An AI-based large-scale model-based optimization method and system for industrial waste heat storage scheduling

Abstract

This invention discloses an industrial waste heat storage scheduling optimization method and system based on an AI large-scale model. The method involves collecting multi-source waste heat data and storage medium operation data, generating a waste heat state parameter set through time-series synchronization and thermal parameter analysis; extracting grade critical drop characteristics and storage capacity matching factors to construct a charging start-up judgment table; inputting the waste heat state parameter set into the AI ​​large-scale model to identify multi-source heat flow superposition characteristics, generating charging coordination parameters, and dynamically adjusting the charging start-up judgment table to generate storage allocation parameters; performing heat load time-series verification on the storage allocation parameters to identify storage capacity boundaries, extracting waste heat supply confidence intervals to generate charging time window locking parameters, verifying the heat balance window to generate a storage scheduling sequence; finally, determining the priority weight of heat energy allocation, and outputting waste heat storage execution instructions based on the storage scheduling sequence to achieve adaptive scheduling optimization of industrial waste heat storage.

Description

Technical Field

[0001] This invention relates to the field of industrial energy conservation and thermal energy management technology, and in particular to an industrial waste heat storage scheduling optimization method and system based on an AI large model. Background Technology

[0002] Process industries such as steel, chemicals, and building materials continuously generate a large amount of waste heat resources during production. The grade of waste heat changes dynamically with the process progress. When multiple processes operate in parallel, the waste heat output of each node overlaps on the time axis. The timing of the charging of the heat storage device and the capacity allocation directly affect the waste heat recovery efficiency and the continuity of heating. However, the heat flux density of multi-source waste heat nodes exhibits a non-linear decay as the grade decreases. The times when different heat sources trigger the charging conditions frequently overlap, leading to channel competition. The coordination constraint between the sudden increase in heat-use load and the intermittency of waste heat supply is difficult to describe uniformly through fixed parameter rules. Existing scheduling methods generally rely on manual experience to set thresholds and sorting rules, resulting in problems such as missing high-grade waste heat windows, lagging coordination of charging channel competition, and static solidification of heat storage capacity boundaries in multi-heat-source dynamic interaction scenarios.

[0003] Therefore, a method is urgently needed to solve at least one of the above problems. Summary of the Invention

[0004] This invention discloses an industrial waste heat storage scheduling optimization method and system based on an AI large model. It aims to integrate multi-source thermal data, construct an adaptive charging judgment basis by extracting heat source grade drop characteristics and analyzing heat flux density changes, identify time-series conflicts of multiple heat sources and complete charging coordination with the help of an AI large model, dynamically determine the heat storage capacity boundary and charging time window by combining heat load time sequence verification and waste heat supply confidence assessment, and finally output waste heat storage execution instructions based on the priority weight of heat energy allocation, so as to realize adaptive and refined management of industrial waste heat storage scheduling.

[0005] The first aspect of this invention proposes an industrial waste heat storage scheduling optimization method based on an AI large model, comprising the following steps: Collect multi-source waste heat data and heat storage medium operation data of industrial processes, and perform time-series synchronization and thermal parameter analysis on the multi-source waste heat data and the heat storage medium operation data to generate a waste heat state parameter set. Based on the waste heat state parameter set, the heat source grade range is divided and the grade critical drop feature is extracted. Based on the waste heat state parameter set, the heat flux density variation is analyzed to generate the heat storage capacity matching factor. Based on the heat storage capacity matching factor and the grade critical drop feature, a heat charging start-up judgment table is constructed. The waste heat state parameter set is input into the AI ​​large model to identify the superposition characteristics of multi-source heat flow and generate charging coordination parameters. Based on the charging coordination parameters, the charging start-up judgment table is dynamically adjusted to generate heat storage allocation parameters. The heat storage allocation parameters are checked for heat load time sequence to identify the heat storage capacity boundary. Based on the heat charging coordination parameters, the waste heat supply confidence interval is extracted to generate the heat charging time window locking parameters. Based on the heat storage capacity boundary and the heat charging time window locking parameters, the heat balance window is verified to generate the heat storage scheduling sequence. Based on the correlation calculation between the heat storage allocation parameters and the heat storage capacity boundary, the priority value of heat energy allocation is determined, and the waste heat storage execution command is output by combining the priority value of heat energy allocation with the heat storage scheduling sequence.

[0006] The second aspect of this invention proposes an industrial waste heat storage scheduling optimization system based on an AI large model, comprising: The data acquisition module is used to collect multi-source waste heat data and heat storage medium operation data of industrial processes, and to perform time-series synchronization of the multi-source waste heat data and the heat storage medium operation data and analyze thermal parameters to generate a waste heat state parameter set. The grade analysis module is used to divide the heat source grade range according to the waste heat state parameter set, extract the grade critical drop characteristics, analyze the heat flux density variation based on the waste heat state parameter set to generate a heat storage capacity matching factor, and construct a heat charging start-up judgment table based on the heat storage capacity matching factor and the grade critical drop characteristics. The conflict coordination module is used to input the waste heat state parameter set into the AI ​​large model to identify the superposition characteristics of multi-source heat flow and generate heat charging coordination parameters, and dynamically adjust the heat charging start-up judgment table based on the heat charging coordination parameters to generate heat storage allocation parameters. The scheduling optimization module is used to perform heat load timing verification on the heat storage allocation parameters to identify the heat storage capacity boundary, extract the waste heat supply confidence interval based on the heat charging coordination parameters to generate heat charging time window locking parameters, and perform heat balance window verification based on the heat storage capacity boundary and the heat charging time window locking parameters to generate a heat storage scheduling sequence. The instruction output module is used to perform correlation calculations based on the heat storage allocation parameters and the heat storage capacity boundary to determine the priority value of heat energy allocation, and to output the waste heat storage execution instruction using the priority value of heat energy allocation combined with the heat storage scheduling sequence.

[0007] The beneficial effects of this invention are reflected in the following points: 1. Time-series synchronization and thermal parameter analysis of multi-source waste heat data and thermal storage medium operation data are performed to construct a unified foundation for describing waste heat status. Based on this, the critical drop characteristics of heat source grade are extracted and the heat flux density variation law is analyzed. A charging heat limit calibration mechanism with dual constraints of grade drop risk and capacity supply-demand deviation is established, enabling the trigger sensitivity of the charging start-up judgment table to adaptively adjust with changes in heat source process status and thermal storage capacity configuration. 2. The waste heat status parameter set is input into the AI ​​large model to extract multi-heat source collaborative fluctuation patterns, identify the superimposed power over-limit period and capacity weighted conflict intensity, and establish a charging coordination mechanism with urgency quantification ranking as the core. This drives the charging start-up judgment table to dynamically correct the trigger thresholds and quota allocations of each node, enabling the priority ranking of multi-heat source charging channels to be automatically completed within a short time window. 3. The heat storage allocation parameters are checked for heat load time sequence and the waste heat supply confidence interval is extracted. The capacity constraint and supply confidence constraint are jointly converged to determine the heat storage capacity boundary and the charging time window. A correlation calculation mechanism between margin recovery probability assessment and heat energy allocation priority weight is established, so that the output of the execution command is adjusted in real time according to the heat storage tension, which improves the adaptability of the scheduling command in the scenario of simultaneous load surge and supply fluctuation. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating an industrial waste heat storage scheduling optimization method based on an AI large model, according to the present invention.

[0009] Figure 2 This is a structural block diagram of an industrial waste heat storage scheduling and optimization system based on an AI large model, according to the present invention. Detailed Implementation

[0010] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0011] It should be noted that all directional indicators (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicator will also change accordingly.

[0012] Furthermore, the use of terms such as "first" and "second" in this application is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed in this application.

[0013] The technical solutions of the embodiments of this application will be described below.

[0014] like Figure 1 As shown, this embodiment of the invention provides an industrial waste heat storage scheduling optimization method based on an AI large model, including the following steps S110-S150: Step S110: Collect multi-source waste heat data and heat storage medium operation data of industrial process, and perform time-series synchronization and thermal parameter analysis on the multi-source waste heat data and heat storage medium operation data to generate waste heat state parameter set.

[0015] Specifically, multi-source waste heat data and heat storage medium operation data are collected from the industrial process. Multi-source waste heat data originates from various waste heat release nodes on the industrial production line. Each node is equipped with a platinum resistance temperature sensor and a heat flux meter, continuously recording temperature and heat flux density readings at 30-second intervals. The multi-source waste heat data covers three physical quantities: instantaneous temperature, heat flux density, and outlet flow rate at each node. The waste heat quality varies significantly across different processes, and the acquisition channels for different heat sources operate independently without interference. Heat storage medium operation data is collected by a sensor array inside the heat storage device. The sensors are evenly distributed along the axial direction of the heat storage tank, recording the medium temperature layer by layer. Simultaneously, readings from the charge / discharge heat flow meter and the medium level gauge are collected. The heat storage medium operation data is reported to the data acquisition unit at 1-minute intervals. When the heat storage medium is heat transfer oil or molten salt, the charge / discharge process is accompanied by significant temperature stratification; the temperature difference between the top and bottom of the tank can exceed 30 degrees Celsius during peak charging periods. Heat load data on the heat consumption side is collected by steam flow meters, hot water flow meters, and heating return water temperature sensors, respectively, for three categories of heat loads: process heat, domestic hot water, and heating. The collection interval is 1 minute, consistent with the operation data of the heat storage medium. Multi-source waste heat data, heat storage medium operation data, and heat load data on the heat consumption side are stored through independent acquisition channels, and the original timestamps retain the original acquisition granularity without pruning.

[0016] Multi-source waste heat data and thermal storage medium operation data are synchronized in time and analyzed using thermal parameters to generate a waste heat status parameter set. Time synchronization uses a 1-minute alignment granularity. Readings at 30-second granularities in the multi-source waste heat data are aggregated to a 1-minute granularity by averaging adjacent time points. The thermal storage medium operation data itself has a 1-minute granularity and requires no interpolation. In thermal parameter analysis, the instantaneous waste heat power at each heat source node is determined by the product of three parameters: mass flow rate, specific heat capacity, and inlet / outlet temperature difference, using the formula Q = G × c_p × ΔT, where Q is the instantaneous waste heat power in kilowatts, G is the mass flow rate in kilograms per second, c_p is the specific heat capacity in joules per kilogram per degree Celsius, and ΔT is the inlet / outlet temperature difference in degrees Celsius. The waste heat power at each node is accumulated along the time axis to obtain the total instantaneous waste heat supply for the entire plant. The difference between the stratification temperature of the thermal storage medium and the reference temperature is... The volume, specific heat capacity, and density parameters of each layer are converted into the enthalpy of heat storage for each layer using the formula: Q_i = ρ × V_i × c_p × (T_i - T_ref), where Q_i is the enthalpy of heat storage for the i-th layer in joules, ρ is the density of the medium in kilograms per cubic meter, V_i is the volume of the medium in the i-th layer in cubic meters, T_i is the current temperature of the i-th layer in degrees Celsius, and T_ref is a reference temperature of 20 degrees Celsius. The sum of the enthalpy values ​​of heat storage for each layer yields the total heat storage capacity of the heat storage device. Missing multi-source waste heat data is completed using nearest-neighbor interpolation. Periods with consecutive missing values ​​exceeding 5 minutes are marked as data quality anomalies. Records in the heat storage medium operation data that exceed the physical range are directly discarded and replaced with the mean of the preceding and following values. The data quality processing results are synchronously written to the quality flag bit of the corresponding time in the waste heat state parameter set. The waste heat state parameter set merges the waste heat power sequence, temperature sequence, heat flux density sequence, total heat storage time sequence of the heat storage device, and the heat load time sequence of each component on the heat consumption side into a two-dimensional structure of time rows and feature columns. The waste heat state parameter set covers a period of no less than 7 days and is organized in a sliding window manner with a window length of 24 hours and a step size of 1 hour. The real-time analysis and calculation of each step are executed independently with the current window segment as input. Features that require historical statistical support, such as the critical drop characteristics of grade, are summarized and statistically analyzed by backtracking the historical windows within the entire coverage period when extracting. The real-time analysis and historical statistics are two independent and non-overlapping operations.

[0017] Step S120: Divide the heat source grade range according to the waste heat state parameter set and extract the grade critical drop characteristics. Analyze the heat flux density variation based on the waste heat state parameter set to generate the heat storage capacity matching factor. Construct a heat charging start-up judgment table based on the heat storage capacity matching factor and the grade critical drop characteristics.

[0018] Specifically, the grade range of heat sources is divided based on the waste heat state parameter set to extract the critical grade drop characteristics. The temperature time series of each heat source node in the waste heat state parameter set is divided into intervals according to the thermodynamic grade, with grade intervals defined in 50-degree Celsius increments from the low-temperature segment to the high-temperature segment. The boundaries of each interval correspond to the temperature thresholds at which the heat storage medium can receive heat under different operating conditions. The boundaries of the low-temperature segment typically correspond to the minimum usable charging temperature of the heat storage medium, while the boundaries of the high-temperature segment correspond to the maximum allowable inlet heating temperature of the medium. Under normal smelting rhythm, the flue gas outlet temperature of the steel plant heating furnace is stably maintained in the high-grade range. When the furnace condition is adjusted or the gas supply fluctuates, the flue gas temperature drops rapidly within a few minutes. During this period, the temperature time series of the heat source corresponding to the waste heat state parameter set exhibits a step-like drop across the grade interval boundaries. The temperature time series of the waste heat state parameter set is used to detect the grade drop rate by the difference between adjacent time points. A drop rate exceeding 5 degrees Celsius per minute and lasting for more than 3 time points is identified as a critical drop event. The occurrence time, drop initiation temperature, drop amplitude, and drop duration of the critical drop event are extracted as drop descriptors. The drop descriptors of all critical drop events within the historical coverage period of the waste heat state parameter set are categorized and organized according to heat source nodes. The drop descriptor sets of each heat source node are sorted by drop amplitude to form the grade critical drop feature. In chemical plants, the rate of heat release from the reactor drops sharply at the end of a batch reaction. The residual heat state parameter is concentrated in the reactor cooling water outlet temperature, which begins to decline continuously 20 minutes before the end of the reaction, with a drop typically between 15 and 25 degrees Celsius. This regular drop pattern forms a high-frequency recurring drop descriptor in the grade critical drop characteristics. Among the grade critical drop characteristics, the descriptor with a large drop amplitude and fast drop rate has the most stringent constraints on the timing of charging start-up and is a key reference for subsequent charging heat limit calibration.

[0019] In some embodiments, the step of generating a heat storage capacity matching factor based on the heat flux density variation analysis of the waste heat state parameter set includes: dividing the waste heat state parameter set into grade distribution maps by classifying the heat source temperature range; extracting the heat flux density values ​​of each grade distribution map to generate a heat flux density sequence; performing trend analysis on the heat flux density sequence to generate a variation decay coefficient; and performing adaptive matching calculations between the variation decay coefficient and the heat flux density sequence to generate a heat storage capacity matching factor.

[0020] The waste heat state parameter set is divided into grade distribution maps by tiered intervals according to the heat source temperature range. The historical highest and lowest temperatures of each heat source node in the waste heat state parameter set determine the effective temperature range of each node. The effective temperature range is divided into tiers with a step size of 20 degrees Celsius. The number of tiers is dynamically determined according to the node temperature span; high-temperature flue gas nodes with a wide temperature span can generate more than 10 tiers, while low-temperature cooling water nodes with a narrow temperature span only generate 3 to 4 tiers. After the temperature readings at each moment in the waste heat state parameter set are assigned to the corresponding tier, the cumulative residence time and residence time percentage of each tier are calculated. Tier with a high residence time percentage indicates that the heat source operates within that temperature range for a long time, while tier with a low residence time percentage indicates that the temperature only briefly passes through that range. The grade distribution map is represented as a heat map with heat source nodes as rows and tier intervals as columns. Each cell in the map records the residence time percentage of the corresponding node within the corresponding tier. High percentage areas in the grade distribution map correspond to the normal operating grade band of the heat source, while low percentage areas correspond to grade deviation periods. When industrial kilns are operating stably, the proportion of high-temperature stages in the grade distribution map is concentrated. However, during the initial stage of annual maintenance and repair, when the kiln temperature is not yet stable, the residence time of high-temperature stages is relatively low, resulting in a wide and dispersed grade distribution map, which is significantly different from the stable operation period. The longer the period covered by the waste heat status parameter set, the more stable the statistical proportion of each stage in the grade distribution map. Short-cycle data leads to a lower proportion of stage residence time for low-frequency grade shift events, which can easily misjudge occasional grade fluctuations as long-term low-proportion stages.

[0021] Heat flux density values ​​for each grade interval are extracted from the grade distribution map to generate a heat flux density sequence. The heat flux density readings corresponding to the waste heat state parameter set within each grade interval in the grade distribution map are extracted and averaged. The average heat flux density of each grade interval is arranged from high to low temperature to form a heat flux density sequence, the length of which is equal to the total number of grade intervals. The heat flux density sequence typically shows a monotonically decreasing trend from high temperature to low temperature. High-temperature intervals correspond to periods of intense process reaction with dense waste heat output, while low-temperature intervals correspond to the end of the process or periods of low load with sparse waste heat output. In the aluminum alloy casting process, the furnace flue gas temperature is high and the heat flux density is large during the charging and melting stages. During the settling stage after unloading, the temperature drops to a lower grade, and the heat flux density decreases simultaneously. The heat flux density sequence in this process shows a significant difference between high and low temperature grades, with the ratio of heat flux density between high-temperature and low-temperature grades reaching 3 to 5 times. The heat storage device's heat charging benefit is much higher in the high-temperature grade interval than in the low-temperature grade interval. In the grade distribution map, the heat flux density sample size for the grade with an extremely low residence time percentage is insufficient. The heat flux density sequence elements for these grades are marked as low-confidence elements, and their weight is halved in subsequent decay coefficient calculations to prevent sparse grade data from causing excessive deviation from the overall decay pattern. Each element in the heat flux density sequence is accompanied by the residence time percentage of its corresponding grade as a confidence weight. When the grade distribution map is updated, the heat flux density sequence is simultaneously re-extracted to reflect the latest heat source operating status.

[0022] A trend analysis of the heat flux density sequence is performed to generate a progressive decay coefficient. The decay amount for each adjacent step in the heat flux density sequence is determined by subtracting the heat flux density of the lower-temperature step from the heat flux density of the higher-temperature step. A positive decay amount indicates that the heat flux density decreases as the grade decreases. Dividing the decay amount by the heat flux density of the higher-temperature step yields the normalized decay rate for that adjacent step pair. The normalized decay rate is typically positive. The normalized decay rates of all adjacent step pairs are arranged in descending order of temperature to form a decay rate sequence. The overall trend of the decay rate sequence reflects the regularity of heat flux density loss as the waste heat grade decreases. A uniform distribution of the decay rate sequence indicates a linear decay of heat flux density with decreasing grade. A peak in the decay rate sequence at a certain step indicates a sudden, abrupt drop in heat flux density near that grade. After the coke oven gas in the coking process is recovered and reused, the waste heat grade experiences a sharp drop in heat flux density near a specific temperature due to the complete release of latent heat from phase change. The peak in the decay rate of the heat flux density sequence at this temperature level indicates a supply cliff in the heat storage device at this grade level. Therefore, heat storage scheduling should prioritize completing the charging process before the grade enters this level, rather than waiting for the grade to continue declining. The progressive decay coefficient is defined as the weighted average of the decay rate sequence, with the weight representing the residence time percentage of the corresponding level. The formula is K_decay=Σ(r_i×w_i) / Σ(w_i), where K_decay is the progressive decay coefficient, r_i is the normalized decay rate of the i-th adjacent level pair, and w_i is the residence time percentage weight of the corresponding level. The larger the value of the progressive decay coefficient, the more severe the decay of heat flux density when the waste heat grade decreases. The higher the priority of the heat storage device in the high grade period, the more attention is paid to the stage corresponding to the peak moment of decay rate in the heat flux density sequence in the progressive decay coefficient calculation. The high grade section above the stage where the peak is located is correspondingly lowered when the heat storage trigger threshold group is calibrated to prioritize the heat storage response of high grade waste heat.

[0023] An adaptive matching factor for thermal storage capacity is generated based on the varying attenuation coefficient and the heat flux density sequence. The varying attenuation coefficient reflects the overall attenuation intensity of heat flux density as the waste heat grade decreases, while the heat flux density sequence reflects the absolute heat supply level of each stage. Combining the two allows determination of the capacity ratio that the thermal storage device should reserve at each temperature level. The capacity demand intensity of each stage is obtained by dividing the heat flux density of each stage in the heat flux density sequence by the maximum value of the sequence and then multiplying it by the varying attenuation coefficient. Stages with higher capacity demand intensity correspond to temperature levels where the thermal storage device should reserve more available capacity. After normalizing the capacity demand intensity of each stage, a capacity demand distribution vector is formed. The difference between the capacity demand distribution vector and the actual current free capacity ratio of each temperature level of the thermal storage device yields the capacity supply-demand deviation. A positive deviation indicates insufficient capacity reservation at that temperature level, while a negative deviation indicates excessive reservation. The thermal storage capacity matching factor represents the overall capacity matching degree by weighted summation of the absolute values ​​of the capacity supply and demand deviations of all temperature layers. The calculation formula is F_match=Σ(|d_k|×e_k) / Σ(e_k), where F_match is the thermal storage capacity matching factor, d_k is the capacity supply and demand deviation of the k-th temperature layer, and e_k is the capacity demand intensity weight of the corresponding temperature layer. Both d_k and e_k are dimensionless. The value of F_match ranges from 0 to 1. Temperature layers with larger absolute deviations contribute more to the weighted summation, ensuring that the impact of severely mismatched temperature layers on the thermal storage capacity matching factor is fully reflected. A higher thermal storage capacity matching factor indicates a worse match between the current thermal storage capacity configuration and the waste heat supply structure. The thermal storage device needs to adjust the allocation ratio of the reserved capacity of each temperature layer. When the gradual decay coefficient is large, the weight of the thermal storage capacity matching factor on the capacity deviation of the high-temperature layer is correspondingly increased to prioritize the reception capacity of high-grade waste heat. When there are structural changes in the time sequence of heat flux densities of each heat source in the waste heat state parameter set, the thermal storage capacity matching factor is recalculated accordingly.

[0024] In some embodiments, constructing a charging start-up determination table based on the heat storage capacity matching factor and the grade critical drop characteristics includes: jointly evaluating the drop amplitude and drop rate of the grade critical drop characteristics to generate a critical drop amplitude set; identifying a drop gradient steep increase segment from the critical drop amplitude set to generate a drop warning indicator; performing a drop rate adaptive charging limit calibration based on the drop warning indicator and the heat storage capacity matching factor to generate a charging trigger threshold group; and constructing a charging start-up determination table based on the charging trigger threshold group.

[0025] A critical drop amplitude and drop rate are jointly evaluated to generate a critical drop amplitude set for the critical drop characteristics. The drop descriptors for each heat source node in the critical drop characteristics carry two key dimensions: drop amplitude and drop rate. Drop amplitude reflects the absolute amount of grade decrease, while drop rate reflects the severity of the grade decrease. When both are simultaneously high, the heat storage device must complete the charging action with the fastest response speed. The drop amplitude and drop rate of each drop descriptor in the critical drop characteristics are normalized and then multiplied to obtain the joint evaluation score, calculated as S_joint = A_norm × V_norm, where S_joint is the joint evaluation score, A_norm is the normalized drop amplitude value, and V_norm is the normalized drop rate value, both dimensionless. The value of S_joint ranges from 0 to 1. A drop event with a high joint evaluation score indicates a deep and rapid grade drop, requiring the most stringent timing for charging. After pouring, the residual heat temperature of the cooling water in the foundry drops by more than 40 degrees Celsius within a very short time. This node exhibits an extremely high drop rate in the critical grade drop characteristics, ranking among the top drop events in the joint evaluation score. The charging response time for this node must be within minutes to capture the final charging window before the drop. Drop events with a joint evaluation score exceeding the mean plus one standard deviation of all descriptors are included in the critical drop amplitude set. Each element in the critical drop amplitude set retains four pieces of information: drop amplitude, drop rate, joint evaluation score, and historical recurrence rate. Drop events with low historical recurrence rates, despite high evaluation scores, are still marked with low confidence to prevent occasional extreme events from dominating the calibration of charging heat limits. The number of elements in the critical drop amplitude set reflects the overall drop risk level of the current heat source group. A high number of elements suggests that the overall charging heat limits need to be tightened to cope with frequent sharp grade drops.

[0026] The critical drop amplitude set is used to identify steeply increasing gradient segments and generate drop warning indicators. After arranging all elements in the critical drop amplitude set in ascending order of drop amplitude, the difference in drop amplitude between adjacent elements forms a gradient sequence. Locations where the difference in the gradient sequence suddenly increases correspond to boundary points where there are clear stratifications in the drop amplitude distribution. Locations where the gradient sequence difference exceeds the mean plus two standard deviations are identified as gradient steepening points. These gradient steepening points naturally divide the critical drop amplitude set into low-amplitude and high-amplitude drop segments. The high-amplitude drop segment corresponds to the group of drop events with severe grade decline and the most urgent heating demand. All elements in the high-amplitude drop segment of the critical drop amplitude set are marked as high-priority drop warning indicators, while the low-amplitude drop segment is marked as low-priority drop warning indicators. The heat source node corresponding to the high-priority drop warning indicator obtains the minimum trigger threshold value at the timed heating limit indicator, ensuring that the heating action is initiated in advance before a significant grade drop. During peak load reduction, a combined heat and power (CHP) unit experienced a sudden decrease in turbine extraction steam, causing the waste heat grade to drop two levels within 15 minutes. The critical drop amplitude concentration resulted in an extremely high joint evaluation score for this event and a stable historical recurrence rate, leading to a high-priority drop warning. Peak load reduction scenarios occur 2-3 times daily during seasons with frequent grid load fluctuations, and the high-priority labeling of the drop warning indicator has the most significant constraint on heat charging decisions during this period. The drop warning indicator carries the corresponding heat source node number and priority label. When there are no steep gradient increases in the critical drop amplitude concentration, all drop warning indicators are assigned low priority, indicating that the amplitude distribution of each drop event is continuous without significant stratification.

[0027] A charging trigger threshold group is generated based on the drop rate adaptive calibration of charging heat supply thresholds using drop warning indicators and thermal storage capacity matching factors. The priority of the drop warning indicator determines the urgency of the charging response, while the thermal storage capacity matching factor determines the current thermal storage capacity's ability to receive waste heat supply. Together, they determine the tightness of the charging trigger threshold. The charging trigger threshold is defined as the lower limit threshold of waste heat power at each heat source node in the corresponding grade cascade. When the real-time waste heat power in the waste heat status parameter set is higher than this threshold, it is determined to be an effective charging supply, and charging is triggered. When the waste heat power is lower than the threshold, it is determined to be insufficient supply, and charging is not triggered. The charging trigger threshold for the high-priority drop warning indicator corresponding to the heat source node is adaptively calibrated based on the grade drop rate. Nodes with fast drop rates have lower threshold values ​​to ensure that charging is triggered when the grade just enters the initial drop phase and the power is still sufficient. Nodes with slow drop rates have threshold values ​​appropriately increased to the power level in the middle of the drop before activation. The difference in threshold calibration methods between the two types of nodes ensures that the charging response timing is accurately matched with the actual grade drop rhythm of each heat source. When the heat storage capacity matching factor is too high, it indicates that the current heat storage capacity configuration is insufficient, and the overall charging trigger threshold is lowered to more actively seize the waste heat supply window. When the heat storage capacity matching factor is too low, the heat storage capacity is sufficient, and the charging trigger threshold is appropriately raised to avoid frequent equipment activation due to ineffective small-scale charging. The charging trigger threshold values ​​of each heat source node and each grade tier combination are summarized to form a charging trigger threshold group. Each element in the charging trigger threshold group carries three pieces of information: the corresponding node number, the grade tier number, and the threshold value. The threshold values ​​of the charging trigger threshold group elements of high-priority drop warning indicator nodes are generally lower than those of low-priority nodes, reflecting the early response strategy to high-risk drop heat sources. The overall threshold level of the charging trigger threshold group is periodically refreshed with the dynamic change of the heat storage capacity matching factor.

[0028] A charging start-up judgment table is constructed based on the charging trigger threshold group. The heat source node number and grade ladder number of each element in the charging trigger threshold group constitute the row and column index of the charging start-up judgment table. The threshold value is filled into the corresponding cell. The charging start-up judgment table is organized by heat source node as the row and grade ladder as the column. Each cell stores the charging trigger threshold of the heat source at the corresponding grade layer. The charging start-up judgment table covers all combinations of heat source nodes and all grade ladders. When a new waste heat node is added to the plant, a corresponding row is added to the charging start-up judgment table, and the threshold is calibrated according to the drop warning indicator and heat storage capacity matching factor of the new node. For the first three analysis windows of the new node, due to insufficient historical data, the initial value of the threshold for the corresponding row is substituted with the historical average value of similar heat sources. Similar heat sources are identified as existing nodes with the same process type and an effective temperature range overlap of more than 60%. After sufficient historical data is accumulated, the values ​​are automatically updated to the accurate calibration values. During the low-load period of the night shift, the overall waste heat quality of each heat source is relatively low. The threshold values ​​for the charging triggering threshold group are correspondingly lowered during this period. The overall trigger sensitivity of the charging start-up judgment table is higher during the night shift than during the day shift, preventing low-grade waste heat from missing charging opportunities due to excessively high thresholds. In the charging start-up judgment table, the threshold values ​​for different quality levels at the same heat source node decrease as the quality increases. A low threshold for the high-grade level means that even if the waste heat power in the high-temperature zone is low, it still meets the triggering conditions, thus prioritizing charging to fully utilize the high-grade heat. A higher threshold for the low-grade level means that low-temperature waste heat is only triggered for charging when the power is significantly higher than the threshold. The difference between the two threshold levels reflects the direct impact of heat quality on the economic value of charging.

[0029] Step S130: Input the waste heat state parameter set into the AI ​​large model to identify the superposition characteristics of multi-source heat flow and generate heat charging coordination parameters. Based on the heat charging coordination parameters, dynamically adjust the heat charging start-up judgment table to generate heat storage allocation parameters.

[0030] In some embodiments, the step of inputting the waste heat state parameter set into an AI large model to identify multi-source heat flow superposition features and generate charging coordination parameters includes: aligning the waste heat state parameter set into the AI ​​large model with multiple heat sources to generate a heat source time series matrix; extracting the heat flow intensity distribution of each heat source from the heat source time series matrix and performing superposition operations to generate multi-source heat flow superposition features; identifying overlapping time window segments based on the multi-source heat flow superposition features and the heat source time series matrix to generate conflict marker groups; and generating charging coordination parameters based on the conflict marker groups and the multi-source heat flow superposition features.

[0031] The waste heat state parameter set is input into the AI ​​large-scale model for multi-heat source time-series alignment to generate a heat source time-series matrix. The waste heat power time-series data of each heat source node in the waste heat state parameter set are collected from different equipment control systems. There are slight deviations in the clocks of each system. The accumulated deviations can reach tens of seconds in high-frequency acquisition scenarios. Direct splicing will cause misalignment of the heat source time sequences on the time axis. The AI ​​large-scale model applies a cross-correlation alignment algorithm to the time sequences of each heat source in the waste heat state parameter set. Using the relative time-series relationship of the waste heat power peaks and troughs as anchor points, it detects the systematic offset of each time sequence. The detected offsets are then used to apply time axis translation correction to each time sequence. The waste heat data of the three kiln lines in the glass furnace plant comes from three independent DCS systems. The maximum clock deviation of each system can reach 45 seconds. The power peaks of the three time sequences in the waste heat state parameter set are staggered in the unaligned state. After alignment by the AI ​​large-scale model, the power peaks of the three channels converge to the same moment, and the actual peak value of the superimposed power of multiple heat sources is accurately restored. After alignment, the time series data of each heat source are stacked row-wise to form a heat source time series matrix. The number of rows in the matrix equals the total number of heat source nodes, and the number of columns equals the total number of time points within the analysis window. Each element in each row of the heat source time series matrix represents the residual heat power value of the corresponding heat source at each time point. After zero-mean, unit-variance normalization, the heat source time series matrix retains a copy of the original physical values. The normalized copy is used for feature extraction within the AI ​​large model, while the original physical value copy is used for subsequent heat calculations. When a new heat source node is added to the residual heat state parameter set, the number of rows in the heat source time series matrix is ​​expanded accordingly.

[0032] The heat flux intensity distributions of each heat source are extracted from the heat source time series matrix and superimposed to generate multi-source heat flux superposition characteristics. The probability distribution of heat flux intensity in each row of the heat source time series matrix is ​​fitted using the kernel density estimation method. The heat flux intensity distribution of each heat source describes the probability of its occurrence at different power levels within the analysis window. The waste heat power values ​​of each row of the heat source time series matrix at the same moment are summed column by column to obtain the instantaneous total waste heat power time series of the entire plant. The peak period of the total waste heat power time series corresponds to the superposition peak of multiple heat sources operating at high output simultaneously. When the production load of the catalytic cracking unit, atmospheric distillation tower, and hydrorefining reactor in the petrochemical enterprise increases simultaneously, the waste heat power of the three sources rises simultaneously. The values ​​of the three rows in the heat source time series matrix are synchronously high during this period. After column summation, the peak value of the total power far exceeds the peak level of any single heat source. The charging pressure faced by the heat storage device during this period is underestimated by the single heat source analysis. The four statistical features of the total waste heat power time series—mean, peak value, peak duration, and peak-to-valley ratio—are extracted and spliced ​​together to form the multi-source heat flow superposition feature. The multi-source heat flow superposition feature also includes the Pearson correlation coefficient matrix between each pair of heat source power time series. Heat source pairs with high correlation coefficients indicate that their superposition behavior on the time axis is highly synchronized. The higher the superposition peak value of the multi-source heat flow superposition feature, the more intense the competition for the heat storage device in the corresponding time period.

[0033] For example, the step of identifying overlapping time window segments and generating conflict marker groups based on the multi-source heat flow superposition features and the heat source time series matrix includes: performing a sliding time window scan on the triggering time of each heat source in the heat source time series matrix to generate a time window density distribution map; extracting multi-source triggering dense segments from the time window density distribution map to generate a dense area identifier set; performing a capacity-weighted conflict intensity assessment on the dense area identifier set in combination with the multi-source heat flow superposition features to generate a capacity-weighted conflict intensity sequence; and identifying overlapping time window segments and generating conflict marker groups based on the capacity-weighted conflict intensity sequence.

[0034] A sliding time window scan is performed on the trigger times of each heat source in the heat source time series matrix to generate a time window density distribution map. The waste heat power time series of each heat source in the heat source time series matrix is ​​checked moment-by-moment according to the original trigger threshold of the charging start-up judgment table to determine whether the trigger conditions are met. Moments that meet the trigger conditions are marked as trigger points. The set of trigger point moments for each heat source is extracted and marked on the time axis. The same heat source may have multiple discrete trigger points within the same analysis window, or it may form a continuous trigger segment. The sliding time window advances minute by minute on the time axis with a width of 10 minutes, counting the number of trigger points from different heat sources within each window position. The trigger point count sequence is arranged in chronological order to form the time window density distribution map. In a textile printing and dyeing plant, waste heat from the setting machine, waste heat from the dyeing vat, and boiler flue gas are the three heat sources that are concentrated before the shift change to complete the work of the shift. Operators simultaneously increase the operating load of each piece of equipment before the end of the shift to meet production targets. The triggering times of the three heat sources are highly concentrated within 30 minutes before the end of the shift, and the time window density distribution chart shows a significant count peak during this period. At the peak, the number of heat sources triggered simultaneously within a single time window reaches the highest level of the day. The overall mean and standard deviation of the count sequence in the time window density distribution chart vary with the production rhythm. During holidays when production is suspended, the trigger points in the heat source time sequence matrix are extremely sparse, and the count in the time window density distribution chart is close to zero throughout. On the first day of factory resumption, due to concentrated preheating and debugging of equipment, the trigger points of each source appear densely in a short period of time, and the time window density distribution chart shows an atypical peak during the resumption period. The count sequence of the time window density distribution chart is normalized for subsequent extraction of dense sections.

[0035] Multi-source triggered dense segments are extracted from the time window density distribution map to generate a dense area identifier set. Time window positions in the time window density distribution map where the count value exceeds the global mean plus one standard deviation are identified as dense points. Three or more consecutive dense points constitute a dense segment. When the interval between adjacent dense segments is less than two time windows, they are merged into one segment. The merging operation ensures that short-term, intermittent concentrated triggering events are identified as a complete dense segment rather than two independent segments. When the distribution of dense points in the time window density distribution map is concentrated, merging them forms a long-term dense segment, corresponding to continuous high-pressure periods triggered synchronously by multiple heat sources. For example, during the winter heating peak, a combined heat and power unit operates at full load simultaneously with the turbine extraction steam and the waste heat boiler. The heating demand increases accordingly for every 1 degree Celsius decrease in the external temperature. The dual-path heat charging channel of the heat storage device is continuously occupied. The time window density distribution map maintains a high count throughout the entire heating peak period. After merging, the dense segment covers a continuous time range of several hours, with a duration far exceeding the triggering event of a single production batch. The dense zone identifier set records three pieces of information for each dense zone segment: the start time, the end time, and the peak count within the segment. The peak count reflects the maximum number of heat sources simultaneously triggered within that segment; a higher peak indicates more intense competition for the charging channels. When there are no dense points in the time window density distribution map, the dense zone identifier set is assigned an empty set, indicating that the trigger times of each heat source within the current analysis window are dispersed, and there is no period of concentrated competition for charging channels among multiple sources. A large number of dense zone identifier segments with long durations suggest that the current production rhythm is creating continuous high pressure on the thermal storage scheduling, and maintenance personnel need to pay close attention to the charging channel allocation plan during this period.

[0036] A capacity-weighted conflict intensity sequence is generated by combining the dense area identifier set with the multi-source heat flow superposition characteristics. The power values ​​of participating heat sources at each moment within each dense area segment of the dense area identifier set are extracted from the original physical values ​​of the multi-source heat flow superposition characteristics. The rated charging power of each heat source is used as a weight to sum the real-time power values ​​of the participating heat sources. The weighted sum is compared with the upper limit of the rated charging power of the thermal storage device to obtain the capacity occupancy rate at that moment. A capacity occupancy rate exceeding 1.0 indicates that the charging power demand of the participating heat sources exceeds the rated upper limit of the thermal storage device. The higher the capacity occupancy rate, the more severe the conflict. Periods with persistently high capacity occupancy rates mean that the thermal storage device is in a state of full charging load for a long time, and any new heat source triggering conditions cannot obtain an effective response. The capacity occupancy rates at each moment within the period covered by the dense area identifier set are arranged in chronological order to form a capacity-weighted conflict intensity sequence. The corresponding position in the capacity-weighted conflict intensity sequence for moments not covered by the dense area identifier set is assigned a value of 0, indicating that there is no charging channel competition at that moment. When heat source pairs with high correlation coefficients in the multi-source heat flow superposition feature appear simultaneously in the same segment of the dense area identifier set, the capacity-weighted conflict intensity sequence value of that segment is superimposed with a synchronization additional coefficient on the basic capacity occupancy rate. The synchronization additional coefficient is linearly mapped to the range of 0.05 to 0.2 based on the relative position between the minimum and maximum values ​​of the correlation coefficients of all highly correlated heat source pairs in the current dense segment. When the correlation coefficient is equal to the minimum value, the additional coefficient is 0.05; when it is equal to the maximum value, the additional coefficient is 0.2. The intermediate values ​​are determined by proportional interpolation, reflecting the synergistic amplification effect of highly correlated heat sources on synchronization triggering. Moments in the capacity-weighted conflict intensity sequence that exceed 1.2 are marked as severe conflict moments. Severe conflict moments are marked with the highest conflict level during the conflict marker group generation stage.

[0037] Conflict marker groups are generated by identifying overlapping time windows based on the capacity-weighted conflict intensity sequence. Consecutive time segments with values ​​exceeding 1.0 in the capacity-weighted conflict intensity sequence are defined as capacity overrun periods. The intersection of these capacity overrun periods and the corresponding densely populated areas in the densely populated area marker set is used to identify overlapping time windows. This intersection operation ensures that the conflict marker group only includes time segments that simultaneously meet both the conditions of triggering densely populated areas and exceeding capacity limits. Time segments that only trigger densely populated areas but still have capacity margins are not included in the conflict marker group. Adjacent overlapping time windows are merged when the interval is less than 3 time points. After merging, the outer boundary is marked as the complete conflict period range. The merging operation uniformly handles continuous conflict scenarios where capacity briefly drops and then exceeds limits again. Overlapping time windows with a concentration of severe conflict times in the capacity-weighted conflict intensity sequence are marked as high-intensity conflict segments in the conflict marker group, while the rest are marked as medium-to-low-intensity conflict segments. High-intensity conflict segments trigger a stricter priority ranking and quota compression mechanism during the heat charging coordination phase. During full-production periods, the combined power of waste heat from the kiln head cooler and waste heat from the kiln tail preheater at a cement plant exceeds the rated charging limit of the thermal storage system. The capacity-weighted conflict intensity sequence consistently exceeds 1.3 during this period, with a high correlation coefficient approaching 0.85. This, coupled with a synchronization factor, raises the peak value to 1.46. The corresponding overlapping time window segment is marked as a high-intensity conflict segment in the conflict marker group. This marking drives the charging coordination to prioritize the kiln head waste heat source with a higher grade at the top of the charging queue. Each element in the conflict marker group carries four pieces of information: the time range of the segment, the conflict intensity level, the participating heat source number, and the average over-limit power. When the capacity-weighted conflict intensity sequence is consistently below 1.0, the conflict marker group is empty, indicating that the triggering demands of all heat sources within the current analysis window are within the capacity range of the thermal storage device and can be simultaneously met, eliminating the need to activate the charging coordination sorting mechanism.

[0038] Based on the conflict marker group and the superimposed characteristics of the multi-source heat flow, charging coordination parameters are generated. The participating heat sources in each conflict segment of the conflict marker group are prioritized according to a comprehensive score of waste heat grade, heat flux density, and grade drop risk in the multi-source heat flow superimposed characteristics. The comprehensive score formula is R_i = 0.4 × Q_grade_i + 0.35 × F_density_i + 0.25 × D_risk_i, where R_i is the priority score of the i-th heat source, Q_grade_i is the normalized waste heat grade, F_density_i is the normalized heat flux density, and D_risk_i is the normalized grade drop risk, taken from the normalized value of the drop amplitude of the corresponding heat source node in the S120 grade critical drop characteristics. Within the same segment of the conflict marker group, the participating heat sources are arranged in descending order of R_i, with the higher-ranked heat source having priority in obtaining the allocation of charging channels. The severity level of a conflict is determined by combining the over-limit power value of each conflict segment in the conflict marker group with the number of participating heat sources. A conflict is classified as high-severity when the over-limit power exceeds 50% of the rated charging power and the number of participating heat sources exceeds 3. Other cases are classified as medium-to-low-severity conflicts. The charging coordination parameters integrate the time range, severity level, priority score of each heat source, and suggested delay duration for all conflict segments. The suggested delay duration is estimated based on the grade drop rate of low-priority heat sources. Heat sources with slow grade drop rates can tolerate longer delays. The charging coordination parameters are organized in a structured list format, with the list length equal to the total number of conflict segments in the conflict marker group.

[0039] In some embodiments, the step of dynamically adjusting the heat charging start-up decision table based on the heat charging coordination parameters to generate heat storage allocation parameters includes: quantifying and sorting the heat charging coordination parameters by urgency to generate an urgency ranking sequence; extracting the heat charging quota of each heat source from the urgency ranking sequence to generate a heat charging quota group; modifying the triggering conditions in the heat charging start-up decision table based on the heat charging quota group to generate an adjustment decision table; and generating heat storage allocation parameters according to the adjustment decision table and the heat charging quota group.

[0040] The urgency of the heat charging coordination parameters is quantified and ranked to generate an urgency ranking sequence. The priority score of each heat source in the heat charging coordination parameters reflects the competition ranking during the current conflict period. However, heat sources with similar priority scores still need to be further distinguished by an urgency index. The urgency comprehensively considers the remaining margin of the heat source's current grade from the critical drop value and the remaining time expected to reach the critical value. The remaining grade margin is determined by the difference between the current temperature of the heat source in the heat charging coordination parameters and the drop start temperature of the corresponding node's critical drop characteristic. The smaller the difference, the closer the grade is to the critical drop edge, and the higher the urgency of the heat charging. The remaining time expected to reach the critical value is estimated by linear extrapolation based on the current grade drop rate. Heat sources with a fast rate of decrease and small margin have the highest urgency. In a certain steelmaking process, the furnace temperature drops rapidly at the end of the converter tapping stage, with only about 8 degrees Celsius of grade remaining margin and the rate of decrease continuing to accelerate. It is expected to reach the critical value within 3 minutes. The comprehensive urgency score of this heat source is significantly higher than that of the sintering process waste heat with stable grade during the same period. The urgency score is a weighted sum of the normalized inverse value of the remaining margin and the normalized inverse value of the remaining time, with weights of 0.5 and 0.5 respectively. The normalized inverse value is defined as 1 minus the normalized value of the corresponding indicator among all participating heat sources in the current conflict period. The larger the normalized inverse value, the smaller the margin or the shorter the remaining time of the heat source, and the higher the urgency of the charging. All heat sources participating in the conflict in the charging coordination parameters are arranged in descending order of urgency score to form an urgency ranking sequence. The length of the urgency ranking sequence is equal to the total number of heat sources competing in the current conflict period. The ranking of different heat sources in the urgency ranking sequence in the same conflict period directly determines the allocation order of the charging channels. Heat sources with higher rankings have priority, while heat sources with lower rankings can only be started after the charging channels become available. The urgency ranking sequence is dynamically rearranged as the conflict period is updated in real time, and the urgency of each heat source is reassessed and the ranking is refreshed every sampling period.

[0041] The charging quota for each heat source is extracted from the urgency ranking sequence to generate charging quota groups. Heat sources ranked higher in the urgency sequence receive priority access to the charging channels of the thermal storage device. The maximum charging power that a charging channel can handle per unit time is determined jointly based on the rated charging power of the thermal storage device and the current thermal storage status. The first heat source in the urgency ranking sequence is allocated 60% of its rated charging power as a base quota, the second-ranked heat source is allocated 50% of its remaining charging power, and subsequent heat sources are allocated decreasing proportions of their remaining power. This decreasing proportion design ensures absolute priority for high-urgency heat sources in the allocation of charging resources, preventing low-urgency heat sources from occupying too many channels and causing a continuous drop in the grade of high-urgency heat sources. After the quota allocation is completed, the upper limit of the charging power for each heat source and the recommended charging duration together constitute the charging quota for a single heat source. The recommended charging duration is estimated by dividing the current grade's remaining margin by the grade decline rate, ensuring sufficient charging reserves are completed before the grade reaches the critical value. The charging quotas of all heat sources are aggregated to form a charging quota group. Each element in the charging quota group carries four pieces of information: heat source node number, charging power limit, recommended charging duration, and quota source ranking. If the remaining quota obtained by the heat source at the bottom of the urgency ranking is lower than its minimum effective charging power threshold, the charging quota of that heat source is marked as unexecutable, and the charging action is postponed to the next time period for re-coordination. Heat source nodes with unexecutable quotas in the charging quota group are not included in the correction scope of this adjustment judgment table. The total quota of the entire charging quota group does not exceed the rated charging power limit of the thermal storage device to meet equipment safety constraints.

[0042] Based on the heat charging quota group, the triggering conditions in the heat charging start-up decision table are modified to generate an adjustment decision table. The upper limit of heat charging power and the recommended heat charging duration for each heat source in the heat charging quota group are used to cover and modify the trigger threshold and continuous heat charging duration constraints of the corresponding nodes in the heat charging start-up decision table. The modified trigger threshold reflects the adjustment result of the coordination ranking on the timing of heat charging start-up. The upper limit of heat charging power for high-ranking heat sources in the heat charging quota group is usually higher than the original allocation value in the heat charging start-up decision table. The trigger threshold is lowered accordingly to more actively capture the heat charging window and ensure that high-urgency heat sources trigger the heat charging response as soon as the grade begins to drop. The upper limit of heat charging power for low-ranking heat sources is compressed, and the trigger threshold is raised so that start-up is only allowed during the gap period after the waste heat supply is obviously sufficient and the heat charging of high-ranking heat sources has been completed, so as to avoid competing with high-ranking heat sources for heat charging resources. When the waste heat from both the roller kiln and the drying kiln in a ceramics factory simultaneously triggers the charging conditions, the flue gas grade at the roller kiln outlet reaches 320 degrees Celsius, far exceeding the 80 degrees Celsius of the drying kiln. The roller kiln ranks first in the charging quota group and has ample quota. In the charging start-up judgment table, the trigger threshold for the roller kiln node is lowered to allow immediate start-up once the grade is met, while the trigger threshold for the drying kiln node is raised to allow start-up only during the downtime after the roller kiln's charging is completed. The charging duration constraints for both nodes are synchronously overridden with the original settings according to the quota recommendation duration. The adjustment judgment table overlays the charging quota group's correction items on top of the charging start-up judgment table. Both the original and corrected thresholds are retained in the adjustment judgment table. The correction source is clearly marked as either a coordinated adjustment or the original setting, facilitating maintenance personnel to trace the source of threshold changes. After the period of conflict in the charging coordination parameters ends, the adjustment judgment table automatically reverts to the original threshold, maintaining adaptability to subsequent operating cycles.

[0043] Based on the adjustment judgment table and the heat charging quota group, heat storage allocation parameters are generated. The three scheduling elements—the corrected trigger threshold, the upper limit of heat charging power, and the recommended heat charging duration—for each heat source node in the adjustment judgment table are extracted and compared with the corresponding quota information in the heat charging quota group. Nodes that match are directly generated as scheduling execution entries. For nodes with discrepancies, the corresponding entries in the adjustment judgment table are overwritten with the data from the heat charging quota group. The source of these discrepancies is usually a slight difference between the approximate rounding during the adjustment judgment table correction process and the precise calculated value of the heat charging quota group. All heat source node scheduling execution entries are integrated into heat storage allocation parameters in urgency order. These parameters are then sorted in ascending order by heat charging execution time, indexed by time period. When multiple heat source scheduling execution entries exist at the same time, their priority execution order is determined by their urgency order. The thermal storage allocation parameters also include the start and end times of the charging window for each heat source node. The charging window is defined by the modified trigger threshold and recommended charging duration in the adjustment judgment table. Charging can only be initiated when the waste heat supply within the window meets the charging conditions. Even if the waste heat power occasionally meets the threshold outside the window, it will not be included in the effective scheduling plan to prevent frequent start-ups and shutdowns of the charging equipment due to unstable supply after occasional waste heat triggering charging outside the window. Heat source nodes with zero conflict severity in the charging coordination parameters do not participate in the coordination sorting. Their thermal storage allocation parameter scheduling execution entries directly use the original settings of the charging start judgment table. The thermal storage allocation parameters are finally output in a structured list format, with the list length equal to the total number of heat source nodes in the current analysis window. Each entry is updated independently without interference.

[0044] Step S140: Perform heat load time sequence verification on the heat storage allocation parameters to identify the heat storage capacity boundary; extract the waste heat supply confidence interval based on the heat charging coordination parameters to generate the heat charging time window locking parameters; and perform heat balance window verification based on the heat storage capacity boundary and the heat charging time window locking parameters to generate the heat storage scheduling sequence.

[0045] In some embodiments, the step of performing heat load time-series verification and identifying heat storage capacity boundaries on the heat storage allocation parameters includes: extracting heat load demand for each time period from the heat storage allocation parameters to generate a heat load demand sequence; identifying load surge warning features from the heat load demand sequence to generate a load surge warning identifier; performing capacity margin calculation on the heat load demand sequence based on the load surge warning identifier to generate a capacity margin value; and determining the heat storage capacity boundary based on the capacity margin value and the load surge warning identifier.

[0046] Heat load demand sequences are generated by extracting heat load requirements for each time period from the thermal storage allocation parameters. The average heat load for each time period is extracted from the charging execution time of each heat source node in the thermal storage allocation parameters, with a granularity of 30 minutes. The average heat load covers the sum of three categories of sub-demands: process heat, heating heat, and domestic hot water. In the paper mill's cooking process, steam consumption jumps from no load to full load within minutes during batch feeding, while the paper machine's drying section continuously and stably consumes steam. The average heat load for the corresponding time period in the thermal storage allocation parameters covers the peak demand resulting from the superposition of these two processes. Analyzing any single process in isolation underestimates the actual heat release pressure of the thermal storage device. The average heat load demand for each time period in the heat load demand sequence is arranged chronologically. The overall trend of the sequence depicts the basic demand intensity of the heat user side at each time period of the day. The higher average heat load demand at the charging plan time in the thermal storage allocation parameters indicates that the charging period overlaps with the peak heat consumption period. The simultaneous occurrence of charging and heat release activities creates dual pressure on the heat budget management of the thermal storage device. A period in the heat load demand sequence where the difference between adjacent time periods is consistently positive indicates that the heat demand is in an expansion phase. The peak-to-valley difference in the heat load demand sequence directly determines the capacity margin of the thermal storage device at different time periods.

[0047] Load surge warning features are identified from the heat load demand sequence to generate load surge warning indicators. The heat load demand sequence is analyzed by differentially detecting the rate of change of demand between adjacent time periods. Time periods with a rate of change exceeding the sequence mean plus 1.5 times the standard deviation are identified as candidate surge points. Two or more consecutive candidate surge points constitute a surge warning segment. The start time, duration, and surge magnitude of the surge warning segment are extracted as warning descriptors. For example, during the concentrated pouring period, the gas consumption and waste heat output of a foundry's smelting furnace increased simultaneously, corresponding to a sharp rise in heating demand. The heat load demand sequence showed four consecutive candidate surge points during this period, with the surge magnitude exceeding 60% of the previous trough. The surge warning segment is compared with the surge patterns of similar time periods in the historical heat load demand sequence. A historical recurrence rate exceeding 0.65 is classified as a periodic surge, while a rate below 0.2 is classified as an occasional surge. The load surge warning indicator consists of warning descriptors and type labels for all surge warning sections. The advance time for the load surge warning indicator corresponding to periodic surges is set to two time periods before the start of the surge, allowing sufficient time for the heat storage device to accumulate heat. The advance time for the load surge warning indicator corresponding to occasional surges is compressed to one time period and a low confidence label is added. When there are no surge candidate points in the heat load demand sequence, the load surge warning indicator is assigned an empty set.

[0048] Based on load surge warning indicators, capacity margin calculations are performed on the heat load demand sequence to generate capacity margin values. Within the lead time period of the load surge warning indicator, the thermal storage device needs to complete its heat storage reserves to cover subsequent surge demands. The capacity margin calculation is obtained by subtracting the cumulative heat release demand during the peak surge period from the expected heat storage capacity calculated from the cumulative heat charging plan in the thermal storage allocation parameters at the end of the warning lead time period. The cumulative heat release demand for each surge segment in the heat load demand sequence is determined by integrating the average heat load at each moment during the surge period. The integration result reflects the total heat that the thermal storage device needs to release to the heat user during the duration of the surge. The difference between the peak period integral value corresponding to the load surge warning indicator and the expected heat storage capacity in the corresponding period in the thermal storage allocation parameters constitutes the capacity margin value for that surge segment. A positive capacity margin value indicates sufficient expected heat storage capacity to cover the surge demand, while a negative capacity margin value indicates insufficient expected heat storage capacity and a risk of heat supply interruption. When there are multiple surge warning sections in the heat load demand sequence, the capacity margin value of each section is calculated separately. The capacity margin values ​​of all sections are arranged in chronological order to form a capacity margin sequence. The negative value section in the capacity margin sequence is the key reference area for the lower limit of the heat storage capacity boundary. When the load surge warning is marked as an empty set, the capacity margin sequence is positive throughout, and the capacity configuration of the heat storage device is sufficient under the current heat charging plan.

[0049] The thermal storage capacity boundary is determined based on the capacity margin value combined with the load surge warning indicator. The minimum value in the capacity margin value sequence is defined as the lower limit of the capacity margin. The lower limit of the capacity margin plus the safety redundancy is mapped to the minimum heat storage capacity that the thermal storage device should maintain. This minimum heat storage capacity is the lower limit of the thermal storage capacity boundary. The safety redundancy is determined based on the type of surge warning segment in the load surge warning indicator. The safety redundancy corresponding to periodic surges is 15% of the cumulative heat release demand of the surge segment, and the safety redundancy corresponding to occasional surges is 20% to cope with sudden demands that are difficult to predict accurately. The upper limit of the thermal storage capacity boundary is 90% of the rated thermal storage capacity of the thermal storage device, with 10% reserved as a safety margin for thermal expansion and equipment. When the high-temperature thermal storage medium is close to full capacity, thermal stress concentrates. Forcibly retaining the upper limit safety margin can delay medium aging and container fatigue. When the capacity margin value sequence is consistently positive and the minimum value is more than twice the safety redundancy, the lower limit of the thermal storage capacity boundary can be appropriately lowered to expand the effective operating range of the thermal storage device. Expanding the operating range means that more waste heat can be added during periods of high supply without triggering overflow limits. For load surge warning indicators, the lower limit of the thermal storage capacity boundary for periods corresponding to high-confidence periodic surges is locally raised. The local increase is 10% of the cumulative heat release demand for the corresponding surge segment, ensuring that the thermal storage device has accumulated sufficient heat storage before the surge. The thermal storage capacity boundary is output as a time-level sequence of upper and lower limit values. The global minimum lower limit remains unchanged as the overall safety benchmark. The local increase only applies to the boundary values ​​during the corresponding surge period. Both exist independently in the time-level sequence, and the boundary values ​​at each moment of the sequence are dynamically adjusted according to the distribution of the load surge warning indicators.

[0050] In some embodiments, the step of extracting the waste heat supply confidence interval based on the heat supply coordination parameters to generate the heat supply time window locking parameters includes: performing waste heat supply fluctuation probability distribution analysis on the heat supply coordination parameters to generate a confidence distribution map; extracting high-confidence continuous time periods from the confidence distribution map to generate a supply confidence interval; performing time window boundary positioning based on the supply confidence interval to generate time window boundary parameters; and generating the heat supply time window locking parameters based on the time window boundary parameters and the supply confidence interval.

[0051] A confidence distribution map is generated by analyzing the probability distribution of waste heat supply fluctuations in the heat charging coordination parameters. The waste heat power time series of each heat source in the waste heat state parameter set, combined with the priority information of the corresponding heat sources in the heat charging coordination parameters, participate in the supply stability analysis. Stability is measured by the historical frequency of waste heat power exceeding the effective heat charging threshold at each moment; a higher historical frequency indicates stronger reliability of waste heat supply at that moment. The waste heat power time series is estimated using kernel density to fit the probability density distribution of waste heat power for each time period. The high confidence interval of the probability density distribution corresponds to the period when the power remains consistently stable within the effective range. In a dyeing and printing plant with a fixed production schedule, the stenter operates continuously from 8:00 AM to 5:00 PM daily. The historical recurrence rate of the waste heat power time series during this period is close to 1.0, and the probability density distribution of the heat charging coordination parameters for this period is concentrated in the high power interval, indicating extremely high confidence. After 5:00 PM, the equipment gradually shuts down, the power time series distribution drifts to lower values, and the confidence drops rapidly. The confidence scores for each time period are arranged in chronological order to form a confidence distribution map. The confidence distribution map presents the supply reliability profile of each heat source at different times of the day. Among the heat supply coordination parameters, the overall level of the confidence distribution map of high-priority heat sources is higher than that of low-priority heat sources, reflecting that high-priority heat sources usually have a more stable production process background.

[0052] High-confidence consecutive time periods are extracted from the confidence distribution map to generate supply confidence intervals. Consecutive time periods with a confidence value exceeding 0.75 in the confidence distribution map are considered high-confidence time periods. Three or more consecutive high-confidence time periods constitute a high-confidence segment. Adjacent high-confidence segments are merged into one segment when the interval is less than two time periods. A concentration of high-confidence segments in a fixed time period indicates strong regularity in the heat source supply. For example, in a car painting line where the baking process operates on two fixed shifts daily, the confidence distribution map shows two clear high-confidence segments during the morning and evening shifts, while the confidence level is close to zero during other times. Merging these two high-confidence segments forms two independent elements of the supply confidence interval. Each element of the supply confidence interval records the start and end times of the segment and the segment's built-in average confidence value. Segments with higher average confidence values ​​correspond to more stable and reliable waste heat supply within the charging window. If no high-confidence segment is identified in the confidence distribution map, the supply confidence interval is given an empty set, indicating that the overall confidence of the current heat source waste heat supply is insufficient. It is recommended that the heat charging coordination parameters mark the heat source's heat charging plan as low reliability and reduce the heat charging quota.

[0053] The time window boundary parameters are generated based on the supply confidence interval. The start and end times of each element in the supply confidence interval constitute the initial boundary candidates for the charging time window. The time window boundary positioning is based on the initial boundary and is adjusted forward by combining the heat source grade drop rate and the charging response delay of the heat storage device. For heat sources with a fast grade drop rate, the charging action needs to be started while the grade is still high. The initial boundary of the charging time window needs to be moved forward to allow for charging preparation time. The response delay comes from three physical processes: opening of the charging valve, establishment of pipeline hot state, and flow stabilization. The typical response delay is 3 to 8 minutes. The effective start boundary is defined as the sum of the grade drop response time and the charging response delay, shifted forward from the start time of the supply confidence interval. The effective end boundary is defined as the sum of the heat storage accumulation buffer time, shifted backward from the end time. In a certain electric arc furnace steelmaking process, the waste heat power begins to decay at the end of the smelting process. The end boundary of the supply confidence interval is set at the point where the power decays to 50% of the effective charging threshold. The effective end boundary is shifted forward by 2 minutes at this point to ensure that the charging action is completed when the waste heat supply is sufficient. The time window boundary parameters are composed of the time value pairs of the effective start boundary and the effective end boundary of each heat source. Multiple segments in the supply confidence interval correspond to multiple pairs of time window boundary parameter elements. The effective start and end times of the time window boundary parameters determine the actual operable window range of the charging execution plan.

[0054] The charging time window locking parameters are generated based on the time window boundary parameters and the supply confidence interval. The effective start and end times of each heat source in the time window boundary parameters determine the time constraint for charging execution, while the mean confidence level of the corresponding segment in the supply confidence interval determines the reliability of the charging plan within that time window. Together, they determine the locking strength of the charging time window locking parameters. Charging plans corresponding to time windows with a mean confidence level exceeding 0.85 are strongly locked, and any external scheduling command that would cause charging interruption within the locked time window must undergo security verification before execution. Charging plans corresponding to time windows with a mean confidence level between 0.75 and 0.85 are softly locked. Softly locked time windows allow charging to be paused when the upper limit of the heat storage capacity boundary is approached, but the planned charging heat must be replenished within the time window. The charging time window locking parameters store three attributes for each heat source: the start and end times of the time window, the locking strength level, and the average confidence level, indexed by the heat source node number. If the effective time window width in the time window boundary parameters is less than the minimum charging duration for a single cycle, the corresponding heat source's charging time window locking parameters are marked as unexecutable, indicating that the supply window for that heat source in the current operating cycle is too short to complete effective charging. The charging time window locking parameters are updated synchronously with the charging coordination parameters. When a new round of coordination changes the heat source priority order, the locking time window boundary of the corresponding heat source is repositioned accordingly.

[0055] A thermal balance window is determined based on the thermal storage capacity boundary and the charging time window locking parameters to generate a thermal storage scheduling sequence. The thermal storage capacity boundary defines the upper and lower limits of the acceptable heat storage capacity of the thermal storage device, while the charging time window locking parameters define the charging execution time window for reliable waste heat supply from each heat source. The thermal balance window determination determines the final charging execution period for each heat source node within the intersection of these two constraints. Charging is prioritized when the lower limit of the thermal storage capacity boundary is approaching to prevent the heat storage capacity from falling below the safety threshold. The charging priority during such periods overrides the locking window limit of the charging time window parameters, allowing temporary extension of the window boundary to initiate emergency charging even if the corresponding heat source is currently at the outer edge of the locking window. In northern winter heating networks, the demand for heating surges during the early morning when temperatures drop sharply, and the heat storage capacity of the thermal storage device is rapidly depleted, approaching the lower limit of the thermal storage capacity boundary. At this time, even if the waste heat from the production line is in a period of low supply stability, a charging response must be initiated to maintain heating continuity. During periods approaching the upper limit of the thermal storage capacity boundary, charging triggering is suppressed. Charging execution entries within the locked window of the charging time window lock parameter are postponed until the stored heat capacity drops below 80% of the upper limit. After the heat balance window is verified, the final charging execution period, planned charging power value, and upper limit of charging duration for each heat source node are arranged chronologically to form a thermal storage scheduling sequence. This sequence covers all moments within the current 24-hour analysis window. Each moment carries a list of active heat sources and a corresponding charging power allocation scheme. The sum of the charging power of the active heat sources at each moment in the thermal storage scheduling sequence does not exceed the upper limit of the rated charging power of the thermal storage device. The thermal storage scheduling sequence is re-verified when either the thermal storage capacity boundary or the charging time window lock parameter is updated.

[0056] Step S150: Based on the correlation calculation between the thermal storage allocation parameters and the thermal storage capacity boundary, determine the priority weight of thermal energy allocation, and use the thermal energy allocation priority weight combined with the thermal storage scheduling sequence to output the waste heat storage execution command.

[0057] In some embodiments, the step of determining the priority weight for heat energy allocation based on the correlation calculation between the heat storage allocation parameters and the heat storage capacity boundary includes: performing capacity margin amplitude distribution statistics on the heat storage capacity boundary to generate a capacity margin distribution interval; performing low-margin dense segment clustering extraction on the capacity margin distribution interval to generate high-risk capacity identifiers; performing margin recovery probability assessment on the heat storage allocation parameters based on the high-risk capacity identifiers to generate recovery probability weights; and performing risk adaptive calculation on the heat storage allocation parameters based on the recovery probability weights to determine the priority weight for heat energy allocation.

[0058] Statistical analysis of the amplitude distribution of the capacity margin is carried out for the heat storage capacity boundary to generate the capacity margin distribution interval. The upper and lower limit time series of the heat storage capacity boundary are subtracted from the total heat storage time series of the regenerative heat storage device in the waste heat state parameter set. Each element of the difference series at each moment corresponds to the upper limit margin and the lower limit margin respectively. The smaller the lower limit margin, the closer the heat storage amount is to the safety bottom line. When it remains low for a long time, it indicates that the regenerative heat storage device has been operating at a low heat storage level for a long time, and there is a potential risk of interruption in heat supply continuity. The historical distribution of the lower limit margin time series is statistically analyzed by a histogram. The horizontal axis of the histogram is the numerical interval of the lower limit margin, and the vertical axis is the count of the corresponding time period. The peak interval of the histogram reflects the heat storage level at which the regenerative heat storage device most frequently operates. If the peak is located in the low margin interval, it indicates that the current charging plan is insufficient to supplement the heat storage amount of the regenerative heat storage device. The capacity margin distribution interval represents the probability distribution of the margin amplitude by the count ratio of each interval in the histogram. The margin interval with a high count ratio corresponds to the heat storage level at which the regenerative heat storage device frequently stays during historical operation. For the regenerative heat storage device supporting the glass furnace, during continuous drawing production, the waste heat supply is abundant, and the heat storage amount is maintained at a medium to high level for a long time, so the count ratio of the medium to high lower limit margin interval in the capacity margin distribution interval is high; during the cold repair and shutdown period of the furnace, the waste heat supply is interrupted, and the heating demand on the heat consumption side continues, so the heat storage amount is quickly consumed, and the count ratio of the low to medium lower limit margin interval in the capacity margin distribution interval rises concentratedly during this period. The morphological differences in the capacity margin distribution intervals between the two states before and after shutdown are obvious, providing a clear statistical stratification basis for the subsequent identification of the low margin dense section. The longer the time period covered by the heat storage capacity boundary, the more stable the statistical analysis of the capacity margin distribution interval, and the clearer the influence of the seasonal operation law on the peak position of the capacity margin distribution interval. The capacity margin distribution interval is re-statistically analyzed synchronously with the update of the heat storage capacity boundary.

[0059] High-risk capacity identifiers are generated by clustering low-margin dense segments within the capacity margin distribution range. A low-margin range is defined as an interval where the lower limit margin is 30% below the distribution mean. Periods where low-margin ranges occur frequently on the time axis indicate that the thermal storage device repeatedly faces the risk of its stored heat capacity approaching its lower limit during those periods. After extracting the corresponding times for low-margin ranges, K-Means clustering is performed. The number of clusters is adaptively determined based on the temporal distribution characteristics of the low-margin times, using the principle of maximizing the silhouette coefficient. The candidate range for the number of clusters is 2 to 5. Before clustering, the timestamps of the low-margin times are normalized to ensure that the impact of time distance on the clustering results is consistent with the margin amplitude. After clustering, each cluster corresponds to a typical low-margin occurrence pattern. For example, in a foundry, after the continuous casting process, the stored heat capacity drops sharply during the concentrated heat release phase. Within 1 to 2 hours after casting, the lower limit margin of the thermal storage device repeatedly falls to a low level. During this period, low-margin times are highly concentrated on the time axis, forming an independent low-margin time cluster after clustering. The high-risk capacity identifier is composed of three pieces of information: the time range of each cluster, the density of low margin moments within the cluster, and the average lower limit margin within the cluster. Clusters with higher low margin moment densities and lower average lower limit margins correspond to periods with higher risk levels in the high-risk capacity identifier. When multiple high-risk periods occur together within the same operating cycle, the overall risk rating of the high-risk capacity identifier is upgraded, indicating that the current thermal charging plan is insufficient to guarantee the maintenance of thermal storage capacity. When there are no low margin intervals in the capacity margin distribution range, the high-risk capacity identifier is assigned an empty set, indicating that the thermal storage device has sufficient thermal storage capacity and low capacity risk within the current operating cycle, and thermal energy allocation does not require special protection for specific high-risk periods.

[0060] Based on high-risk capacity identification, a margin recovery probability assessment is performed on the thermal storage allocation parameters to generate recovery probability weights. Each risk period in the high-risk capacity identification reflects the frequent approach of the thermal storage device's stored heat capacity to the lower limit during that period. The margin recovery probability assessment quantifies the likelihood that the charging plans of each heat source in the thermal storage allocation parameters can bring the stored heat capacity back to a safe level before the arrival of the high-risk period. The upper limit of charging power and the charging duration constraint of each heat source before the high-risk period in the thermal storage allocation parameters determine the expected replenishable stored heat capacity. The ratio of the expected replenishment amount to the lower limit margin gap serves as the basic assessment value for the margin recovery probability. A ratio greater than 1.0 indicates that the charging plan can theoretically completely compensate for the margin gap; a ratio less than 0.5 indicates that even if the charging plan is fully executed, it may still be impossible to restore the stored heat capacity to a safe level. The charging duration constraint for such heat source nodes needs to be further tightened in conjunction with the risk level of the high-risk capacity identification. Historical data shows that the actual execution rate of the heat source's charging plan before similar high-risk periods is used as a correction coefficient. The actual execution rate is determined by dividing the number of times the charging plan was completed in the historical version of the heat storage allocation parameters by the total number of planned executions. Heat source nodes with low execution rates are usually constrained by the intermittency of waste heat supply or changes in production scheduling. The basic assessment value of the margin recovery probability is adjusted downwards after being multiplied by the execution rate correction. The recovery probability weight is composed of the corrected margin recovery probability of each heat source node before each high-risk period. The higher the recovery probability weight value, the stronger the margin recovery capability of the heat source's charging plan before the corresponding high-risk period. When the high-risk capacity is marked as an empty set, all recovery probability weights are assigned a value of 1.0, indicating that there is no need to adjust the charging priority for specific risk periods, and each heat source participates in heat energy allocation according to the normal priority order.

[0061] Based on the recovery probability weights, risk-adaptive calculations are performed on the thermal storage allocation parameters to determine the priority weights for heat energy allocation. Heat source nodes with higher recovery probability weights contribute significantly to the recovery of heat storage capacity before high-risk periods and should therefore receive higher priority weights in resource competition. This priority increase ensures that the heat source's charging plan is not preempted by lower-priority nodes when sharing charging channels with other heat sources. The risk-adaptive calculation normalizes the upper limit of the charging power of each heat source node in the thermal storage allocation parameters and multiplies it by the corresponding recovery probability weight. The product reflects the effective charging contribution intensity of the heat source under the current risk context; the higher the effective charging contribution intensity, the higher the heat energy allocation priority score. The comprehensive score of the heat energy allocation priority also incorporates a real-time feedback term on the current lower limit margin of the thermal storage capacity boundary. The smaller the lower limit margin, the higher the weight of the feedback term, driving the heat energy allocation priority to concentrate on the heat source nodes with the largest contribution to margin recovery. This prevents the allocation of charging channels with equal priority during critical periods of heat storage capacity, which would lead to low recovery efficiency. The comprehensive score formula is W_i = λ × R_recover_i × P_norm_i + (1-λ) × M_margin, where W_i is the priority weight for heat energy allocation of the i-th heat source, R_recover_i is the corresponding recovery probability weight, P_norm_i is the normalized upper limit of charging power, M_margin is the normalized inverse value of the current lower limit margin (1 minus the current normalized lower limit margin value; the smaller the lower limit margin, the larger M_margin), and λ is a tradeoff coefficient dynamically adjusted according to the tightness of the lower limit margin; the lower the lower limit margin, the smaller λ is to increase the weight of the margin feedback term. The heat energy allocation priority scores of all heat source nodes are summarized by time period to form a priority weight matrix. The rows of the matrix correspond to heat source nodes, and the columns correspond to time periods. The node with the highest score in the matrix obtains the optimal charging channel allocation right in the corresponding time period. When there are multiple high-scoring nodes in the heat storage allocation parameters at the same time period and the charging channels are limited, they are allocated in descending order of score to ensure that the most urgent charging demand is responded to first.

[0062] The waste heat storage execution command is output by combining the heat energy allocation priority weight with the heat storage scheduling sequence. The list of activated heat sources and the charging power allocation scheme at each moment in the heat storage scheduling sequence provide the basic framework for charging execution. Based on this, the heat energy allocation priority weight makes the final adjustment of the execution order and power allocation ratio of the activated heat sources at each moment. At a certain moment, three heat sources are activated in the heat storage scheduling sequence, but the heat storage device currently has only two charging channels remaining. The two heat sources with the highest heat energy allocation priority weight are activated for charging first, and the third heat source is delayed until the next moment when the charging channel becomes available. The delay time is dynamically determined based on the score difference of the heat energy allocation priority weight. The smaller the score difference, the shorter the delay time to avoid low-priority heat sources being shelved for a long time. The roller kiln in the ceramics plant resumes ignition and preheating ahead of the night shift shutdown and maintenance window. The waste heat power rapidly increases from zero. In the heat storage scheduling sequence, the roller kiln node switches from standby to active at this moment. Based on the current low heat storage capacity boundary, the roller kiln is determined to be the highest priority node for heat energy allocation. The corresponding waste heat storage execution command immediately issues a charging start command and allocates the maximum allowable charging power. Waste heat storage execution commands are generated sequentially using the equipment node number as an index. Each command includes four execution parameters: charging start time, target charging power, continuous charging duration, and safety interruption conditions. Safety interruption conditions include three trigger scenarios: heat storage reaching the upper limit of the heat storage capacity boundary, waste heat power falling below the minimum effective charging threshold, and equipment protection actions. Heat source nodes whose heat energy allocation priority score is less than 50% of the current time period average of the priority value matrix will not generate waste heat storage execution instructions. The corresponding node's heat charging action will be postponed until the heat energy allocation priority value recovers and is re-evaluated. The waste heat storage execution instruction list is arranged in ascending order of execution time, with the highest priority execution instruction placed at the top of the list to ensure that critical heat charging actions are responded to first.

[0063] To implement the above-described method embodiments, a large-scale AI model-based industrial waste heat storage scheduling optimization method is proposed to achieve the corresponding functionalities and technical effects. See also... Figure 2 , Figure 2 This diagram illustrates a structural block diagram of an industrial waste heat storage scheduling optimization system 200 based on an AI large-scale model, as provided in an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The industrial waste heat storage scheduling optimization system 200 based on an AI large-scale model provided in this application includes: Data acquisition module 201 is used to acquire multi-source waste heat data and heat storage medium operation data of industrial processes, and to perform time-series synchronization of the multi-source waste heat data and the heat storage medium operation data and thermal parameter analysis to generate a waste heat state parameter set. Grade analysis module 202 is used to divide the heat source grade range according to the waste heat state parameter set, extract the grade critical drop feature, analyze the heat flux density change based on the waste heat state parameter set to generate the heat storage capacity matching factor, and construct the heat charging start-up judgment table based on the heat storage capacity matching factor and the grade critical drop feature. The conflict coordination module 203 is used to input the waste heat state parameter set into the AI ​​large model to identify the superposition characteristics of multi-source heat flow and generate heat charging coordination parameters, and dynamically adjust the heat charging start-up judgment table based on the heat charging coordination parameters to generate heat storage allocation parameters. The scheduling optimization module 204 is used to perform heat load timing verification on the heat storage allocation parameters to identify the heat storage capacity boundary, extract the waste heat supply confidence interval based on the heat charging coordination parameters to generate heat charging time window locking parameters, and perform heat balance window verification based on the heat storage capacity boundary and the heat charging time window locking parameters to generate a heat storage scheduling sequence. The instruction output module 205 is used to perform correlation calculations based on the heat storage allocation parameters and the heat storage capacity boundary to determine the priority value of heat energy allocation, and to output the waste heat storage execution instruction using the priority value of heat energy allocation combined with the heat storage scheduling sequence.

[0064] The aforementioned AI-based large-scale model-based industrial waste heat storage scheduling optimization system 200 can implement the AI-based large-scale model-based industrial waste heat storage scheduling optimization method described in the above-described method embodiments. Optional embodiments described above are not detailed here. The remaining content of this application's embodiments can be referred to the content of the above-described method embodiments, and will not be repeated in this embodiment.

[0065] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

[0066] The above embodiments are not an exhaustive list based on the present invention, and there may be many other embodiments not listed. Any substitutions and improvements made without departing from the concept of the present invention are within the protection scope of the present invention.

Claims

1. A method for optimizing the scheduling of industrial waste heat storage based on an AI large-scale model, characterized in that, include: Collect multi-source waste heat data and heat storage medium operation data of industrial processes, and perform time-series synchronization and thermal parameter analysis on the multi-source waste heat data and the heat storage medium operation data to generate a waste heat state parameter set. Based on the waste heat state parameter set, the heat source grade range is divided and the grade critical drop feature is extracted. Based on the waste heat state parameter set, the heat flux density variation is analyzed to generate the heat storage capacity matching factor. Based on the heat storage capacity matching factor and the grade critical drop feature, a heat charging start-up judgment table is constructed. The waste heat state parameter set is input into the AI ​​large model to identify the superposition characteristics of multi-source heat flow and generate charging coordination parameters. Based on the charging coordination parameters, the charging start-up judgment table is dynamically adjusted to generate heat storage allocation parameters. The heat storage allocation parameters are checked for heat load time sequence to identify the heat storage capacity boundary. Based on the heat charging coordination parameters, the waste heat supply confidence interval is extracted to generate the heat charging time window locking parameters. Based on the heat storage capacity boundary and the heat charging time window locking parameters, the heat balance window is verified to generate the heat storage scheduling sequence. Based on the correlation calculation between the heat storage allocation parameters and the heat storage capacity boundary, the priority value of heat energy allocation is determined, and the waste heat storage execution command is output by combining the priority value of heat energy allocation with the heat storage scheduling sequence.

2. The method according to claim 1, characterized in that, The step of generating a heat storage capacity matching factor based on the heat flux density variation analysis of the waste heat state parameter set includes: The waste heat state parameter set is divided into grade distribution maps by stepping intervals according to the heat source temperature range; Heat flux density values ​​for each grade interval are extracted from the grade distribution map to generate a heat flux density sequence; The heat flux density sequence is subjected to trend analysis to generate a trend decay coefficient; The heat storage capacity matching factor is generated by adaptive matching calculation based on the changing attenuation coefficient and the heat flux density sequence.

3. The method according to claim 1, characterized in that, The method of constructing a charging start-up determination table based on the heat storage capacity matching factor and the grade critical drop characteristics includes: A critical drop amplitude set is generated by jointly evaluating the drop amplitude and drop rate of the aforementioned critical drop characteristics. From the set of critical drop amplitudes, identify sections with steeply increasing drop gradients and generate drop warning indicators; Based on the drop warning identifier and the heat storage capacity matching factor, the heat storage capacity limit calibration is performed to generate a heat storage trigger threshold group by adaptively adjusting the drop rate. A heat charging start determination table is constructed based on the aforementioned heat charging trigger threshold group.

4. The method according to claim 1, characterized in that, The step of inputting the waste heat state parameter set into the AI ​​large model to identify the superposition characteristics of multi-source heat flow and generate heat-combining parameters includes: The waste heat state parameter set is input into the AI ​​large model and multi-heat source time-series alignment is performed to generate a heat source time-series matrix; The heat flux intensity distribution of each heat source is extracted from the heat source time series matrix and superimposed to generate a multi-source heat flux superposition feature. Based on the multi-source heat flow superposition characteristics and the heat source time series matrix, conflict marker groups are generated to identify overlapping time window segments. Heat charging coordination parameters are generated based on the conflict marker group and the superposition characteristics of the multi-source heat flow.

5. The method according to claim 1, characterized in that, The step of dynamically adjusting the heat storage allocation parameters based on the heat charging coordination parameters to generate the heat charging start-up judgment table includes: The urgency ranking sequence is generated by quantifying and sorting the heat charging coordination parameters. Heat charging quotas for each heat source are extracted from the urgency sorting sequence to generate heat charging quota groups; Based on the aforementioned heat charging quota group, the triggering conditions in the heat charging start-up decision table are modified to generate an adjustment decision table. Based on the adjustment judgment table and the heat storage allocation group, heat storage allocation parameters are generated.

6. The method according to claim 1, characterized in that, The step of performing heat load time-series verification and identifying heat storage capacity boundaries for the heat storage allocation parameters includes: The heat load demand for each time period is extracted from the heat storage and allocation parameters to generate a heat load demand sequence; The load surge warning features are identified from the heat load demand sequence to generate a load surge warning identifier; Based on the load surge warning indicator, the capacity margin is calculated for the heat load demand sequence to generate a capacity margin value; The thermal storage capacity boundary is determined based on the capacity margin value and the load surge warning indicator.

7. The method according to claim 1, characterized in that, The step of extracting the waste heat supply confidence interval based on the heat charging coordination parameters and generating the heat charging time window locking parameters includes: The waste heat supply fluctuation probability distribution of the aforementioned heat supply coordination parameters is analyzed to generate a confidence distribution map; High-confidence continuous time periods are extracted from the confidence distribution map to generate supply confidence intervals; Time window boundary parameters are generated based on the supply confidence interval for time window boundary positioning. The charging time window locking parameters are generated based on the time window boundary parameters and the supply confidence interval.

8. The method according to claim 1, characterized in that, The step of determining the priority value for heat energy allocation based on the correlation calculation between the heat storage allocation parameters and the heat storage capacity boundary includes: The capacity margin amplitude distribution of the heat storage capacity boundary is statistically analyzed to generate a capacity margin distribution interval; The high-risk capacity identifier is generated by clustering low-margin dense segments within the capacity margin distribution interval. Based on the high-risk capacity identifier, a margin recovery probability assessment is performed on the thermal storage allocation parameters to generate recovery probability weights; Based on the recovery probability weight, risk adaptive calculation is performed on the thermal energy allocation parameters to determine the priority weight for thermal energy allocation.

9. The method according to claim 4, characterized in that, The step of generating conflict marker groups based on the multi-source heat flux superposition characteristics and the heat source time series matrix to identify overlapping time windows includes: A sliding time window scan is performed on the trigger time of each heat source in the heat source time series matrix to generate a time window density distribution map; Extract multi-source triggered dense segments from the time window density distribution map to generate a dense area identifier set; The capacity-weighted conflict intensity sequence is generated by combining the dense area identifier set with the multi-source heat flow superposition characteristics. Based on the capacity-weighted conflict intensity sequence, the overlapping segments of the time window are identified to generate conflict marker groups.

10. An industrial waste heat storage scheduling and optimization system based on an AI large model, characterized in that, include: The data acquisition module is used to collect multi-source waste heat data and heat storage medium operation data of industrial processes, and to perform time-series synchronization of the multi-source waste heat data and the heat storage medium operation data and analyze thermal parameters to generate a waste heat state parameter set. The grade analysis module is used to divide the heat source grade range according to the waste heat state parameter set, extract the grade critical drop characteristics, analyze the heat flux density variation based on the waste heat state parameter set to generate a heat storage capacity matching factor, and construct a heat charging start-up judgment table based on the heat storage capacity matching factor and the grade critical drop characteristics. The conflict coordination module is used to input the waste heat state parameter set into the AI ​​large model to identify the superposition characteristics of multi-source heat flow and generate heat charging coordination parameters, and dynamically adjust the heat charging start-up judgment table based on the heat charging coordination parameters to generate heat storage allocation parameters. The scheduling optimization module is used to perform heat load timing verification on the heat storage allocation parameters to identify the heat storage capacity boundary, extract the waste heat supply confidence interval based on the heat charging coordination parameters to generate heat charging time window locking parameters, and perform heat balance window verification based on the heat storage capacity boundary and the heat charging time window locking parameters to generate a heat storage scheduling sequence. The instruction output module is used to perform correlation calculations based on the heat storage allocation parameters and the heat storage capacity boundary to determine the priority value of heat energy allocation, and to output the waste heat storage execution instruction using the priority value of heat energy allocation combined with the heat storage scheduling sequence.

Images