs a joint optimization objective function based on speed, and solves for a baseline power scheduling scheme to maintain temperature in each temperature zone. An interaction processing module monitors user interaction preparation signals with the water dispenser. When such a signal is detected, it triggers an immediate intent parsing process, combining the current state of the dynamic thermal coupling graph to generate fine-tuning instructions for the baseline power scheduling scheme. An execution and feedback module executes these fine-tuning instructions and collects energy consumption data and user water consumption behavior data during the actual water supply process, forming a closed-loop feedback dataset. A feedback update module uses this closed-loop feedback dataset to synchronously update the process of predicting the group demand trend and the heat transfer coefficient in the dynamic thermal coupling graph.

[0008] Compared with the prior art, the present invention has the following advantages: 1. The invention is based on long-term trend prediction of user group water usage habits and generates a baseline power scheduling scheme. From a macro perspective, it avoids the ineffective standby energy consumption of traditional water dispensers during periods of no use, and accurately matches energy consumption with actual demand, thereby reducing the overall operating cost of the equipment.

[0009] 2. By monitoring user approach and other interactive preparatory signals, this invention can predict the user's immediate water usage intention in advance and trigger real-time fine-tuning of the power, transforming the water dispenser from a passive response device into an active service device. It can prepare the water temperature before the user operates, reducing the waiting time for the user to get water during peak hours or after the device has been idle for a long time.

[0010] 3. By establishing a closed-loop feedback mechanism, the system can compare actual operating data with model predictions and continuously calibrate its internal physical thermodynamic model and user behavior prediction model using the deviation. This enables it to automatically adapt to changes in user habits, seasonal fluctuations in the environment, and performance degradation caused by aging of the equipment itself. It can maintain optimal operating efficiency throughout its entire life cycle without human intervention. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating a dynamic energy efficiency optimization method for multi-temperature zone on-demand water supply in a water dispenser, according to an embodiment of the present invention.

[0012] Figure 2 This is a parameter curve diagram of multiple temperature zones operating with a dynamic energy efficiency strategy during a certain period of time in an embodiment of the present invention.

[0013] Figure 3 This is a schematic diagram of the structure of a dynamic energy efficiency optimization system for a water dispenser with multi-temperature zones and on-demand water supply, according to an embodiment of the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, 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.

[0015] Reference Figure 1 One embodiment of the present invention proposes a dynamic energy efficiency optimization method for multi-temperature zone on-demand water supply in water dispensers. It adopts a control strategy of dynamic physical modeling, behavior prediction analysis and closed-loop feedback learning, which can perform power scheduling and real-time fine-tuning based on the prediction of water demand, thereby reducing operating energy consumption while ensuring rapid response to user needs.

[0016] The method described in this embodiment specifically includes: S1. Construct the physical topology of the internal heat exchange network of the water dispenser, identify the thermal coupling relationship between each node in the physical topology, and generate a dynamic thermal coupling relationship diagram. The dynamic thermal coupling relationship diagram includes the real-time thermal state parameters of each node and the thermal conductivity coefficient of the connecting edge. Optionally, generating the dynamic thermal coupling diagram includes: Construct the physical topology between the heating modules, cooling modules, water storage containers, and connecting pipes inside the water dispenser; The system controls the water dispenser to sequentially execute predefined standard power test actions, while simultaneously collecting data sequences from the temperature sensors and power meters of each node in the physical topology. By analyzing the data sequence, the correlation strength and delay time between the power changes of different nodes and the temperature changes of other nodes are calculated, and the real-time thermal state parameters of each node in the physical topology and the thermal conductivity coefficient of the connecting edges are quantified. Based on the real-time thermal state parameters, the thermal conductivity coefficient, and the physical topology, a directed weighted graph characterizing the thermal coupling strength is generated as a dynamic thermal coupling relationship graph.

[0017] In this embodiment, through active testing and data analysis, the heat transfer relationship between the internal components of the water dispenser is quantified and constructed into a computable dynamic thermal coupling relationship diagram. First, the physical topology of the water dispenser is constructed. This structure is a digital model, in which each heating module, cooling module, water storage containers in different temperature zones, and connecting pipes are defined as nodes. The physical connections between nodes, such as pipes or contact surfaces, are defined as edges, forming a basic connection diagram.

[0018] To quantify the thermodynamic characteristics of each side in this diagram, the system will control the water dispenser to enter an offline calibration test mode. In this mode, the system will sequentially execute a series of predefined standard power test actions. For example, the system will first apply a constant 800W power pulse to the heating module in the hot water zone for 180 seconds. During this period and the subsequent cooling period, the system will continuously collect and record the power meter reading of the heating module at a sampling frequency of 1 Hz, as well as the temperature data sequences of all temperature sensors, such as NTC thermistors, distributed in the cold water tank, warm water tank, and key locations in the pipeline. After this action is completed, the system will wait for the temperature of all nodes to recover to a quasi-steady state before performing similar power test actions on the cooling module or other heating modules, until all active thermodynamic components have been individually stimulated and tested. This series of operations produces multiple sets of data sequences containing power changes from a single excitation source and the temperature response they cause throughout the system.

[0019] Next, we move into the data analysis phase, precisely extracting the thermodynamic parameters between nodes from the collected data sequence. For any pair of nodes, i.e., node i applying power and node j responding to temperature changes, the system quantifies their coupling relationship using the following algorithm. First, the thermal conductivity coefficient of the connection edge is calculated, which characterizes the intensity of the thermal effect. This coefficient is calculated using the following formula: , in, represents the thermal conductivity coefficient from node i to node j. Its physical meaning is the maximum temperature rise that can be caused at node j by a unit of energy input at node i, and the unit is Kelvin per joule (K / J). It is the maximum temperature change of node j relative to its initial temperature during this test. This is the total energy applied to node i in this test, calculated by adjusting the power... During the test duration Obtained through inner integration, i.e. equal and The product of the power applied signal and the temperature response signals of each node. Simultaneously, the system determines the delay time by analyzing the time relationship between the applied power signal and the temperature response signals of each node. Delay time From the moment the power begins to be applied Temperature change rate to node j The time difference between the first occurrence of a preset sensitivity threshold, such as 0.05 degrees Celsius per second. Real-time thermal state parameters refer to the temperature values ​​directly collected by the temperature sensors at each node at the current moment. .

[0020] Based on the above analysis results, the system generates a dynamic thermal coupling graph. This graph is a directed weighted graph used as a digital twin model for thermodynamic simulation. Each node in the graph corresponds to a component in the physical topology and is accompanied by its real-time thermal state parameters, primarily the current temperature. A directed edge connecting two nodes represents a heat transfer path. The weight of the edge is a tuple containing the quantized thermal conductivity coefficient. and delay time For example, the weight of the edge pointing from the heating module to the hot water tank accurately describes the efficiency and speed of heat transfer to the hot water tank when the heating module is working. Since the thermal state parameters of each node in the diagram are updated in real time, the entire diagram is dynamic and can accurately reflect the internal thermodynamic state of the water dispenser at any given moment.

[0021] S2. Obtain historical water usage data and environmental parameters of user groups, conduct multi-temperature zone water usage habit analysis, and generate group demand trend predictions; Optionally, the generation of group demand trend prediction includes: Obtain historical water usage data for user groups; Acquire historical water usage data of user groups, and decompose the historical water usage data of user groups into a time-cycle-based cyclical pattern, an event-based burst pattern, and random fluctuation residuals; Obtain environmental parameters, which include at least ambient temperature and humidity and date type labels representing weekdays or holidays; By using the cyclic mode, the sudden mode, the random fluctuation residual, and the environmental parameters to perform time series prediction, the probability distribution of water demand in each temperature zone of the water dispenser within multiple preset time periods in the future is obtained, thus constituting a group demand trend prediction.

[0022] In this embodiment, discrete historical water usage behavior and environmental information are transformed into a predictive model for group demand trends that provides a forward-looking and probabilistic description of future water demand across multiple temperature zones. This process begins with the collection and integration of data sources. The system obtains at least three months of historical water usage data from a cloud server or local storage. This data is in time-series format, with each record containing a timestamp accurate to the second, a water temperature zone identifier (e.g., hot water, warm water, or cold water), and the corresponding water output in milliliters. Simultaneously, the system acquires environmental parameters synchronized with the historical water usage data timestamps via built-in sensors or a network API. These parameters include at least the ambient temperature in degrees Celsius, the relative humidity as a percentage, and a date type label indicating whether it is a weekday, weekend, or public holiday.

[0023] After acquiring the data, the system performs time-series decomposition on the historical water consumption data for each temperature zone to reveal its inherent structure. The system employs seasonality and trend decomposition algorithms to break down the original water consumption time series into three main parts: a cyclical pattern based on time cycles, a sudden event pattern based on events, and random fluctuation residuals. The cyclical pattern captures regular water consumption peaks and troughs on a daily or weekly basis, such as coffee breaks at 9 AM or tea breaks at 3 PM on weekdays. The sudden event pattern (Bt) identifies non-periodic large-scale water consumption events by setting a dynamic threshold, such as water consumption exceeding the cyclical pattern's predicted value by three standard deviations for that period, and records the time and intensity of these events. The random fluctuation residuals (Rt) are the unpredictable portion remaining after removing the cyclical and sudden event patterns; the system analyzes their statistical characteristics, such as mean and variance.

[0024] Finally, the system constructs and trains a time series forecasting model to generate predictions of population demand trends. This model uses a predetermined time period in the future, such as 24 hours, with each 15-minute interval as the prediction target. The model's inputs include the future extension values ​​of the cyclical patterns obtained from the aforementioned decomposition, the probability of recurring sudden patterns based on historical patterns, and environmental parameters corresponding to the prediction period, such as the predicted temperature and a defined date type label. The model calculates the probability distribution parameters of water demand in temperature zone z at future time t using the following formula: , in, These are the mean and standard deviation of the predicted water consumption Gaussian distribution, which together constitute the predicted probability distribution. This represents a trained prediction function, such as a gradient booster or long short-term memory network model. It is the cyclic mode component of the z-temperature region at time t. It is the probability of a sudden water use event occurring at time t based on historical statistics. and These are the ambient temperature and humidity at time t and the date type label, respectively. These are the statistical characteristics of historical random fluctuation residuals. The model output, i.e., this series of probability distributions ordered by time, These factors together constitute a quantitative forecast of future water demand in various temperature zones, i.e., a forecast of group demand trends, providing a basis for decision-making in subsequent baseline power dispatching schemes.

[0025] S3. Based on the dynamic thermal coupling relationship diagram and the predicted group demand trend, establish a joint optimization objective function for energy efficiency and response speed, and solve for the baseline power scheduling scheme for maintaining the temperature in each temperature zone. Optionally, the obtained reference power scheduling scheme for maintaining temperature in each temperature zone includes: Based on the aforementioned baseline power scheduling scheme, the total system energy consumption is obtained; Based on the probability distribution in the predicted group demand trend and the dynamic thermal coupling relationship diagram, the waiting time when a user requests water in each temperature zone is calculated under the baseline power scheduling scheme, and the expected response delay penalty term for multiple temperature zones is obtained. Under the constraint of meeting the preset upper and lower limits of water temperature safety in each temperature zone, the optimization algorithm is used to find the power allocation scheme that minimizes the weighted sum of the estimated total energy consumption of the system and the expected response delay penalty, and a baseline power scheduling scheme is obtained. The weighting coefficients used for weighting are dynamically adjusted according to the real-time electricity price signal or the preset energy-saving strategy level.

[0026] In this embodiment, based on predictions of future water demand and an understanding of the thermodynamic characteristics of the equipment, a baseline power scheduling scheme that achieves the optimal balance between energy consumption and user experience is calculated. This scheme is a pre-planned sequence of power allocation instructions for a future time period, such as the next hour. It is achieved by constructing and solving a joint optimization objective function, which quantitatively weighs operating energy consumption against service response speed.

[0027] The system constructs the total system energy consumption term based on the baseline power scheduling scheme to be solved, i.e., the power sequence of the heating or cooling modules in each temperature zone over a future time period. This term is one of the core costs of the scheme. The total system energy consumption term is calculated using the following formula: , in, This represents the total energy consumption expected within the optimization time window T, expressed in joules. It is the instantaneous power allocated by the system to the z-th temperature zone, in watts. This is the decision variable that the optimization algorithm needs to solve. This represents the sum of all temperature zones, such as hot water, warm water, and cold water.

[0028] The system constructs a multi-temperature zone expected response delay penalty term to quantify the quality of user experience. The calculation of this term deeply integrates group demand trend prediction and dynamic thermal coupling relationship graph. The system first calculates the water demand probability distribution for each future time point t given in the group demand trend prediction. and The system simulates user requests. For each possible request, the system utilizes a dynamic thermal coupling graph and the current baseline power scheduling scheme. Perform rapid thermal simulation to calculate the waiting time from when a user initiates a request until the water temperature in that temperature zone reaches a usable standard, such as 85 degrees Celsius for hot water. The expected response delay is the mathematical expectation of the waiting time under the probability distribution of demand. To achieve a unified dimension for optimization with the energy consumption term, the system introduces a weighting coefficient to convert the time delay into an equivalent energy penalty, resulting in the expected response delay penalty term D, which is calculated as shown in the formula: , Where D represents the total delay penalty cost, in joules. It is a dynamically adjusted weighting coefficient, measured in watts, and its physical meaning is the extra power cost that the system is willing to pay to shorten the user's waiting time by one second. It is the expected waiting time for temperature zone z at time t, obtained by taking a probability-weighted average of the waiting times under all possible water intake scenarios.

[0029] Finally, under the strict constraints of meeting the preset upper and lower safety limits of water temperature in each temperature zone (e.g., 4 to 10 degrees Celsius for cold water and 85 to 95 degrees Celsius for hot water), the system uses optimization algorithms such as nonlinear programming or model predictive control to find the power allocation scheme that minimizes the following joint optimization objective function J. The joint optimization objective function J is: , The solution to this optimization problem is the optimal power sequence. This is the final output baseline power scheduling scheme. Among them, the weighting coefficients... Dynamic adjustment is key to achieving energy efficiency strategies. During peak electricity price periods, The value of will decrease, for example, to between 0.1 and 0.5, making the optimization objective more biased towards minimizing energy consumption E. Conversely, during off-peak electricity periods or in the user-defined "performance mode," The value of will increase, for example, between 1.0 and 2.0, making the optimization algorithm more inclined to sacrifice some energy consumption in exchange for lower response latency.

[0030] S4. Monitor the interaction preparation signal between the user and the water dispenser. When the interaction preparation signal is detected, trigger the real-time intent parsing process and generate a fine-tuning instruction for the baseline power scheduling scheme in combination with the current state of the dynamic thermal coupling relationship diagram. Optionally, the monitoring of user interaction preparation signals with the water dispenser, and the triggering of the real-time intent parsing process when the interaction preparation signal is detected, includes: The system detects when a user enters a trigger area centered on the water dispenser with a preset radius, and generates an interaction preparation signal. After generating the interaction preparation signal, the user's identity identifier is obtained, and the user's movement trajectory and dwell time in the trigger area are recorded to form a short-stay behavior pattern; By integrating the user's identity identifier, the short-stay behavior pattern, the current time, and the predicted group demand trend, the probability of the user's intention pointing to each temperature zone is calculated, and an instant probability vector is generated.

[0031] In this embodiment, the physical act of a user approaching the water dispenser is analyzed in real time as a quantified instantaneous intent probability distribution pointing to a specific temperature zone, thereby providing input for instantaneous fine-tuning of the power.

[0032] The system continuously scans a virtual circular area, or trigger zone, centered on the water dispenser and with a preset radius of 1.5 to 3 meters, using built-in millimeter-wave radar or image sensors. Once a user is detected entering this area, the system determines this event as an interaction preparation signal and immediately triggers the real-time intent resolution process.

[0033] After the interaction preparation signal is generated, the system immediately initiates multimodal information acquisition. It acquires the user's identity using Bluetooth Low Energy beacons or near-field facial recognition technology. Simultaneously, sensors continuously track the user's movement coordinate sequence within the triggered area, recording the complete time and path from entry to final stop in front of the water dispenser. By analyzing the path's curvature, average speed, and final stopping position, the system quantifies and encodes this dynamic information, forming a short-term dwelling behavior pattern vector representing the individual's behavioral habits.

[0034] For example, the millimeter-wave radar outputs dynamic point cloud data of objects entering the trigger area at a frequency of 10 Hz. This data is processed into a two-dimensional coordinate sequence. The system extracts the following four key features and encodes them into normalized behavioral feature vectors. Here, it is set as a 4-dimensional vector: (1) Trajectory linearity index The formula used to distinguish between targeted "water collection" and unintentional "passing by" is as follows: Approaching 1 indicates heading straight for the water dispenser, while approaching 0 indicates lingering. (2) Average approach speed in the last meter The average speed was calculated from data segments less than 1 meter away from the water dispenser. Higher speeds are usually associated with the user's urgency (weighted association with the demand for quick water dispensing). This physical value was normalized to the [0, 1] interval by Max-Min scaling. (3) Interactive forward angle deviation The cosine of the angle between the human torso orientation vector and the water dispenser normal is used for millimeter-wave radar detection. That is, facing it directly increases the probability of getting water; That is, turning sideways, possibly just passing by. (4) Micro-pause count value The proportion of frames where the user's speed is detected to be below a threshold, such as 0.1 m / s, and who do not actually remain at the water intake is recorded to capture hesitant behavior. Therefore, the brief pause behavior pattern is concretized as follows: The 4-dimensional floating-point eigenvector.

[0035] The system performs intent fusion calculations to generate a final real-time probability vector. This vector is a normalized array representing the user's probability of demanding each temperature zone, such as hot water, warm water, and cold water. The system takes the acquired user identity, the encoded short-stay behavior pattern vector, the current precise time, and the baseline probability of water usage in each temperature zone at the current moment extracted from the group demand trend prediction as input, and feeds them into a pre-trained lightweight classification model, such as a Bayesian classifier or a small neural network. This model is calculated using the following formula: , in, It is the instantaneous probability vector of the output, and its components are... This represents the probability that the user intends to target the z-temperature region. The softmax function ensures that the sum of all components is 1. This represents the classification model. U is the input user identity identifier. M is the quantized short-stay behavior pattern vector. G represents the current time characteristic, such as the number of minutes in a day. G is the prior probability vector for the current moment provided by the group demand trend prediction. This calculated result, the instantaneous probability vector, provides a direct, probabilistic decision-making basis for subsequent power fine-tuning.

[0036] Optionally, generating fine-tuning instructions for the reference power scheduling scheme based on the current state of the dynamic thermal coupling diagram includes: The urgency of demand for each temperature zone is determined based on the portion of the instantaneous probability vector that exceeds a preset baseline probability threshold. Based on the thermal state parameters in the dynamic thermal coupling diagram and the urgency of the demand, the required instantaneous power increase is calculated to obtain the power adjustment amount; Based on the power adjustment amount and the thermal conductivity coefficient of the dynamic thermal coupling diagram, calculate the expected thermal disturbance to other temperature zones caused by applying the power adjustment amount. With the goal of offsetting the expected thermal disturbance and maintaining the system's energy efficiency balance, control commands are generated for real-time adjustment of the power distribution in each temperature zone, serving as fine-tuning commands.

[0037] In this embodiment, as Figure 2 As shown, the probabilistic prediction of user interaction intent is transformed into a set of power fine-tuning commands in real time. This allows for rapid response to immediate user needs and proactive management of internal thermodynamic side effects without compromising baseline energy efficiency strategies. This process begins with the analysis of the real-time probability vector. Based on the probability values ​​of each temperature zone within the vector, the urgency of the demand is determined. The system identifies probability values ​​exceeding a preset baseline probability threshold, such as 0.5, as valid demand signals. Demand Urgency Defined as the instantaneous probability of that temperature range A positively correlated function used to quantify the intensity of demand.

[0038] Based on the urgency of this requirement, the system combines the real-time thermal state parameters of this temperature zone in the dynamic thermal coupling diagram, mainly the current water temperature. Calculate the instantaneous power increase required to meet potential user demand, i.e., the power adjustment amount. This calculation aims to raise the water temperature to the target usable temperature within an estimated, extremely short preparation time, such as 5 to 10 seconds. With this as the objective, the power adjustment is a function of the urgency of demand and the temperature difference, ensuring an immediate response.

[0039] After determining the power adjustment amount for the main target temperature range Subsequently, the system utilizes a dynamic thermal coupling relationship diagram to perform forward-looking perturbation analysis. This is based on the power adjustment amount and the thermal conductivity coefficients connecting the target temperature zone z with other temperature zones i in the diagram. The system calculates the expected thermal disturbances to other temperature zones caused by applying this power, which manifest as temperature changes over a short period of time. This step transforms the thermodynamic coupling effect from a passive influence into a predictable and manageable active control variable.

[0040] Finally, the system generates fine-tuning instructions for compensation and adjustment. The core objective of these instructions is to counteract the anticipated thermal disturbance while maintaining overall system efficiency. For the non-target temperature range i affected by the disturbance, the system calculates a compensation power. This is to offset the expected temperature rise or drop. The compensation power is calculated as shown in the formula: , in, It is the compensation power adjustment applied to temperature zone i. It is a dimensionless damping coefficient ranging from 0 to 1, used to adjust the compensation intensity to balance energy efficiency, for example, set to 0.8. It is the thermal conductivity coefficient from the primary target temperature region z to the affected temperature region i. It is the self-heating conductivity coefficient of temperature zone i itself, that is, the efficiency of its own heating or cooling module. Both coefficients are directly read from the dynamic thermal coupling relationship diagram. This is the power adjustment amount for the main target temperature range. Ultimately, this set of main adjustment amounts... and various compensation adjustments The resulting power adjustment vector will be superimposed on the currently executing baseline power scheduling scheme as a fine-tuning instruction, forming the final control signal sent to each temperature zone actuator.

[0041] S5. Execute the fine-tuning instruction and collect energy consumption data and user water consumption behavior data during the actual water supply process to form a closed-loop feedback dataset. Optionally, the closed-loop feedback dataset includes: Execute the fine-tuning command, collect the water temperature response curve, real-time power curve and execution timestamp of each temperature zone of the water dispenser after the fine-tuning command is executed, and generate the actual operating status sequence. The baseline power scheduling scheme and the fine-tuning instruction are used as expected comparison standards. The actual operating state sequence is compared with the expected comparison standards in the time domain to calculate the deviation magnitude and generate an execution performance deviation vector. The correlation between the execution performance deviation vector and the user identity and short-stay behavior pattern is analyzed, outlier noise caused by hardware failures is removed, and a closed-loop feedback dataset containing the mapping relationship between input conditions and output deviation is selected and generated. In this embodiment, the actual physical response after the system executes the control command is precisely quantified and compared with the expected target, thereby constructing a structured closed-loop feedback dataset that includes input conditions, expected response, actual results and performance deviations, providing high-quality empirical evidence for the subsequent adaptive update of the model.

[0042] After the system sends the fine-tuning command to the power control unit of each temperature zone, it immediately starts to continuously record the operating status of the water dispenser at a sampling frequency of, for example, 1 Hz, and collects and generates an actual operating status sequence that includes a precise execution timestamp, the water temperature response curve of each temperature zone, and the real-time power curve.

[0043] The baseline power scheduling scheme prior to the system call issuing the fine-tuning instruction, along with the fine-tuning instruction itself, serves as the expected benchmark. This benchmark defines the system's desired power output and temperature response under an ideal model. By aligning the actual operating state sequence with this expected benchmark in the time domain—that is, using the execution timestamp of the fine-tuning instruction as a common zero point—the system compares the actual response with the expected target point by point, and generates an execution performance deviation vector using the following formula. : , in, It is a multidimensional vector, and its components quantify the performance deviations in different dimensions. It is the difference between the actual water temperature and the target water temperature in temperature zone z at the end of the response, in degrees Celsius. It is the difference between the actual time to reach the target water temperature and the time predicted by the system, expressed in seconds. It is the difference between the actual electrical energy consumed during this response and the electrical energy predicted by the system based on the dynamic thermal coupling relationship diagram, expressed in joules.

[0044] System analysis execution performance deviation vector The statistical correlation between the user's identity and short-term behavior patterns that triggered the fine-tuning was established. During this process, the system applied anomaly detection algorithms, such as the 3σ criterion based on historical deviation distribution, to identify and eliminate outlier noise data points caused by sporadic hardware failures, such as aging heating elements or pump blockage, rather than inaccurate model predictions. After filtering, the system integrated and stored each valid input-output relationship—including user identity, behavior patterns, initial thermal state, fine-tuning instructions, and the final performance deviation vector—to form the final closed-loop feedback dataset. This dataset accurately establishes the mapping relationship between user and environmental inputs and system performance output deviations, providing a foundation for subsequent model self-correction.

[0045] S6. Using the closed-loop feedback dataset, synchronously update the process of predicting the group demand trend and the heat conduction coefficient in the dynamic thermal coupling relationship diagram.

[0046] Optionally, the process of updating the group demand trend forecast includes: Extract the distribution characteristics of actual water consumption and water consumption time from the closed-loop feedback dataset, calculate the difference between the actual water consumption and the corresponding probability distribution in the group demand trend prediction, and generate a demand prediction residual sequence. The demand forecast residual sequence is decomposed into the dimensions corresponding to the cyclic mode, the burst mode, and the random fluctuation residual, and the drift of each dimension is calculated to quantify and generate the long-term and short-term correction factors of the model. The process of updating time series forecasts using the long-term and short-term correction factors of the model.

[0047] In this embodiment, the closed-loop feedback data collected during actual system operation is used to self-correct and iteratively optimize the group demand trend prediction model, thereby continuously improving its accuracy in predicting future water usage behavior. This process is achieved through a periodic model update mechanism, for example, executed every 24 hours. The system extracts the actual water consumption records of all users in the past period from the closed-loop feedback dataset, including water consumption timestamps and water consumption amounts, and aggregates them according to a preset time granularity, such as 15 minutes, to obtain the actual water consumption time series for each temperature zone.

[0048] The system compares the actual water consumption sequence with the predicted values ​​generated by the group demand trend prediction model for the same time period, calculates the difference between the two, and generates a demand prediction residual sequence. This residual sequence precisely quantifies when, in what temperature range, and to what extent the model prediction deviated from the actual situation.

[0049] The system performs pattern decomposition on the demand forecast residual sequence. Through signal processing techniques such as filtering and correlation analysis, the system decomposes the total residual and attributes it to the three basic dimensions that constitute the forecast model. If the residual exhibits obvious daily or weekly periodicity, its main component is attributed to the correction amount of the cyclical pattern. If the residual shows a few isolated, large-amplitude spikes, it is identified as unpredicted events or intensity prediction biases in the burst pattern. The remaining residual components without obvious patterns are used to update the statistical characteristics of the random fluctuation residual. By accumulating or moving averaged the residual components in each dimension, the system calculates the drift amount for each dimension. This drift amount intuitively reflects the systematic bias of the model in each pattern.

[0050] Finally, the system uses these quantified drift values ​​to generate long-term and short-term correction factors for the model, and uses these to update the time series forecasts. This update process is implemented using the following formula: , In formula H, and These are the model's internal parameter sets before and after the update, respectively. and These are the model's long-term and short-term correction factors. They are essentially learning rates of different dimensions, typically ranging from 0.01 to 0.1. They are used to control the step size of updates and prevent the model from overfitting due to a single large deviation. It is a parameter gradient derived from the cyclical mode drift, used to adjust the long-term periodic behavior of the model. This is a parameter gradient derived from the burst pattern drift, used to quickly adapt to newly emerging short-term event patterns. By applying this update formula, time series forecasting models can continuously learn from the gradual and abrupt changes in user group habits, achieving adaptive evolution of predictive capabilities.

[0051] Optionally, updating the thermal conductivity coefficient in the dynamic thermal coupling graph includes: Filter the data segments in the closed-loop feedback dataset that are in non-user water intake periods and whose temperature change rate is less than a preset change rate threshold, extract the ambient temperature, water temperature decay rate of each node and maintenance power, and generate static heat preservation verification data. Substitute the static thermal insulation verification data into the current dynamic thermal coupling relationship diagram to perform reverse thermal simulation, calculate the Euclidean distance between the theoretical temperature change and the actual temperature change of each node based on the current thermal conductivity coefficient, and generate the thermal conductivity model error matrix. With the goal of minimizing the error matrix of the thermal conductivity model, the weights of the edges between nodes in the physical topology are iteratively corrected to obtain the calibrated thermal conductivity coefficient.

[0052] In this embodiment, the heat transfer coefficient in the dynamic thermal coupling diagram is continuously calibrated online using the natural heat dissipation data of the water dispenser in standby mode. This ensures the long-term accuracy of the physical model and compensates for model drift caused by equipment aging or environmental changes. The system traverses the dataset to find data segments during non-user water consumption periods where the temperature change rate of all temperature zone nodes is less than a preset extremely low change rate threshold, such as 0.1 degrees Celsius per minute. From these segments, the system accurately extracts the ambient temperature, the natural temperature decay rate of each water storage container node, and the small sustaining power provided by the power control unit to maintain this quasi-steady state, typically in the range of 5 to 15 watts.

[0053] The system utilizes the extracted static insulation calibration data to perform a reverse thermal simulation of the current dynamic thermal coupling relationship diagram. The core logic of this simulation is to calculate the theoretically expected rate of temperature change for each node using the existing thermal conductivity coefficients in the model, given the actual measured sustaining power and ambient temperature. Subsequently, the system compares this theoretical rate of temperature change vector with the actual temperature decay rate vector extracted from the static insulation calibration data, and calculates the thermal conductivity model error matrix between the two using the following formula I, which is typically quantified as a scalar error value in practice: , Where M represents the overall error metric of the current thermal conductivity model. This represents summing over all nodes i. It is the actual temperature decay rate recorded by node i in the static insulation verification data, in degrees Celsius per second. The inverse thermal simulation is the theoretical rate of temperature change calculated based on the current thermal conductivity. Its calculation considers not only the sustaining power of node i, but also the thermal effects of other nodes coupled through the thermal conductivity and the heat dissipation to the environment. The physical meaning of this error M is the squared Euclidean distance between the model-predicted heat loss rate and the real-world heat loss rate.

[0054] The system initiates an iterative correction process with the optimization objective of minimizing the error M of the thermal conductivity model. The system employs gradient descent or a similar optimization algorithm to calculate the weights of the edges between nodes in the physical topology based on the error M, i.e., the thermal conductivity coefficients. The partial derivatives are used to determine the correction direction. The algorithm iteratively adjusts the values ​​of each thermal conductivity coefficient in small steps, with each adjustment aiming to make the calculated theoretical temperature change rate closer to the actual observed value. The correction process stops when the error M converges to a minimum or reaches the preset number of iterations. At this point, the weights of each edge in the model are the thermal conductivity coefficients calibrated with real data, allowing the dynamic thermal coupling relationship diagram to more accurately reflect the current true thermodynamic characteristics of the water dispenser.

[0055] Based on the same inventive concept, such as Figure 3 As shown, the present invention also provides a dynamic energy efficiency optimization system for multi-temperature zone on-demand water supply in a water dispenser, characterized in that the system comprises: The heat network modeling module is used to construct the physical topology of the heat exchange network inside the water dispenser, identify the thermal coupling relationship between each node in the physical topology, and generate a dynamic thermal coupling relationship diagram. The dynamic thermal coupling relationship diagram includes the real-time thermal state parameters of each node and the thermal conductivity coefficient of the connecting edge. The behavior prediction module is used to acquire historical water usage data and environmental parameters of user groups, conduct multi-temperature zone water usage habit analysis, and generate group demand trend predictions. The solution-solving module is used to establish a joint optimization objective function of energy efficiency and response speed based on the dynamic thermal coupling relationship diagram and the predicted group demand trend, and solve for the baseline power scheduling scheme for maintaining the temperature in each temperature zone. The interaction processing module is used to monitor the interaction preparation signal between the user and the water dispenser. When the interaction preparation signal is detected, the real-time intent parsing process is triggered. Combined with the current state of the dynamic thermal coupling relationship diagram, a fine-tuning instruction for the baseline power scheduling scheme is generated. The execution and feedback module is used to execute the fine-tuning instructions and collect energy consumption data and user water consumption behavior data during the actual water supply process to form a closed-loop feedback dataset. The feedback update module is used to synchronously update the process of predicting the group demand trend and the heat conduction coefficient in the dynamic thermal coupling relationship diagram using the closed-loop feedback dataset.

[0056] To verify the feasibility and specific effects of this invention in practice, it was applied to a commercial multi-temperature zone water dispenser deployed in an office environment accommodating 50 people. This water dispenser has three independent temperature zones: a hot water zone, a warm water zone, and a cold water zone.

[0057] The water dispenser's physical configuration includes a 5-liter hot water tank equipped with a 2000-watt heating module; a 2-liter cold water tank equipped with a 75-watt compressor cooling module; and a 3-liter warm water tank that regulates temperature by mixing hot and cold water. The system collects data at a frequency of 1 Hz using NTC thermistors distributed in each water tank, key piping nodes, and the environment; power meters monitoring the heating and cooling modules; and a millimeter-wave radar sensor with a detection range of 3 meters. The test environment was a constant 22°C office.

[0058] To generate the dynamic thermal coupling diagram, the system performed an offline calibration test. The system applied a power pulse of 1500 watts for 120 seconds to the heating module (node ​​H) in the hot water zone. The total energy applied during this period was... The result is 500 * 120 = 180,000 Joules. The system collected data showing that under this excitation, the hot water tank's own temperature rose by 21.5 K, while simultaneously, due to heat conduction, the cold water tank (node ​​C) experienced its maximum temperature rise. The maximum temperature rise of the warm water tank (node ​​W) is 1.2 K. The value is 3.5 K. The thermal conductivity between nodes can be calculated using the formula. For example, the thermal conductivity from the hot water tank to the cold water tank. The calculation is as follows: , Simultaneously, the system recorded the delay time at which the cold water tank temperature change rate first exceeded the 0.05 K / s threshold. The time is 45 seconds. By performing this type of test on each module in sequence, the system constructs a directed weighted graph that includes the weights of the thermal conductivity coefficients and delay times between each node, i.e., a dynamic thermal coupling graph.

[0059] Secondly, the system generates a trend forecast of group demand based on the collected user water usage history data from the past three months. Data shows a significant peak in hot water usage at 9:00 AM on weekdays, primarily for brewing coffee and tea, with an average demand of 4.5 liters. The system uses time series decomposition to identify this pattern as a cyclical pattern. Based on the formula, and combining historical data with the date type (weekday), the system predicts that the probability distribution of hot water demand between 8:55 AM and 9:10 AM on the next weekday will be a Gaussian distribution with a mean μ = 4.5 liters and a standard deviation σ = 0.7 liters.

[0060] Based on this prediction and the established dynamic thermal coupling graph, the system solves for the baseline power scheduling scheme. To ensure a good user experience during the 9:00 peak period, the system adjusts the weighting coefficient of the expected response latency penalty term during this time period. Setting the power to a higher 1.8 watts indicates that the system is willing to expend an additional 1.8 watts of power to reduce the waiting time by 1 second. The optimization objective function J=E+D shows that to keep the average user waiting time below 5 seconds, the hot water tank temperature needs to reach 94℃ before 9:00 AM. Considering the current water temperature of 88℃ and ambient heat dissipation, the baseline power scheduling scheme instructs the heating module to operate at an average power of 600 watts for 8 minutes starting at 8:50 AM to achieve pre-stored energy.

[0061] In real-time operation, the instant response mechanism of this invention was verified. At 9:02 AM, the millimeter-wave radar detected a user entering the 2.5-meter trigger zone. Based on the user's identity identifier, historical data showed the user to be a frequent hot beverage user, and their movement trajectory towards the water dispenser, the interaction processing module calculated an instant probability vector. [Hot water: 0.90, Warm water: 0.08, Cold water: 0.02]. At this point, the hot water tank has been partially used, and the water temperature has dropped to 93°C. In response to this high-probability immediate intention, the system determines the target water temperature to be 95°C and calculates the required power adjustment of 500 watts for 10 seconds. To quickly reach that temperature. Simultaneously, to offset the thermal disturbance caused by this power adjustment to other temperature zones, the system calculates the compensation power for the cold water tank. This is done by substituting the previously measured thermal conductivity coefficient. Based on the relevant parameters, the compensation power is calculated. The value is -15 watts. The system ultimately generates and executes a fine-tuning instruction: the power of the hot water module is increased by 500 watts from the baseline, while the power of the cooling module is increased by 15 watts to offset the negative adjustment, lasting for 10 seconds.

[0062] After this interaction, the execution and feedback module collected data showing an actual water output waiting time of 4 seconds and energy consumption of 5.2 kJ. Based on the formula, the system generated an execution performance deviation vector. This data is then stored in the closed-loop feedback dataset.

[0063] Finally, the closed-loop feedback dataset is used for continuous model updates. For example, the system monitored actual hot water demand at 9:00 AM on weekdays for several consecutive weeks and found it to be consistently around 4.8 liters, higher than the original forecast of 4.5 liters. Based on this, the feedback update module calculates the demand forecast residual and uses the model's long-term and short-term correction factors according to the formula. The parameters of the cyclic mode in the time series prediction model were updated. Meanwhile, during off-peak hours at night, the system screened static insulation verification data and found that the actual natural cooling rate of the hot water tank was 5% faster than the model prediction. Based on Formula I, the system iteratively corrected the heat transfer coefficient of the hot water tank to the environment, aiming to minimize the thermal conductivity model error matrix M, thereby improving the accuracy of the dynamic thermal coupling relationship diagram.

[0064] It should be noted that the electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any method of indirect connection is applicable to the embodiments of the present invention as long as it achieves the purpose of the present invention. The above description is merely an exemplary embodiment of the present invention and should not be construed as limiting the scope of the present invention.

[0065] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A dynamic energy efficiency optimization method for multi-temperature zone on-demand water supply in a water dispenser, characterized in that... The method includes: Construct the physical topology of the internal heat exchange network of the water dispenser, identify the thermal coupling relationship between each node in the physical topology, and generate a dynamic thermal coupling relationship diagram. The dynamic thermal coupling relationship diagram includes the real-time thermal state parameters of each node and the thermal conductivity coefficient of the connecting edges. Acquire historical water usage data and environmental parameters of user groups, conduct multi-temperature zone water usage habit analysis, and generate group demand trend predictions; Based on the dynamic thermal coupling relationship diagram and the predicted demand trend of the group, a joint optimization objective function for energy efficiency and response speed is established, and the baseline power scheduling scheme for maintaining the temperature in each temperature zone is obtained by solving the problem. The system monitors the user's interaction preparation signal with the water dispenser. When the interaction preparation signal is detected, an instant intent parsing process is triggered. Based on the current state of the dynamic thermal coupling relationship diagram, a fine-tuning instruction for the baseline power scheduling scheme is generated. The fine-tuning instructions are executed, and energy consumption data and user water consumption behavior data are collected during the actual water supply process to form a closed-loop feedback dataset. Using the closed-loop feedback dataset, the process of predicting the group demand trend and the heat transfer coefficient in the dynamic thermal coupling relationship diagram are updated synchronously.

2. The dynamic energy efficiency optimization method for multi-temperature zone on-demand water supply in a water dispenser according to claim 1, characterized in that, The generation of the dynamic thermal coupling relationship diagram includes: Construct the physical topology between the heating modules, cooling modules, water storage containers, and connecting pipes inside the water dispenser; The system controls the water dispenser to sequentially execute predefined standard power test actions, while simultaneously collecting data sequences from the temperature sensors and power meters of each node in the physical topology. By analyzing the data sequence, the correlation strength and delay time between the power changes of different nodes and the temperature changes of other nodes are calculated, and the real-time thermal state parameters of each node in the physical topology and the thermal conductivity coefficient of the connecting edges are quantified. Based on the real-time thermal state parameters, the thermal conductivity coefficient, and the physical topology, a directed weighted graph characterizing the thermal coupling strength is generated as a dynamic thermal coupling relationship graph.

3. The dynamic energy efficiency optimization method for multi-temperature zone on-demand water supply in a water dispenser according to claim 2, characterized in that, The generated group demand trend prediction includes: Obtain historical water usage data for user groups; Acquire historical water usage data of user groups, and decompose the historical water usage data of user groups into a time-cycle-based cyclical pattern, an event-based burst pattern, and random fluctuation residuals; Obtain environmental parameters, which include at least ambient temperature and humidity and date type labels representing weekdays or holidays; By using the cyclic mode, the sudden mode, the random fluctuation residual, and the environmental parameters to perform time series prediction, the probability distribution of water demand in each temperature zone of the water dispenser within multiple preset time periods in the future is obtained, thus forming a group demand trend prediction.

4. The dynamic energy efficiency optimization method for multi-temperature zone on-demand water supply in a water dispenser according to claim 3, characterized in that, The obtained baseline power scheduling scheme for maintaining temperature in each temperature zone includes: Based on the aforementioned baseline power scheduling scheme, the total system energy consumption is obtained; Based on the probability distribution in the predicted group demand trend and the dynamic thermal coupling relationship diagram, the waiting time when a user requests water in each temperature zone is calculated under the baseline power scheduling scheme, and the expected response delay penalty term for multiple temperature zones is obtained. Under the constraint of meeting the preset upper and lower limits of water temperature safety in each temperature zone, the optimization algorithm is used to find the power allocation scheme that minimizes the weighted sum of the estimated total energy consumption of the system and the expected response delay penalty, and a baseline power scheduling scheme is obtained. The weighting coefficients used for weighting are dynamically adjusted according to the real-time electricity price signal or the preset energy-saving strategy level.

5. The dynamic energy efficiency optimization method for multi-temperature zone on-demand water supply in a water dispenser according to claim 4, characterized in that, The monitoring of user interaction preparation signals with the water dispenser, and the triggering of the instant intent parsing process when the interaction preparation signal is detected, includes: The system detects when a user enters a trigger area centered on the water dispenser with a preset radius, and generates an interaction preparation signal. After generating the interaction preparation signal, the user's identity identifier is obtained, and the user's movement trajectory and dwell time in the trigger area are recorded to form a short-stay behavior pattern; By integrating the user's identity identifier, the short-stay behavior pattern, the current time, and the predicted group demand trend, the probability of the user's intention pointing to each temperature zone is calculated, and an instant probability vector is generated.

6. The dynamic energy efficiency optimization method for multi-temperature zone on-demand water supply in a water dispenser according to claim 5, characterized in that, The step of generating fine-tuning instructions for the baseline power scheduling scheme based on the current state of the dynamic thermal coupling relationship diagram includes: The urgency of demand for each temperature zone is determined based on the portion of the instantaneous probability vector that exceeds a preset baseline probability threshold. Based on the thermal state parameters in the dynamic thermal coupling diagram and the urgency of the demand, the required instantaneous power increase is calculated to obtain the power adjustment amount; Based on the power adjustment amount and the thermal conductivity coefficient of the dynamic thermal coupling diagram, calculate the expected thermal disturbance to other temperature zones caused by applying the power adjustment amount. With the goal of offsetting the expected thermal disturbance and maintaining the system's energy efficiency balance, control commands are generated for real-time adjustment of the power distribution in each temperature zone, serving as fine-tuning commands.

7. The dynamic energy efficiency optimization method for multi-temperature zone on-demand water supply in a water dispenser according to claim 1, characterized in that, The closed-loop feedback dataset includes: Execute the fine-tuning command, collect the water temperature response curve, real-time power curve and execution timestamp of each temperature zone of the water dispenser after the fine-tuning command is executed, and generate the actual operating status sequence. The baseline power scheduling scheme and the fine-tuning instruction are used as expected comparison standards. The actual operating state sequence is compared with the expected comparison standards in the time domain to calculate the deviation magnitude and generate an execution performance deviation vector. The correlation between the execution performance deviation vector and the user identity and short-term dwell behavior pattern is analyzed. Outlier noise caused by hardware failure is removed, and a closed-loop feedback dataset containing the mapping relationship between input conditions and output deviation is selected and generated.

8. The dynamic energy efficiency optimization method for multi-temperature zone on-demand water supply in a water dispenser according to claim 1, characterized in that, The process of updating the forecast of the group's demand trends includes: Extract the distribution characteristics of actual water consumption and water consumption time from the closed-loop feedback dataset, calculate the difference between the actual water consumption and the corresponding probability distribution in the group demand trend prediction, and generate a demand prediction residual sequence. The demand forecast residual sequence is decomposed into the dimensions corresponding to the cyclic mode, the burst mode, and the random fluctuation residual, and the drift of each dimension is calculated to quantify and generate the long-term and short-term correction factors of the model. The process of updating time series forecasts using the long-term and short-term correction factors of the model.

9. The dynamic energy efficiency optimization method for multi-temperature zone on-demand water supply in a water dispenser according to claim 1, characterized in that, Updating the thermal conductivity coefficient in the dynamic thermal coupling graph includes: Filter the data segments in the closed-loop feedback dataset that are in non-user water intake periods and whose temperature change rate is less than a preset change rate threshold, extract the ambient temperature, water temperature decay rate of each node and maintenance power, and generate static heat preservation verification data. Substitute the static thermal insulation verification data into the current dynamic thermal coupling relationship diagram to perform reverse thermal simulation, calculate the Euclidean distance between the theoretical temperature change and the actual temperature change of each node based on the current thermal conductivity coefficient, and generate the thermal conductivity model error matrix. With the goal of minimizing the error matrix of the thermal conductivity model, the weights of the edges between nodes in the physical topology are iteratively corrected to obtain the calibrated thermal conductivity coefficient.

10. A dynamic energy efficiency optimization system for multi-temperature zone on-demand water supply in a water dispenser, characterized in that, The system includes: The heat network modeling module is used to construct the physical topology of the heat exchange network inside the water dispenser, identify the thermal coupling relationship between each node in the physical topology, and generate a dynamic thermal coupling relationship diagram. The dynamic thermal coupling relationship diagram includes the real-time thermal state parameters of each node and the thermal conductivity coefficient of the connecting edge. The behavior prediction module is used to acquire historical water usage data and environmental parameters of user groups, conduct multi-temperature zone water usage habit analysis, and generate group demand trend predictions. The solution-solving module is used to establish a joint optimization objective function of energy efficiency and response speed based on the dynamic thermal coupling relationship diagram and the predicted group demand trend, and solve for the baseline power scheduling scheme for maintaining the temperature in each temperature zone. The interaction processing module is used to monitor the interaction preparation signal between the user and the water dispenser. When the interaction preparation signal is detected, the real-time intent parsing process is triggered. Combined with the current state of the dynamic thermal coupling relationship diagram, a fine-tuning instruction for the baseline power scheduling scheme is generated. The execution and feedback module is used to execute the fine-tuning instructions and collect energy consumption data and user water consumption behavior data during the actual water supply process to form a closed-loop feedback dataset. The feedback update module is used to synchronously update the process of predicting the group demand trend and the heat conduction coefficient in the dynamic thermal coupling relationship diagram using the closed-loop feedback dataset.

Images