Ice maker intelligent control method and system based on large model analysis

Abstract

The invention discloses an intelligent control method and system for an ice maker based on large model analysis, and relates to the technical field of intelligent control. In a to-be-defrosted monitoring stage, an ice layer re-melting change feature group is obtained through a sensor, after space standardization processing, a re-melting dense change interval is identified, historical ice making process data is utilized, and the ice layer re-melting dense change interval is identified; obtaining the theoretical defrosting heat energy input upper limit under the corresponding time condition, dynamically correcting the fitting ice layer thickness under the corresponding time condition, calculating the defrosting heat energy demand under the corresponding time condition after dynamic correction, and executing correction operation in combination with the theoretical defrosting heat energy input upper limit under the corresponding time condition. And finally, according to the change difference of the re-melting degree of the ice layer after each round of ice making of the ice maker, whether pre-risk exists in re-melting after the current round of ice making or not is recognized, a jump point is determined, a hot gas bypass starting node is determined according to the jump point, and defrosting operation is achieved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and in particular to an intelligent control method and system for an ice maker based on large model analysis. Background Art

[0002] This invention relates to the field of intelligent control technology, particularly the intelligent optimization branch of industrial thermal management and state perception. More specifically, it relates to intelligent control technology for ice-making equipment within refrigeration systems. Furthermore, it particularly relates to state monitoring, heat regulation, and predictive control during the defrosting process of ice-making equipment after ice making. During operation, an ice-making machine forms a solid ice mass that adheres tightly to the evaporator surface. To achieve subsequent cycles, the system must input heat to the evaporator at the appropriate time to smoothly release the ice from the mold surface.

[0003] In existing ice-making equipment, defrost operations are often triggered based on fixed time delays or empirical rules. This ignores the impact of complex factors such as external temperature differences and changes in melting trends during different ice-making cycles on defrosting behavior. This can easily lead to problems such as misjudgment of defrost timing, deviation in heat input, and demolding failure. For example, some ice-making machines defrost immediately after ice-making, heating according to a fixed time. This makes it difficult to detect whether the ice has begun to melt and whether to trigger hot air bypass in advance. This rigid control logic, when faced with large fluctuations in ice thickness and frequent environmental disturbances, can lead to problems such as premature defrosting resulting in incomplete ice solidification or delayed defrosting causing partial melting and damage to the ice layer, reducing system efficiency and increasing failure rates.

[0004] The cause of this problem is that existing ice-making control systems lack real-time sensing of meltback and adaptive feedback mechanisms for thermal control strategies. Since the meltback process often involves certain jumps, if the system cannot identify the period of intense meltback and intervene in advance, it can easily miss the optimal window for defrosting, causing the ice layer to adhere and refreeze on the evaporator surface. Furthermore, ice blocks melt and then condense due to local overheating, resulting in a three-phase adhesion of ice, water, and refreeze on the mold surface. This makes demolding difficult and may cause mold sticking, affecting the mechanical efficiency of the next round of demolding and increasing mechanical vibration, which can damage the demolding device over time. Summary of the Invention

[0005] In view of the deficiencies in the prior art, the present invention provides an ice-making machine intelligent control method and system based on large model analysis, which solves the problems in the above-mentioned background technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: an ice making machine intelligent control method based on large model analysis, comprising the following steps:

[0007] During the defrost monitoring phase, sensors are used to obtain ice layer melt change feature groups, and after spatial normalization, the melt intensive change intervals are identified.

[0008] Using historical ice-making process data, the theoretical upper limit of defrosting heat energy input under corresponding time conditions is obtained;

[0009] Dynamically correct the fitted ice thickness under the corresponding time conditions, calculate the defrost heat energy demand under the corresponding time conditions after dynamic correction, and perform correction operations based on the theoretical upper limit of defrost heat energy input under the corresponding time conditions;

[0010] Based on the difference in the degree of ice melting after each round of ice making, it is possible to identify whether there is a risk of premature melting after the current round of ice making, and determine the jump point. Based on the jump point, the hot gas bypass start node is determined to achieve defrosting operation.

[0011] Preferably, after the ice making cycle ends, the ice maker enters the defrost monitoring stage, and an infrared thermal sensor array, an acoustic wave array module, and a visual module are deployed in the near-field area of ​​the ice surface to obtain a set of ice layer melting change characteristics, including the ice layer surface temperature gradient distribution, thickness change, and the distribution map of the melting starting area;

[0012] By performing spatial normalization on various types of information in the ice melt change feature group, the standardized perception tensor after spatial normalization is obtained. Combined with the physical properties of the ice material, the ice mass is estimated and the mass time series is generated, specifically: M ice (t) = ρ ice *A*H ice (t), where M ice (t) is the mass of ice in the entire controlled area at time t, ρ ice is the ice density, A is the projected area of ​​the controlled area, H ice (t) is the thickness of the ice layer at time t;

[0013] The mass time series is converted into a melt mass change curve. According to the melt mass change curve, a fitting function is constructed. Based on the fitting function, the fitted ice thickness and melt rate at different times are obtained. The melt rate is calculated using the fitting derivative method.

[0014] Then, the second-order derivative of the ice mass is calculated to obtain the melting acceleration at different times, generate an acceleration sequence, and perform density estimation on the acceleration sequence to form a discriminant, which is specifically: Among them, D time is the number of time points that meet the conditions, T is the time window interval, t is the time number, a m (t) is the melting acceleration at time t, Th is the acceleration threshold, and 1[*] is the indicator function;

[0015] If the number of time points that meet the conditions exceeds Time window interval, the time point that meets the conditions will fall into the interval of intensive melting change.

[0016] Preferably, the change in ice thickness during the defrosting process of the evaporator, the change in the evaporator inner wall temperature, the historical water temperature sequence, and the spray fluctuation range are extracted from the historical ice making process data. Based on the historical water temperature sequence and the change in the evaporator inner wall temperature, the temperature difference gradient and the standard deviation of the spray intensity are calculated to obtain a feature set in the historical ice making data.

[0017] The regression model trained using the feature set of historical ice making data is used to obtain the upper limit of theoretical defrosting heat input, which is in the form of: Q def (t) = γ*H ice，actual (t)+λ*ΔW(t)+ξ*σ spr +∈, where Q def (t) is the upper limit of theoretical defrosting heat input at time t, H ice，actual (t) is the ice thickness collected at time t in the historical ice making data, σ spr is the standard deviation of the spray intensity, γ is the ice thickness influence coefficient, λ is the temperature difference response coefficient, ξ is the spray disturbance weight, and ∈ is the bias term;

[0018] The fitted ice thickness under the corresponding time conditions is substituted into the regression model, and the ice thickness influence coefficient, temperature difference response coefficient and spray disturbance weight are output by fitting using the least squares method to obtain the upper limit of the theoretical defrosting heat energy input under the corresponding time conditions.

[0019] Preferably, according to the melting intensive change interval, the actual melting rate and the measured ice thickness estimate in the corresponding time in the melting intensive change interval are extracted, the current thickness change speed is determined according to the actual melting rate, and the fitted ice thickness under the corresponding time conditions is dynamically corrected, specifically as follows: H ice (t)' is the correction value of the fitted ice thickness under the corresponding time conditions, R m (t) is the melting rate at time t, θ H (t) is the valuation deviation, is the melting correction coefficient, sgn(θ H (t)) is the sign function, Δt is the time step;

[0020] Among them, the valuation deviation is obtained as follows: θ H (t)=H(t)-H ice (t), where H(t) is the estimated ice thickness during the period of intensive melting. If H(t)≤H ice (t), in this case the sign function is plus, otherwise it is minus;

[0021] According to the fitting ice thickness under the corresponding time conditions after dynamic correction, the defrosting heat energy demand under the corresponding time conditions after dynamic correction is calculated, specifically: Q real (t) = ρ ice *A*H ice (t)'*L, where Q real (t) is the defrost heat energy requirement under the corresponding time conditions after dynamic correction, and L is the specific melting heat of unit volume of ice.

[0022] Preferably, the defrosting heat energy demand under the corresponding time conditions after dynamic correction is compared with the theoretical upper limit of defrosting heat energy input under the corresponding time conditions. real (t)≤Q def (t), indicating that the demand after correction is within the theoretical budget range, the ice maker is operating normally, and the current defrost control strategy continues to be executed without temporary intervention. If Q real (t)>Q def (t) indicates that the actual heat energy required by the ice maker has exceeded the upper limit of the budget. At this time, the slope of the thermal control curve in the current defrost control strategy will be corrected until the ice maker operates normally, and the correction operation will be stopped.

[0023] Preferably, according to the difference in the degree of ice melting after each round of ice making by the ice maker under corresponding time interval conditions, it is identified whether there is a risk of premature melting after the current round of ice making. The specific steps are:

[0024] Determine the melting intensive change interval corresponding to each round of ice making to construct a time series interval sequence;

[0025] Extract the lower limit of the time point in each interval of intensive change of the re-melting in the temporal interval sequence;

[0026] According to the time series interval sequence, the difference calculation of the lower limit of each time point in the intensive melting change interval is carried out in turn to obtain the point difference, and the deviation value of the corresponding point difference after the current round of ice making is calculated using the Z-Score standard deviation method;

[0027] If the deviation value of the corresponding spread after the current round of ice making is greater than 3, it is determined that there is a risk of early ice melting in the corresponding time window after the current round of ice making. Otherwise, there is no risk of early ice melting.

[0028] The time window interval corresponding to the risk of early refinancing is marked as the jumping point.

[0029] Preferably, the hot gas bypass start node is determined according to the jump point, and the specific steps are:

[0030] Traverse the melting intensive change intervals in different time windows after the current round of ice making to obtain the jump point set;

[0031] Extract the earliest time window interval corresponding to the trip point from the set of trip points and record it as the critical candidate interval for defrost triggering;

[0032] The lower limit of the time point in the critical candidate interval of defrost triggering is used as the hot gas bypass starting node to activate the hot gas bypass valve and realize the defrost operation.

[0033] An intelligent control system for ice making machine based on large model analysis, including:

[0034] The first recognition module, during the defrost monitoring phase, uses sensors to obtain ice layer melt change feature groups, and identifies the intervals of intensive melt changes after spatial normalization.

[0035] The first analysis module uses historical ice making process data to obtain the theoretical upper limit of defrosting heat energy input under corresponding time conditions;

[0036] The second analysis module dynamically corrects the fitted ice thickness under the corresponding time conditions, calculates the defrosting heat energy demand under the dynamically corrected corresponding time conditions, and performs correction operations based on the theoretical upper limit of defrosting heat energy input under the corresponding time conditions;

[0037] The second recognition module identifies whether there is a risk of premature melting after the current round of ice making based on the changes in the degree of ice melting after each round of ice making by the ice maker, and determines the jump point. Based on the jump point, the hot air bypass start node is determined to realize the defrost operation.

[0038] The present invention provides an ice-making machine intelligent control method and system based on large model analysis, which has the following beneficial effects:

[0039] (1) This method introduces a multimodal sensor in the defrost monitoring stage. By collecting the ice layer melting change feature group and performing spatial normalization processing, the melting intensive change interval is identified, and the ice layer melting mutation trend is effectively captured. The dynamic response to the defrost prediction is achieved, thereby avoiding abnormal phenomena such as demoulding failure and ice breakage caused by early or late defrosting. A bidirectional regulation mechanism based on thermal control boundaries and demand is constructed. By training a regression model using historical ice-making data, the theoretical upper limit of defrost heat input under corresponding time conditions is obtained. Dynamically correcting the upper limit based on the actual melt rate and fitted ice thickness, the defrost heat demand value is calculated to be closer to the actual operating state. This enables precise regulation of heat input, reduces system energy consumption, and avoids the problem of localized ice overmelting caused by heat input deviation. After each ice-making cycle, the system compares the changes in ice melt levels. Based on the sequence of dense melt change intervals and the jump point identification mechanism, the risk of premature melt is assessed. The activation timing of the hot gas bypass is dynamically adjusted accordingly to ensure that the defrost action is closely aligned with the ice release state, improving system operational stability and demolding completeness. In summary, this method, through in-depth analysis of the ice melt state and dynamic optimization of the thermal control path, not only improves the intelligence level of the defrost process, but also enhances the adaptability and reliability of the ice-making machine under different operating conditions. It is particularly suitable for ice-making equipment operating continuously for multiple cycles or in scenarios with drastic temperature and humidity fluctuations.

[0040] (2) The present invention constructs a heat consumption prediction function by using a regression model obtained through training using a historical feature set. It comprehensively considers factors such as ice thickness, temperature difference, and spray intensity disturbance, and quantifies the upper limit of theoretical defrosting heat energy input, so that the system can perform energy budgeting for ice-making cycles at different times and under different environmental conditions to prevent problems such as excessive heat input or insufficient energy.

[0041] (3) The present invention introduces a multi-round melting trend time series comparison mechanism, combined with the Z-Score standard deviation algorithm, to quantitatively analyze and identify jumps in the melting time point after each ice-making cycle, effectively enhancing the ice-making system's ability to judge the risk of early melting and dynamically control the timing of defrosting. Specifically, after each round of ice-making, the lower limit of the time point in the corresponding melting intensive change interval is extracted, and a time series interval sequence is constructed. The system can identify the difference in melting response behavior between adjacent ice-making rounds, quantify the rate of change of the melting start time point, and then analyze whether there is abnormal melting behavior that deviates from the normal trend in the current round, so as to identify the risk of early melting and realize the pre-perception of defrosting risk, further avoiding problems such as late defrosting caused by system inertia control or fixed time series defrosting logic, resulting in melting of the ice layer boundary structure, demolding failure, etc. For rounds that are judged to have the risk of early meltback, the system further selects the earliest critical candidate interval in the jump point set, and sets its lower time limit as the early trigger node for hot gas bypass, thereby realizing integrated closed-loop control based on trend changes, jump positioning and strategy adjustment, effectively improving the response sensitivity and strategy accuracy of the defrost thermal control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flow chart of an ice-making machine intelligent control method based on large model analysis according to the present invention;

[0043] Figure 2 This is a logic diagram of an ice maker intelligent control method based on large model analysis of the present invention;

[0044] Figure 3 This is a block diagram of an ice-making machine intelligent control system based on large model analysis in the present invention. DETAILED DESCRIPTION

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0046] Example 1

[0047] See also Figure 1 and Figure 2 The present invention provides an ice making machine intelligent control method based on large model analysis, comprising the following steps:

[0048] During the defrost monitoring phase, sensors are used to obtain ice layer melt change feature groups, and after spatial normalization, the melt intensive change intervals are identified.

[0049] Using historical ice-making process data, the theoretical upper limit of defrosting heat energy input under corresponding time conditions is obtained;

[0050] Dynamically correct the fitted ice thickness under the corresponding time conditions, calculate the defrost heat energy demand under the corresponding time conditions after dynamic correction, and perform correction operations based on the theoretical upper limit of defrost heat energy input under the corresponding time conditions;

[0051] Based on the difference in the degree of ice melting after each round of ice making, it is possible to identify whether there is a risk of premature melting after the current round of ice making, and determine the jump point. Based on the jump point, the hot gas bypass start node is determined to achieve defrosting operation.

[0052] The large-scale model analysis includes a multi-factor state mapping network built based on historical operating data, a time series trend prediction module and a thermal control strategy optimization engine, which supports high-dimensional dynamic reasoning and control recommendation generation based on current melting behavior, ice layer status and external environmental disturbances.

[0053] In this embodiment, through the deep integration of multi-dimensional perception, melting trend modeling, heat consumption prediction and demolding control, an intelligent transition of thermal control strategy from static rule-driven to data prediction-driven is achieved.

[0054] Specifically, the present invention no longer uses a fixed time strategy to trigger the defrost operation. Instead, during the defrost monitoring phase, it integrates multimodal sensor information such as infrared thermal sensing, acoustic wave reflection, and visual images to obtain a set of ice layer melt change characteristics and extract them. By performing spatial normalization on these characteristics, a dynamic perception tensor at a unified scale is constructed, thereby identifying intervals of intensive melt changes, further realizing real-time recognition of ice layer melt trends and identifying abnormal ice melt fluctuations.

[0055] The system leverages historical data on temperature gradients, water temperature fluctuations, and spray intensity fluctuations during ice-making, combined with the current melt rate and fitted ice thickness, to dynamically estimate the theoretical upper limit of defrost heat input using a regression model. The fitted ice thickness is then dynamically corrected to accurately reflect the ice layer's state, thereby inferring a more realistic defrost heat demand. For example, if the ambient temperature suddenly changes and the ice layer thickens locally, the system dynamically adjusts the heat input slope based on the corrected ice thickness and real-time melt rate, preventing ice surface collapse caused by overheating and improving the efficiency of each ice-making and demolding cycle.

[0056] During multiple rounds of ice making, the system compares the difference between the lower limits of the time points in each round of intensive melting changes, and uses the Z-Score standard deviation method to determine whether there is a risk of premature melting in the current ice making round. Once the risk jump point is identified, it will be used as a candidate hot air bypass start node to adjust the defrost action in advance to avoid local damage to the ice body caused by delayed thermal control response. For example, in a certain ice making round, due to abnormal external temperature control, the ice body tends to melt prematurely. The system detects a point difference deviation value of >3, immediately determines the jump point, and starts the hot air bypass 1.5 minutes in advance, effectively suppressing the problem of failure to debond the bottom of the ice body due to delayed defrost, and further avoiding the phenomenon of ice-water-refrozen three-phase adhesion on the mold surface.

[0057] During each round of defrosting, the system will record the deviation between the corrected thermal energy demand and the theoretical thermal energy upper limit. If the budget limit is exceeded, the current thermal control curve slope and ice thickness fitting parameters will be automatically corrected, gradually realizing self-correction of the model and self-optimization of the control strategy, forming a learnable and evolvable intelligent defrosting closed-loop logic.

[0058] In summary, the present invention further realizes the advantages of high timeliness, high energy efficiency and high stability of ice maker defrost control through multiple intelligent thermal control chains of perception, prediction, judgment, control and correction. It shows strong adaptability under multiple rounds of ice making and complex environmental fluctuations, and provides innovative solutions for energy-saving operation and automatic control of ice making systems.

[0059] Example 2

[0060] Please refer to Figure 1 Specifically: After the ice-making cycle ends, the ice-making machine enters the defrost monitoring phase. By deploying an infrared thermal sensor array, an acoustic array module, and a visual module in the near-field area of ​​the ice surface, a set of ice layer melting change characteristics is obtained, including the ice surface temperature gradient distribution, thickness changes, and the distribution map of the melting starting area.

[0061] The near-field area of ​​the ice surface refers to a limited space area close to the ice layer surface, heat exchange interface or evaporator mold boundary during the ice-making process or after the ice-making cycle. Its spatial range is usually a few millimeters to a few centimeters. It is mainly used for fine perception of changes in ice layer state, heat exchange behavior or initial melting characteristics.

[0062] After the ice-making cycle is completed, the ice-making machine enters the defrost monitoring stage, which is a key state identification window before the thermal control behavior is started.

[0063] The infrared thermal sensor array is used to obtain the temperature gradient distribution on the ice surface. The acoustic wave array module will obtain the position fluctuations of the ice-water interface based on the changes in ultrasonic reflection time to extract thickness change data. The visual module will be used to monitor changes in surface color and glossiness, identify water film formation areas, and form a distribution map of the melting start area. Among them, the distribution map of the melting start area refers to: a two-dimensional distribution matrix generated based on the analysis of the surface visual sensor (or infrared thermal image), which is a graphical expression of which areas on the ice surface have or are about to show signs of melting.

[0064] When the ice layer has just finished freezing, some areas may have begun to melt naturally due to uneven heat transfer or water flow disturbances, especially near the evaporator surface or the edge of the ice. At this time, the hot gas defrost stage has not yet begun, but tiny temperature gradients, changes in acoustic wave impedance and surface reflectivity have begun to form inside the ice layer. These differences are the key features for determining whether to defrost in advance and adjust the thermal control amount.

[0065] By performing spatial normalization on various types of information in the ice melt change feature group, unifying the scale based on ice mass estimation, obtaining the standardized perception tensor after spatial normalization, and combining it with the material physical properties of ice to estimate the ice mass, a mass time series is generated, specifically: M ice (t) = μ ice *A*H ice (t), where M ice (t) is the mass of ice in the entire controlled area at time t, ρ ice is the ice density, A is the projected area of ​​the controlled area, H ice (t) is the thickness of the ice layer at time t, that is, the average thickness of the ice layer in the vertical direction;

[0066] The material physical properties of ice include ice density;

[0067] The projected area of ​​the controlled area refers to the effective area on the horizontal plane of the area controlled or sensed by the system during the ice making process.

[0068] The above-mentioned method of obtaining the mass of the ice layer is to approximately obtain it by multiplying the density, bottom area and thickness of the ice layer. This actually comes from the basic expression formula of mass in physics: mass = density × volume, and volume can be expressed as: volume = bottom area × thickness; its function is to allow the perception data of multimodal sensors (mainly to obtain thickness) to be converted into actual physical mass, so that the subsequent heat loss modeling and melting rate prediction have physical real semantic support.

[0069] Spatial normalization is to unify the physical reference scale and eliminate the influence of spatial distribution, so as to eliminate dimensional inconsistency and perceptual offset and form a unified description of the ice state;

[0070] The mass time series is converted into a melt mass change curve. According to the melt mass change curve, a fitting function is constructed. Based on the fitting function, the fitted ice thickness and melt rate at different times are obtained. The melt rate is calculated using the fitting derivative method, specifically: R m (t) is the melting rate at time t, dM ice (t) is the small change in ice mass, and dt is the small change in time;

[0071] The fitting function refers to performing polynomial regression, spline interpolation or Gaussian kernel fitting on the ice mass time series to transform the original discrete point sequence into a continuous, differentiable and predictable mass function curve;

[0072] Then, the second-order derivative of the ice mass is calculated to obtain the melting acceleration at different times, generate an acceleration sequence, and perform density estimation on the acceleration sequence to form a discriminant, which is specifically: Among them, D time is the number of time points that meet the conditions, T is the time window interval, t is the time number, a m (t) is the melting acceleration at time t, Th is the acceleration threshold, 1[*] is the indicator function, which is 1 if the logic in the brackets is true, otherwise it is 0;

[0073] The time point when the conditions are met here refers to the time when a m (t) time point ≥Th;

[0074] Acceleration is an important indicator for determining abnormal or sudden intensification of remelting. When acceleration is positive and rising rapidly, it means that the system may be experiencing uncontrolled accumulation of heat conduction and requires a timely response control strategy. If this indicator is ignored, the system will only determine the defrost timing based on a fixed time, which will lead to lag problems and make it difficult to perceive the risk of sudden changes in advance.

[0075] Among them, when the melting acceleration at time t is greater than or equal to the acceleration threshold, it is used to mark time t as a high-risk concentrated change area;

[0076] The acceleration sequence includes the melt acceleration at different moments;

[0077] If the number of time points that meet the conditions exceeds If the time window interval is set, the time points that meet the conditions will fall into the interval of dense re-melting change to form a time density marker, otherwise they will not fall into the interval of dense re-melting change.

[0078] The time density marker is a set of time windows extracted based on the melting acceleration fluctuation density, which is used to calibrate high-risk melting segments and constitute an early warning mechanism for defrost control.

[0079] The intensive melting change interval is a time window within which the system detects continuous and drastic changes in the melting rate of the ice layer, which is likely to represent a critical period when melting becomes active, the melting trend changes suddenly, or the risk of premature melting increases. It is used to capture areas where concentrated outbreaks of melting behavior occur within a certain period of time.

[0080] In this embodiment, after the ice-making cycle ends, the system does not immediately defrost. Instead, it enters a defrost monitoring phase. Using an infrared thermal sensor array, acoustic array module, and visual module deployed near the ice surface, the system constructs a signature set of ice melt changes, including surface temperature gradients, thickness variations, and a distribution map of the melt initiation zone. For example, if the temperature at the edge of the ice suddenly rises, the reflectivity in the visual image dims, and the acoustic reflection delay exhibits an abnormal shift, the system can initially identify the beginning of ice melt. This step eliminates dimensional differences in physical quantities across different data sources, forming a unified dynamic signature trajectory that represents changes in the ice state.

[0081] By correlating the spatially normalized sensor tensor with material properties such as ice density, a time series of ice mass is constructed. This series is then fitted using a fitting function. Furthermore, a fitted derivative method is introduced to calculate the melt rate and melt acceleration. This overcomes the noise susceptibility of direct differential calculations, allowing for better capture of sudden changes in trends and fluctuations during the melt process. For example, during the actual ice-making process, if the melt rate curve shows a continuous and steep increase within a certain period of time, and the acceleration exceeds a preset threshold, it can be inferred that the ice is in the initial stages of large-scale rapid melt, and defrost control measures must be prepared in advance.

[0082] Based on acceleration sequence analysis, this invention introduces a time density estimation mechanism. By counting the number of time points within a time window that are above a threshold, it identifies intervals of dense meltback changes. This interval represents the critical behavioral segment where the risk of meltback begins to increase in the system. This provides a time reference anchor point for subsequent decisions such as defrost trigger node determination and thermal control input curve correction. For example, if more than 70% of the acceleration data in a continuous time window is above the threshold, it indicates that the ice layer is in a state of intense dynamic meltback. This time period can be used as a candidate for the defrost judgment trigger critical point to avoid delayed system response.

[0083] Therefore, the present invention overcomes the shortcomings of fixed defrosting action time and lack of trend judgment control in traditional ice-making equipment, and further realizes a continuous closed-loop response mechanism of multi-source data fusion, dynamic derivative judgment, intelligent time window identification and adaptive defrost control, providing support for improving ice-making integrity and reducing heat loss.

[0084] Example 3

[0085] Please refer to Figure 1Specifically, the ice thickness change of the evaporator during the defrosting process, the evaporator inner wall temperature change, the water temperature history sequence, and the spray fluctuation range are extracted from the historical ice making process data. Based on the water temperature history sequence and the evaporator inner wall temperature change, the temperature difference gradient and the spray intensity standard deviation are calculated to obtain the feature set in the historical ice making data. Specifically, ΔW(t) = W wall (t)-W wtr (t), where ΔW(t) is the temperature gradient, W wall (t) is the inner wall temperature of the evaporator, W wtr (t) is the water temperature history sequence; the temperature gradient is used to represent the heat transfer driving force during the defrosting process, reflecting the heat exchange capacity between the evaporator wall and the ice layer, and directly determines the time and heat input required for defrosting;

[0086] Historical ice-making process data refers to the continuous and structured set of operating parameters recorded and collected by the ice-making machine in multiple complete operating cycles. These data are the core foundation for the system to build key modules such as intelligent prediction models, regression functions, and correction mechanisms. They include but are not limited to changes in ice thickness during the defrosting process of the evaporator, changes in the evaporator inner wall temperature, historical water temperature sequences, and spray fluctuation ranges.

[0087] The higher the spray volatility, the more complex the ice crystal structure. Therefore, the spray intensity standard deviation reflects the local complexity of the ice layer structure and the stability of defrost energy consumption. Obtaining the temperature gradient and spray intensity standard deviation is a balance between model stability, heat consumption representativeness, and control feedback efficiency, providing a solid foundation for theoretical defrost heat energy estimation.

[0088] The water temperature history series is used to record the basic thermal background of the melt, and the spray fluctuation range is used to reflect the uneven heat conduction caused by external water flow disturbances.

[0089] The regression model trained using the feature set of historical ice making data is used to obtain the upper limit of theoretical defrosting heat input, which is in the form of: Q def (t) = γ*H ice，actual (t)+λ*ΔW(t)+ξ*σ spr +∈, where Q def (t) is the upper limit of theoretical defrosting heat input at time t, H ice，actual (t) is the ice thickness collected at time t in the historical ice making data, σ spris the standard deviation of the spray intensity, which indicates the degree of fluctuation of the spray water flow and is used to estimate the complexity of the ice layer structure. γ is the ice thickness influence coefficient, which is the linear coefficient of the effect of ice thickness on heat consumption. λ is the temperature difference response coefficient, which is the effect of each 1°C increase in temperature difference on heat consumption. ξ is the spray disturbance weight, which describes the weight that the stronger the spray fluctuation, the greater the need for system thermal control compensation. ∈ is the bias term, which is used to compensate for the energy baseline in the system that is difficult to quantify, such as equipment consumption and startup loss.

[0090] A regression model is a mathematical model used to predict numerical outcome variables. It establishes a functional relationship between input variables (independent variables) and output variables (dependent variables) to achieve the ability to infer target quantities from known states.

[0091] The theoretical defrost heat energy is equivalent to the heat demand in the heat demand forecast, and it is a key value output by the forecast module.

[0092] Defrost heat energy refers to the total amount of heat input required to achieve demoulding or ice removal;

[0093] The fitted ice thickness under the corresponding time conditions is substituted into the regression model, and the ice thickness influence coefficient, temperature difference response coefficient and spray disturbance weight are output by fitting using the least squares method to obtain the upper limit of the theoretical defrosting heat energy input under the corresponding time conditions.

[0094] In this embodiment, by systematically extracting the changes in ice layer thickness, evaporator inner wall temperature, water temperature change sequence and spray fluctuation parameters during the historical ice-making process, a historical feature set containing multiple factors such as temperature difference gradient and spray intensity standard deviation is formed. The system can learn the pattern of ice layer heat consumption behavior under different operating environments. The temperature difference gradient reflects the driving strength of heat exchange; the spray intensity standard deviation is used to measure the uniformity of the ice layer affected by disturbances during formation; and the ice layer thickness sequence is used to determine the ultimate amount of heat energy required to complete demolding.

[0095] The regression model uses the least squares method to learn coefficients, ensuring that the output reflects the actual defrost heat demand at different times, improving the consistency between predictions and actual system performance. Because the model dynamically inputs the ice thickness value fitted to the current cycle and makes corrections based on prior historical patterns, it further enhances the control system's environmental adaptability.

[0096] In summary, this method achieves dynamic and accurate estimation of the theoretical heat consumption budget by integrating historical operating data with current ice-making conditions, providing a reliable and self-evolving energy efficiency boundary support basis for subsequent defrost control strategies such as energy consumption boundary, thermal control curve slope, and adaptive start timing, and demonstrating higher control flexibility and system stability under multiple rounds and complex working conditions.

[0097] Example 4

[0098] Please refer to Figure 1 Specifically, according to the melting intensive change interval, the actual melting rate and the measured ice thickness estimate in the corresponding time of the melting intensive change interval are extracted. The current thickness change speed is determined according to the actual melting rate, and the fitted ice thickness under the corresponding time conditions is dynamically corrected. Specifically, H ice (t)' is the correction value of the fitted ice thickness under the corresponding time conditions, R m (t) is the melting rate at time t, θ H (t) is the valuation deviation, is the melting correction coefficient, which indicates the thickness compensation ratio corresponding to the unit melting rate change, and is derived from the empirical model or data fitting. H (t)) is the sign function, Δt is the time step, i.e., the acquisition interval;

[0099] Among them, the valuation deviation is obtained as follows: θ H (t)=H(t)-H ice (t), where H(t) is the estimated ice thickness during the period of intensive melting. If H(t)≤H ice (t) indicates that the system is underestimated and an additive correction is performed. In this case, the sign function is plus. Otherwise, it indicates that the system is underestimated and an additive correction is performed. The sign function is negative.

[0100] According to the fitting ice thickness under the corresponding time conditions after dynamic correction, the defrosting heat energy demand under the corresponding time conditions after dynamic correction is calculated, specifically: Q real (t) = ρ ice *A*H ice (t)'*L, where Q real (t) is the defrost heat demand under the corresponding time conditions after dynamic correction, L is the specific heat of fusion of unit volume of ice, which represents the energy required to completely melt 1 kg of ice into water starting from 0℃.

[0101] Compare the defrosting heat energy demand under the corresponding time conditions after dynamic correction with the theoretical defrosting heat energy input upper limit under the corresponding time conditions. If Q real (t)≤Q def (t), indicating that the demand after correction is within the theoretical budget range, the ice maker is operating normally, and the current defrost control strategy continues to be executed without temporary intervention. If Q real (t)>q def (t) indicates that the actual heat energy required by the ice maker has exceeded the upper limit of the budget. At this time, the slope of the thermal control curve in the current defrost control strategy will be corrected until the ice maker operates normally, and the correction operation will be stopped.

[0102] Among them, correcting the slope of the thermal control curve in the current defrost control strategy refers to increasing the instantaneous heat flow rate;

[0103] The ice thickness estimation is corrected so that the mapping relationship between ice thickness and heat consumption can more accurately determine the energy demand gradient at that time.

[0104] Dynamic correction of the fitted ice thickness under corresponding time conditions is used to improve the dynamic response accuracy of heat loss boundary prediction and suppress the problem of excessive heat input caused by early erroneous ice thickness estimation.

[0105] In this embodiment, the present invention introduces a mechanism for identifying periods of high melt rate fluctuations, combined with a dynamic ice thickness correction function and a heat consumption demand comparison mechanism, to construct an intelligent defrost energy consumption response process tailored to sudden melt events. Specifically, the present invention dynamically matches the real-time melt rate collected during periods of high melt rate fluctuations with the measured ice thickness estimate to implement an ice thickness correction mechanism based on dynamic melt feedback. This establishes a differential adjustment mechanism between the three elements of time, heat, and structure, thereby suppressing the cumulative amplification of ice thickness fitting errors over consecutive time periods.

[0106] The corrected ice thickness data will be used as input to the heat energy demand calculation formula to obtain the dynamic heat consumption demand that truly matches the current physical state and compare the difference with the theoretical heat energy input upper limit generated by the previous historical model. real (t)>Q def (t) triggers a temporary correction program to dynamically increase the instantaneous heat flow slope of the thermal control curve to avoid defrosting delays caused by insufficient energy, and further ensure that the demoulding action is performed under the optimal temperature control state.

[0107] Example 5

[0108] Please refer to Figure 1 Specifically: Based on the difference in the degree of ice melting after each round of ice making, the ice maker can identify whether there is a risk of premature melting after the current round of ice making. The specific steps are as follows:

[0109] Determine the melting intensive change interval corresponding to each round of ice making to construct a time series interval sequence;

[0110] Extract the lower limit of the time point in each intensive melting interval in the temporal interval sequence; for example, one of the intensive melting intervals is [t a , t b ], where t a is the lower limit of the time point of the intensive melting interval, t b The upper limit of the time point of the intensive melting range;

[0111] According to the time series interval sequence, the difference calculation of the lower limit of each time point in the intensive melting change interval is carried out in turn to obtain the point difference, and the deviation value of the corresponding point difference after the current round of ice making is calculated using the Z-Score standard deviation method;

[0112] If the deviation value of the corresponding spread after the current round of ice making is greater than 3, it is determined that there is a risk of early ice melting in the corresponding time window after the current round of ice making. Otherwise, there is no risk of early ice melting.

[0113] The time window interval corresponding to the risk of early refinancing is marked as the jumping point.

[0114] The calculation method of the spread is: Dc(r)=|t a (r)-t a (r-1)|, where Dc(r) is the difference after the rth round of ice making, t a (r) and t a (r-1) are the lower limits of the time points in the intensive melting interval corresponding to the rth round of ice making and the r-1th round of ice making, respectively, and r is the number of the ice making round;

[0115] The Z-Score standard deviation method is a standardized indicator that measures how far a sample point is from the population mean;

[0116] The calculation method of the deviation value of the corresponding spread after the corresponding round of ice making is: Among them, Z(r) is the deviation value of the corresponding point difference after the rth round of ice making, Dc avg is the mean spread, Dc σ is the standard deviation of the spread.

[0117] Determine the hot gas bypass start-up node based on the trip point. The specific steps are as follows:

[0118] Traverse the melting intensive change intervals in different time windows after the current round of ice making to obtain the jump point set;

[0119] Extract the earliest time window interval corresponding to the trip point from the set of trip points and record it as the critical candidate interval for defrost triggering;

[0120] The lower limit of the time point in the critical candidate interval of defrost triggering is used as the starting node of the hot gas bypass to activate the hot gas bypass valve and realize the defrost operation.

[0121] Hot gas bypass is a thermal control structure in the refrigerant system that uses a control valve to bypass the condenser and directly introduce high-temperature and high-pressure gaseous refrigerant into the evaporator. Its purpose is to heat the evaporator, reverse the direction of the refrigerant and achieve rapid defrosting.

[0122] In this embodiment, the present invention realizes adaptive judgment and response to the early defrost timing in multiple rounds of continuous operation scenarios by introducing a combination mechanism of melt-back intensive change interval sequence analysis, Z-Score fluctuation anomaly detection and jump point driven defrost trigger strategy in the ice maker operation cycle.

[0123] The present invention first extracts the intervals of intense melting after each round of icemaking, establishing a temporal sequence of intervals after multiple rounds of icemaking. This sequence records the start time of ice melt in each round of defrosting, forming a thermal behavior time spectrum with historical depth, enabling the identification of evolving trends. For example, if the start time of melt in previous rounds was approximately three minutes after the end of icemaking, but in one round it was advanced to two minutes, this discrepancy would be explicitly identified through subsequent analysis.

[0124] By calculating the difference between the lower limits of these time points, the system determines the time offset for the start of re-thaw after each round of ice making. The system then uses the Z-Score standard deviation method to normalize the difference for the current round. If the deviation is greater than 3 (i.e., more than 3 standard deviations from the mean), the system automatically identifies the risk of premature re-thaw, thus forming a judgment logic for abnormal sudden changes. This judgment mechanism avoids missing the defrost window due to a single sudden change going undetected, and is particularly useful in scenarios with drastic ambient temperature fluctuations or reduced equipment heat dissipation capacity.

[0125] After detecting the time window where there is a risk of early meltback, the system extracts the area with the earliest time point from all the jump points as the candidate critical interval for defrost triggering, and uses the lower limit time of this interval as the hot gas bypass starting node. This approach ensures that the defrost operation can not only respond to the risk of mutation in a timely manner, but also achieve the earliest accurate triggering under the time granularity, avoiding damage to the ice mold structure or failure of ice adhesion caused by excessive meltback diffusion. In summary, the present invention effectively improves the stability and energy efficiency adaptability of the ice-making system under complex multi-round operation by constructing a defrost start decision mechanism with historical evolution tracking, dynamic deviation identification and node precision control functions. It is particularly suitable for intelligent ice-making equipment that needs to work continuously or is frequently disturbed by the environment.

[0126] Example 6

[0127] Please refer to Figure 3 ,Specifically: An intelligent control system for ice making machine based on large model analysis, including,

[0128] The first recognition module, during the defrost monitoring phase, uses sensors to obtain ice layer melt change feature groups, and identifies the intervals of intensive melt changes after spatial normalization.

[0129] The first analysis module uses historical ice making process data to obtain the theoretical upper limit of defrosting heat energy input under corresponding time conditions;

[0130] The second analysis module dynamically corrects the fitted ice thickness under the corresponding time conditions, calculates the defrosting heat energy demand under the dynamically corrected corresponding time conditions, and performs correction operations based on the theoretical upper limit of defrosting heat energy input under the corresponding time conditions;

[0131] The second recognition module identifies whether there is a risk of premature melting after the current round of ice making based on the changes in the degree of ice melting after each round of ice making by the ice maker, and determines the jump point. Based on the jump point, the hot air bypass start node is determined to realize the defrost operation.

[0132] Some of the data in the above formulas are dimensionless and numerically calculated. Meanwhile, the contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0133] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent control method for an ice maker based on large model analysis, characterized by: The following steps are included: During the defrost monitoring phase, sensors are used to obtain ice layer melt change feature groups, and after spatial normalization, the melt-intensive change intervals are identified. Using historical ice-making process data, the theoretical upper limit of defrosting heat energy input under corresponding time conditions is obtained; Dynamically correct the fitted ice thickness under the corresponding time conditions, calculate the defrost heat energy demand under the corresponding time conditions after dynamic correction, and perform correction operations based on the theoretical upper limit of defrost heat energy input under the corresponding time conditions; Based on the difference in the degree of ice melting after each round of ice making, it is possible to identify whether there is a risk of premature melting after the current round of ice making, and determine the jump point. Based on the jump point, the hot gas bypass start node is determined to achieve defrosting operation.

2. The ice-making machine intelligent control method based on large model analysis according to claim 1 is characterized in that: After the ice-making cycle ends, the ice-making machine enters the defrost monitoring phase. By deploying an infrared thermal sensor array, an acoustic array module, and a visual module in the near-field area of ​​the ice surface, the system acquires a set of ice layer melting change characteristics, including the ice surface temperature gradient distribution, thickness changes, and the distribution map of the melting starting area. By performing spatial normalization on various types of information in the ice melt change feature group, the standardized perception tensor after spatial normalization is obtained. Combined with the physical properties of the ice material, the ice mass is estimated and the mass time series is generated, specifically: M ice (t) = ρ ice *A*H ice (t), where M ice (t) is the mass of ice in the entire controlled area at time t, ρ ice is the ice density, A is the projected area of ​​the controlled area, H ice (t) is the thickness of the ice layer at time t; The mass time series is converted into a melt mass change curve. According to the melt mass change curve, a fitting function is constructed. Based on the fitting function, the fitted ice thickness and melt rate at different times are obtained. The melt rate is calculated using the fitting derivative method. Then, the second-order derivative of the ice mass is calculated to obtain the melting acceleration at different times, generate an acceleration sequence, and perform density estimation on the acceleration sequence to form a discriminant, which is specifically: Among them, D time is the number of time points that meet the conditions, T is the time window interval, t is the time number, a m (t) is the melting acceleration at time t, Th is the acceleration threshold, and 1[*] is the indicator function; If the number of time points that meet the conditions exceeds Time window interval, the time point that meets the conditions will fall into the interval of intensive melting change.

3. The ice-making machine intelligent control method based on large model analysis according to claim 2 is characterized in that: The ice thickness change, evaporator inner wall temperature change, water temperature history sequence, and spray fluctuation range of the evaporator during defrosting are extracted from the historical ice-making process data. Based on the water temperature history sequence and evaporator inner wall temperature change, the temperature difference gradient and spray intensity standard deviation are calculated to obtain the feature set of the historical ice-making data. The regression model trained using the feature set of historical ice making data is used to obtain the upper limit of theoretical defrosting heat input, which is in the form of: Q def (t) = γ*H ice，actual (t)+λ*ΔW(t)+ξ*σ spr +∈, where Q def (t) is the upper limit of theoretical defrosting heat input at time t, H ice，actual (t) is the ice thickness collected at time t in the historical ice making data, σ spr is the standard deviation of the spray intensity, γ is the ice thickness influence coefficient, λ is the temperature difference response coefficient, ξ is the spray disturbance weight, and ∈ is the bias term; The fitted ice thickness under the corresponding time conditions is substituted into the regression model, and the ice thickness influence coefficient, temperature difference response coefficient and spray disturbance weight are output by fitting using the least squares method to obtain the upper limit of the theoretical defrosting heat energy input under the corresponding time conditions.

4. The ice-making machine intelligent control method based on large model analysis according to claim 3 is characterized in that: According to the melting intensive change interval, the actual melting rate and the estimated ice thickness in the corresponding time period of the melting intensive change interval are extracted. The current thickness change speed is determined based on the actual melting rate, and the fitted ice thickness under the corresponding time conditions is dynamically corrected. Specifically, H ice (t)' is the correction value of the fitted ice thickness under the corresponding time conditions, R m (t) is the melting rate at time t, θ H (t) is the valuation deviation, is the melting correction coefficient, sgn(θ H (t)) is the sign function, Δt is the time step; Among them, the valuation deviation is obtained as follows: θ H (t)=H(t)-H ice (t), where H(t) is the estimated ice thickness during the period of intensive melting. If H(t)≤H ice (t), in this case the sign function is plus, otherwise it is minus; According to the fitting ice thickness under the corresponding time conditions after dynamic correction, the defrosting heat energy demand under the corresponding time conditions after dynamic correction is calculated, specifically: Q real (t) = ρ ice *A*H ice (t) ‘ *L, where Q real (t) is the defrosting heat energy requirement under the corresponding time conditions after dynamic correction, and L is the specific melting heat of unit volume of ice.

5. The ice-making machine intelligent control method based on large model analysis according to claim 4 is characterized in that: Compare the defrosting heat energy demand under the corresponding time conditions after dynamic correction with the theoretical defrosting heat energy input upper limit under the corresponding time conditions. If Q real (t)≤Q def (t), indicating that the demand after correction is within the theoretical budget range, the ice maker is operating normally, and the current defrost control strategy continues to be executed without temporary intervention. If Q real (t)>Q def (t) indicates that the actual heat energy required by the ice maker has exceeded the upper limit of the budget. At this time, the slope of the thermal control curve in the current defrost control strategy will be corrected until the ice maker operates normally, and the correction operation will be stopped.

6. The ice-making machine intelligent control method based on large model analysis according to claim 5, characterized in that: Based on the difference in the degree of ice melting after each round of ice making, the ice maker can identify whether there is a risk of premature melting after the current round of ice making. The specific steps are as follows: Determine the melting intensive change interval corresponding to each round of ice making to construct a time series interval sequence; Extract the lower limit of the time point in each interval of intensive melting change in the temporal interval sequence; According to the time series interval sequence, the difference calculation of the lower limit of each time point in the intensive melting change interval is carried out in turn to obtain the point difference, and the deviation value of the corresponding point difference after the current round of ice making is calculated using the Z-Score standard deviation method; If the deviation value of the corresponding spread after the current round of ice making is greater than 3, it is determined that there is a risk of early ice melting in the corresponding time window after the current round of ice making; otherwise, there is no risk of early ice melting. The time window interval corresponding to the risk of early refinancing is marked as the jumping point.

7. The ice-making machine intelligent control method based on large model analysis according to claim 6, characterized in that: Determine the hot gas bypass start-up node based on the trip point. The specific steps are as follows: Traverse the melting intensive change intervals in different time windows after the current round of ice making to obtain the jump point set; Extract the earliest time window interval corresponding to the jump point from the jump point set and record it as the critical candidate interval for defrost triggering; The lower limit of the time point in the critical candidate interval of defrost triggering is used as the hot gas bypass starting node to activate the hot gas bypass valve and realize the defrost operation.

8. An intelligent control system for an ice maker based on large model analysis, used to implement the intelligent control method for an ice maker based on large model analysis as described in any one of claims 1 to 7, characterized in that: include, The first recognition module, during the defrost monitoring phase, uses sensors to obtain ice layer melt change feature groups, and identifies the intervals of intensive melt changes after spatial normalization. The first analysis module uses historical ice making process data to obtain the theoretical upper limit of defrosting heat energy input under corresponding time conditions; The second analysis module dynamically corrects the fitted ice thickness under the corresponding time conditions, calculates the defrosting heat energy demand under the dynamically corrected corresponding time conditions, and performs correction operations based on the theoretical upper limit of defrosting heat energy input under the corresponding time conditions; The second recognition module identifies whether there is a risk of premature melting after the current round of ice making based on the changes in the degree of ice melting after each round of ice making by the ice maker, and determines the jump point. Based on the jump point, the hot air bypass start node is determined to realize the defrost operation.

Images