Heating complementary control method and system based on AI technology

Abstract

The application belongs to the technical field of heating control, and relates to a heating complementary control method and system based on AI technology, which trains a deviating prediction relationship by identifying a historical period in which a heat pump heating performance deviates from a carbon emission factor direction; predicts a future deviating period and a deviation degree based on current environmental conditions; when a deviating period is predicted, determines a heat pump heating power reduction ratio and a heat release power compensation ratio of a heat storage device according to the deviation degree and a heat storage state; runs at a compensation power before the deviation and guides surplus heat into the heat storage device; reduces the heat pump power during the deviating period and compensates for the shortage by releasing heat from the heat storage device, and supplements the shortage by an auxiliary heat source when the heat storage device is insufficient; and restores normal heating and charges the heat storage device after the deviation ends. The application introduces the carbon emission factor into the scheduling decision, realizes complementary control of the heat pump and the heat storage device in the time and power dimensions, and reduces the dependence on the auxiliary heat source during the deviating period.

Description

Technical Field

[0001] This invention belongs to the field of heating control technology, and relates to a heating complementary control method and system based on AI technology. Background Technology

[0002] In heat pump heating scenarios, the instantaneous heating efficiency of the heat pump fluctuates dynamically with changes in outdoor environmental conditions. Simultaneously, the carbon emission factor of the regional power grid also varies over time, exhibiting peak-valley alternation characteristics within the power dispatch cycle. When the period of high heat pump heating efficiency coincides with the period of rising grid carbon emission factor, the efficient operation of the heat pump will come at the cost of higher carbon emissions per unit of heating, creating a conflict between the goals of energy efficiency improvement and carbon emission reduction.

[0003] In existing technologies, the coordinated control of heat pumps and thermal storage devices typically employs the following methods: using time-of-use electricity prices as a guiding signal to store heat during off-peak hours and release heat during peak hours; using outdoor temperature or heating load as a trigger threshold to activate the thermal storage device to supplement heating when the environment deteriorates or the load increases; and using the coefficient of performance of the heat pump as an optimization target to adjust the heat pump to maintain maximum efficiency operation, with any shortfall being supplemented by thermal storage or auxiliary heat sources.

[0004] The aforementioned existing technologies have at least the following shortcomings: 1. The scheduling logic of the existing control methods does not take into account the temporal changes of the grid carbon emission factor, and cannot identify the time overlap between the rising period of heat pump heating efficiency and the rising period of grid carbon emission factor. When the heat pump heating performance and the changing direction of grid carbon emission factor deviate, it is impossible to predict and adjust the operation strategy before the deviation occurs, resulting in the heat pump maintaining high power consumption during high carbon emission periods, causing an unnecessary increase in the carbon footprint of heating.

[0005] 2. The execution conditions of existing dispatching strategies, whether it's electricity price range switching or environmental threshold triggering, are all reactive control logics, executing dispatching actions only after the relevant signals have changed. This lag means that existing technologies lack the time margin to store heat in advance before a divergence trend forms, and during the divergence period, they can only rely on high-carbon-emission auxiliary heat sources to fill the heating gap.

[0006] 3. Existing technologies drive optimization and control based on a single indicator of heat pump heating efficiency. During periods of rising carbon emission factors, this will still push heat pumps to increase output to compensate for the heating loss caused by decreased efficiency, exacerbating the conflict between energy efficiency improvement and carbon emission reduction goals. Summary of the Invention

[0007] In view of this, in order to solve the problems mentioned in the background technology, a heating complementary control method and system based on AI technology is proposed.

[0008] The objective of this invention can be achieved through the following technical solutions: The first embodiment of this invention provides a heating-complementary control method based on AI technology, including: identifying the time period in the historical operation range of the heat pump where the heating performance and the direction of change of carbon emission factors deviate, extracting the environmental condition sequence and carbon emission factor sequence corresponding to the time period of deviation, and training the deviation prediction relationship based on a mapping network constructed by recursive quantile partitioning.

[0009] Collect current environmental conditions and current carbon emission factors, and use the deviation prediction relationship to output the prediction results and the degree of deviation for the directional deviation period within a preset future time period.

[0010] When a period of directional deviation is predicted, the reduction ratio of the heating power of the heat pump and the compensation ratio of the heat release power of the heat storage device are determined based on the degree of deviation and the status of the heat storage device during the period of directional deviation.

[0011] In the preparatory stage before entering the directional divergence period, the heat pump is controlled to operate at a compensating power level, and the surplus heat exceeding the real-time heating load is introduced into the heat storage device.

[0012] When entering the period of directional deviation, the heating power of the heat pump is reduced according to the proportion of the decrease in heating power, and the heat storage device is controlled to release heat to compensate for the deficiency according to the heat release power compensation ratio. When the heat storage device has insufficient heat release capacity, it is supplemented by the auxiliary heating branch.

[0013] When the period of directional deviation ends, the control heat pump resumes normal heating and performs recharging of the heat storage device.

[0014] The second embodiment of the present invention provides a heating-complementary control system based on AI technology, including: a prediction training module, a prediction output module, a power matching determination module, a heat storage compensation preparation module, a heat storage compensation execution module, and a recovery and recharge control module.

[0015] The prediction training module is connected to the prediction output module, the prediction output module is connected to the power matching determination module, the power matching determination module is connected to the heat storage compensation preparation module, the heat storage compensation preparation module is connected to the heat storage compensation execution module, and the heat storage compensation execution module is connected to the recovery and recharge control module.

[0016] The prediction training module identifies periods within the historical operating range of the heat pump where the heating performance and carbon emission factor change directions diverge. It extracts the environmental condition sequence and carbon emission factor sequence corresponding to the divergent periods and trains the divergence prediction relationship based on a mapping network constructed by recursive quantile partitioning.

[0017] The prediction output module collects current environmental conditions and current carbon emission factors, and uses the deviation prediction relationship to output the prediction results and the degree of deviation for the directional deviation period within a preset future time period.

[0018] The power matching determination module determines the reduction ratio of the heat pump's heating power and the compensation ratio of the heat storage device's heat release power during the predicted period of directional deviation, based on the degree of deviation and the status of the heat storage device.

[0019] The heat storage compensation preparation module controls the heat pump to operate at compensating power during the preparation stage before entering the directional deviation period, and introduces the surplus heat exceeding the real-time heating load into the heat storage device.

[0020] The heat storage compensation execution module reduces the heat pump's heating power according to the ratio of the decrease in heating power when entering the period of directional deviation. At the same time, it controls the heat storage device to release heat to compensate for the deficiency according to the heat release power compensation ratio. When the heat storage device's heat release capacity is insufficient, it is supplemented by the auxiliary heating branch.

[0021] When the recharge control module is restored and the period of directional deviation ends, the heat pump is controlled to resume normal heating and the heat storage device is recharged.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention identifies the time periods in the historical operation range of the heat pump where the transient heating coefficient and carbon emission factor change direction deviate, extracts the environmental condition sequence and carbon emission factor sequence corresponding to the deviating time period, and trains the deviating prediction relationship. Under the current environmental conditions and carbon emission factor input, it outputs the prediction result of whether a deviating time period may occur in the future preset time period. The grid carbon emission factor is directly incorporated into the scheduling decision-making process, providing a preliminary judgment basis for subsequent operation adjustments, overcoming the defects of the prior art that only uses electricity price or environmental threshold as triggering conditions and cannot predict the timing of the deviating.

[0023] (2) After predicting the existence of a period of directional deviation, the present invention controls the heat pump to operate at a compensating power during the preparation stage and introduces the surplus heat exceeding the real-time heating load into the heat storage device; when the deviation period arrives, the heat pump heating power is reduced according to the ratio of the decrease in heating power, and the heat storage device is controlled to release heat to compensate for the deficit according to the ratio of the heat release power compensation. It can store alternative heat before the deviation period arrives, and the heat storage device releases the stored heat to fill the heating deficit during the deviation period, reducing the dependence on emergency supplementary heat sources.

[0024] (3) This invention determines the reduction ratio of heating power and the compensation ratio of heat release power based on the degree of deviation and the status of the heat storage device. The ratio of the maximum allowable heat release power of the heat storage device to the baseline power of the heating load demand is used as the upper limit ratio of substitution. When the heating power deficit ratio exceeds the upper limit ratio of substitution, the reduction ratio of heating power is limited to the upper limit ratio of substitution and the auxiliary heating branch is activated to make up the difference. This constraint ensures that the reduction in the heating power of the heat pump never exceeds the compensable range of the heat storage device, guarantees the matching relationship between the reduction in heating power and substitution in complementary control, and avoids the expansion of the heating gap caused by excessive reduction in heating power.

[0025] (4) When the directional divergence period ends, the present invention controls the heat pump to resume normal heating and performs recharging on the heat storage device. The total heat released by the heat storage device during the divergence period is counted, and the heat is replenished to the heat storage device with additional recharging power within the allowed recharging time window. This recharging mechanism enables the heat storage device to restore its energy storage level after a divergence event ends, providing usable energy storage reserves to cope with the next possible directional divergence period, forming a continuous energy storage cycle between adjacent divergence events. Attached Figure Description

[0026] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a flowchart illustrating the implementation steps of the method of the present invention;

[0028] Figure 2 This is a schematic diagram of the identification logic of the present invention for periods in the historical operating range of heat pumps where the heating performance and the direction of change of carbon emission factors deviate from each other.

[0029] Figure 3 This is a schematic diagram of the module connection of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Please see Figure 1 As shown, the first embodiment of the present invention provides a heating-complementary control method based on AI technology, including: S1. Identifying the time periods in the historical operation range of the heat pump where the heating performance and the direction of change of carbon emission factors deviate, extracting the environmental condition sequence and carbon emission factor sequence corresponding to the time periods of deviation, and training the deviation prediction relationship.

[0032] Specifically, prior to step S1, the following data acquisition process occurs: the control unit retrieves the transient heating coefficient sequence of the heat pump within the historical operating range (e.g., 30 consecutive complete heating days), the synchronously recorded environmental condition sequence, and the regional power grid carbon emission factor sequence from its own operation log database or the historical data storage area of ​​the upper-level monitoring system.

[0033] The transient heating coefficient sequence uses the transient heating coefficient of the heat pump as the primary data source. The transient heating coefficient is calculated in real time from the ratio of the heat exchange on the evaporator side of the heat pump to the input electrical power of the compressor. When the evaporator side heat exchange sensor is not configured or is in a faulty state, the heat output per unit time is used instead. The heat output per unit time is obtained by multiplying the supply water temperature difference on the condenser side by the circulating water flow rate. Once the data source is selected, it remains consistent throughout the entire training and prediction process without being switched midway.

[0034] The environmental condition sequence includes an outdoor temperature sequence and a relative humidity sequence, which are taken from temperature and humidity sensors deployed on the outdoor unit of the heat pump or on the periphery of the building, and recorded synchronously after being aligned with the same timestamp as the transient heating coefficient.

[0035] The regional power grid carbon emission factor sequence is taken from the node marginal carbon emission factors published by the power dispatching agency at a fixed release period (e.g., 15 minutes or 1 hour). When the node marginal carbon emission factor is not covered in the region, the average carbon emission factor of the region is used instead, and the data source type is marked in the sequence.

[0036] Based on this, refer to Figure 2 As shown, the identification of the time period in which the heating performance and carbon emission factor change direction deviate within the historical operating range of the heat pump includes: applying a sliding time window to the transient heating coefficient sequence and carbon emission factor sequence of the historical operating range respectively, and extracting continuous segments.

[0037] The width of the sliding time window is determined based on the release cycle of the carbon emission factor sequence: the interval carbon emission factor sequence is released by the power dispatching agency at a fixed cycle, which is read from the power information disclosure interface during method deployment initialization. The width of the sliding time window is then taken as an integer multiple of the release cycle, preferably 3 to 6, so that each segment contains at least 3 carbon emission factor data points, ensuring sufficient sample size within the segment to support reliable extraction of the direction of change. Additionally, when the sampling cycle of the transient heating coefficient is shorter than the carbon emission factor release cycle, the heating performance sequence is resampled to align with the start time of the release cycle, ensuring consistency in the temporal granularity between the two sequences.

[0038] Within each sub-segment, a pruned least-squares regression is performed using a trend extraction operator to obtain a trend scalar. This pruned least-squares regression is based on least-squares regression and iteratively removes outliers and refits the data to make the fitting results insensitive to outliers within the sub-segment. The specific process is as follows: an initial linear fit is performed on the sequence points within the current sub-segment. The fitting method is least-squares regression. The slope and intercept of the fitted line are solved with the goal of minimizing the sum of the absolute values ​​of the fitting residuals. The slope of the fitted line is taken as the trend scalar. A positive slope indicates an upward trend, and a negative slope indicates a downward trend. The absolute value of the slope characterizes the strength of the trend.

[0039] The residuals of each sequence point relative to the fitted line are obtained by subtracting the actual value from the corresponding value on the fitted line and taking the absolute value. The standard deviation of the residuals is then calculated.

[0040] Outliers with residual absolute values ​​exceeding a preset multiple of the residual standard deviation are removed, and the retained sequence points are refitted; the preset multiple is 3.

[0041] The process iteratively performs residual calculation, outlier removal, and refitting until no new outliers are added or the preset iteration limit is reached. The preset iteration limit is determined based on the number of sequence points in the sub-segment, specifically taking the logarithm of the number of sequence points in the sub-segment with base 2 and rounded up, and must not be less than 3 iterations. This value is based on the convergence characteristic of outlier removal in pruning least squares regression: after each iteration removes the subset of outliers with the largest residuals, the magnitude by which the fitted line moves closer to the true trend direction decreases exponentially with the number of iterations. Therefore, a logarithmic number of iterations can cover the number of iterations required for convergence.

[0042] The slope of the straight line obtained from the last fitting is used as the final tendency scalar output of the current sub-segment.

[0043] The tendency scalar is compared with the upper and lower bounds determined based on the local fluctuation amplitude of the segment: when the tendency scalar is greater than the upper bound, an upward flag is assigned; when it is less than the lower bound, a downward flag is assigned; and when it is between the upper and lower bounds (including the endpoint values), a direction flag of the immediate preceding segment is assigned.

[0044] After traversing all sub-segments, the heating performance direction marker sequence and the carbon emission factor direction marker sequence are compared point by point. Continuous time intervals with directional differences are merged into directional divergence events. During the merging process, if two adjacent back ion segments are separated only by an isolated beat that restores consistency in direction, and the duration of the isolated beat is less than twice the release cycle of the carbon emission factor sequence, then it is merged together with the two back ion segments before and after it into a continuous divergence event. The basis for this judgment is that when the isolated beat contains less than two carbon emission factor data points, the direction marker of the beat is supported by only a very small number of data points, and its directional reliability is insufficient to constitute an effective confirmation of the interruption of the divergence trend. Merging can avoid the fragmentation of divergence events caused by the discrete sampling characteristics of the carbon emission factor sequence.

[0045] Record the start and end times of the deviation, the degree of deviation, and the environmental condition sequence and carbon emission factor sequence during the deviation period. Among them, the difference between the transient heating coefficient at the end and start times of the deviation is taken as the change in heating performance, and the difference between the carbon emission factor values ​​at the end and start times of the deviation is taken as the change in carbon emission factor. The degree of deviation is equal to the absolute value of the change in heating performance divided by the absolute value of the change in carbon emission factor, which represents the change in heating performance corresponding to a unit change in carbon emission factor.

[0046] It should be noted that the process of determining the upper and lower bounds based on the local fluctuation amplitude of the sub-segment is as follows: obtain all the sequence points that are finally retained by the trimmed least squares regression within the current sub-segment, calculate the fitted value of each sequence point at the corresponding sampling time, and subtract the fitted value from the actual value of each sequence point to obtain the detrended residual sequence.

[0047] Calculate the standard deviation of the detrended residual series, denoted as the residual standard deviation.

[0048] Extract all sampling times within the current sub-segment, calculate the standard deviation of the sampling times, and denot it as the time dispersion. The time dispersion is used to characterize the degree of dispersion of the distribution of each sampling point on the time axis within the sub-segment. The larger the time dispersion, the wider the time range covered by the sampling points and the lower the uncertainty of the slope estimation.

[0049] The slope uncertainty is obtained by dividing the standard deviation of the residual by the product of the time dispersion and the square root of the total number of retained sequence points.

[0050] We take twice the slope uncertainty as the upper bound and the opposite of twice the slope uncertainty as the lower bound.

[0051] In this embodiment, the training of the divergence prediction relationship includes: for each historical divergence event, extracting the environmental condition sequence and carbon emission factor sequence within a preset backtracking width before the event starting point, calculating the multi-order statistical characteristics, trend slope, extreme value difference of the environmental conditions within the preset backtracking width, as well as the mean, cumulative rate of change, and degree of fluctuation of the carbon emission factors, to form a condition attribute vector.

[0052] The process for determining the preset backtracking width is as follows: Select the carbon emission factor sequence within the historical operating interval, calculate the correlation coefficient between the sequence and its own lagged version according to different lag steps, record the lag step corresponding to the first decay of the correlation coefficient to below 0.5, multiply the lag step by the carbon emission factor release cycle to obtain the autocorrelation retention time of the carbon emission factor. This autocorrelation retention time characterizes the longest duration for which the carbon emission factor maintains the same trend of change statistically. The preset backtracking width is taken as the autocorrelation retention time of the carbon emission factor, and is not less than 3 times the carbon emission factor release cycle.

[0053] The result attributes are composed of the lag time and the degree of deviation of the event relative to the current moment.

[0054] A recursive partitioning method is used to establish a mapping network composed of hierarchical condition decision nodes. The execution process includes root node initialization, node partitioning and decision, recursive generation of child nodes, and assignment of leaf node parameters: Root node initialization: Pair the condition attribute vectors of all historical direction deviation events with the result attributes, and assign them to the root node as the training sample set; let the training sample set have a total of Groups of samples, each group containing one A conditional attribute vector and a result attribute consisting of lag duration and degree of bias.

[0055] Node partitioning determination: Check whether the number of samples covered by the current node to be partitioned is lower than the preset minimum number of partitioned samples, and whether the coefficient of variation of the deviation degree of all samples within the node (the standard deviation of the deviation degree within the node divided by its mean) is lower than the preset homogeneity judgment limit; the minimum number of partitioned samples is 3, and the homogeneity judgment limit is a constant between 0.15 and 0.25. The specific value is determined during method initialization as follows: when the number of historical deviation events is not less than 3, calculate the coefficient of variation of the deviation degree sequence of all historical deviation events. If the coefficient of variation falls within the interval [0.15, 0.25], directly take its value. If it is less than 0.15, take 0.15, and if it is greater than 0.25, take 0.25; when the number of historical deviation events is less than 3, directly take 0.2 as the homogeneity judgment limit.

[0056] When the number of samples is less than the minimum number of samples to split, or the coefficient of variation of the bias is less than the homogeneity threshold, the current node to be split is marked as a leaf node and no further splitting is performed; otherwise, the splitting operation is performed: traversing the current node to cover all conditional attribute vectors in the sample. For each dimension, perform a tentative quantile partitioning: sort all samples within a node according to the value of the current dimension in ascending order, and when the number of samples within a node is greater than or equal to... The process involves using a ternary partitioning method. After sorting, the samples are divided into three equal intervals based on their quantity, and the two boundary values ​​between adjacent intervals are recorded. When the number of samples in a node is less than... The quartile partitioning method is used. After sorting, the samples are divided into two equal intervals according to the number of samples, and a boundary value between the intervals is recorded.

[0057] After completing the exploratory quantile division of all dimensions, a partition quality index is calculated for each dimension. The partition quality index is taken as the sample size weighted average of the variance of the degree of deviation within each interval after partitioning.

[0058] Select the dimension that minimizes the quality index, use this dimension as the decision dimension of the current node, and use the corresponding quantile interval boundary value as the comparison rule of the current node, storing them in the decision parameter set of the current node.

[0059] Child node recursively generated: Based on the decision dimension and boundary value selected by the current node, all samples covered by the current node are assigned to various child nodes. For quartiles, samples with values ​​less than or equal to the boundary value on the decision dimension enter the first child node, and those with values ​​greater than the boundary value enter the second child node; for tertiles, samples with values ​​less than or equal to the lower boundary value enter the first child node, those with values ​​greater than the lower boundary value and less than or equal to the upper boundary value enter the second child node, and those with values ​​greater than the upper boundary value enter the third child node.

[0060] Each child node is treated as a new node to be partitioned, and the node partitioning and determination phase is recursively executed until all branch paths terminate with leaf nodes.

[0061] Leaf node parameter assignment: Count all training samples covered by each leaf node, calculate the arithmetic mean of the sample lag time as the predicted deviation time offset, calculate the arithmetic mean of the sample deviation degree as the predicted deviation degree, and store the number of samples covered by the leaf node together with the two predicted values.

[0062] Online step-by-step determination: Input the current conditional attribute vector into the mapping network from the root node. At each non-leaf node, read the determination dimension index and boundary value from the determination parameter set. Extract the value of the current conditional attribute vector in the determination dimension and compare it with the boundary value. Based on the comparison result, proceed to the next level node along the corresponding branch. Repeat the comparison until the leaf node is reached. Read the prediction deviation time period offset and prediction bias degree stored in the leaf node as the prediction output.

[0063] It should be noted that the multi-order statistical characteristics of environmental conditions include: the first-order statistic is the arithmetic mean of the environmental condition sequence within the backtracking width; the second-order statistic is the variance; the third-order statistic is the skewness, characterizing the asymmetry of the sequence distribution; and the fourth-order statistic is the kurtosis, characterizing the steepness of the sequence distribution. The trend slope is the slope obtained by performing linear regression on the environmental condition sequence within the preset backtracking width. Here, the linear regression uses the same trimmed least-squares regression method as the trend extraction operator to ensure that the slope is not contaminated by local perturbations. The extreme value difference is the difference between the maximum and minimum values ​​of the environmental conditions within the backtracking width.

[0064] On the carbon emission factor side, the mean is the arithmetic mean of the carbon emission factor sequence within the backtracking width; the cumulative rate of change is the ratio of the difference between the carbon emission factor at the end time and the start time of the preset backtracking width to the value at the start time; the degree of fluctuation is the standard deviation of the carbon emission factor sequence within the preset backtracking width.

[0065] The above features are concatenated into a conditional attribute vector. Each dimension of the conditional attribute vector corresponds to a feature. The total number of dimensions is equal to the number of environmental condition feature dimensions plus the number of carbon emission factor feature dimensions.

[0066] It should also be noted that when the number of historical deviation events is less than the preset minimum sample size (set to 10), the deviation prediction relationship is temporarily not enabled, and the system defaults to operating in a control mode targeting the transient heating coefficient. The transient heating coefficient of the heat pump is monitored in real time, with compressor frequency and expansion valve opening as control variables. An extreme value search algorithm is used to apply a small disturbance to the current control parameters under the current environmental conditions and observe the direction of change of the transient heating coefficient: if the transient heating coefficient increases, the system continues to step forward by a preset step size along the current disturbance direction; if it decreases, the system steps forward by a preset step size in the opposite direction of the current disturbance direction, until the control parameter value that maximizes the transient heating coefficient is found. The disturbance amplitude corresponding to the small disturbance is specifically 1% of the current value of the control parameter, and the preset step size is twice the control parameter adjustment accuracy. The system will switch to a different mode after accumulating sufficient deviation event samples.

[0067] S2. Collect current environmental conditions and current carbon emission factors, and use the deviation prediction relationship to output the prediction results and degree of deviation for the future preset time period.

[0068] Specifically, current environmental conditions and current carbon emission factors are collected, and the current condition attribute vector is obtained by using the same window width and feature extraction method as the constructed condition attribute vector.

[0069] The current conditional attribute vector is input into the mapping network. After being determined by each branch, it reaches the corresponding leaf node. This leaf node is the final termination node in the mapping network where the current conditional attribute vector finally falls, and is denoted as the target node.

[0070] The output target node stores the predicted deviation time offset as the expected starting offset, the target node stores the arithmetic mean of the duration of deviation events in all training samples covered by the target node as the expected duration, and the target node stores the prediction bias.

[0071] Extract a fixed-length segment from the end of the environmental condition sequence within the current backtracking width. Perform linear fitting on the sequence points within this segment. Use the sign of the fitted slope as the direction of the nearest change trend, and the absolute value of the fitted slope as the current nearest change rate. The length of the fixed-length segment is taken as a fraction of the backtracking width. .

[0072] Extract all historical divergence event samples associated with the target node, perform linear fitting on each sample within the same fixed-length sub-segment at the end of its backtracking width, take the sign of the fitting slope of each sample as the direction of change trend, and count the direction with the highest frequency as the typical trend direction of the target node.

[0073] When the direction of the approaching trend is consistent with the typical trend direction of the target node, a trend strengthening judgment is performed: the absolute values ​​of the slopes obtained by linear fitting of all historical divergence event samples within a fixed-length sub-segment at the end are arranged in ascending order, and the upper third value is taken as the trend strengthening reference value; when the current approaching rate of change is greater than the trend strengthening reference value, the expected starting offset is adjusted by one step size towards the direction closer to the current moment, and the prediction deviation is multiplied by the ratio of the current approaching rate of change to the trend strengthening reference value; the step size is half of the standard deviation of the lag time of all historical divergence event samples.

[0074] If the direction of the nearby trend is inconsistent with the typical trend direction of the target node, no correction will be performed, and the expected starting offset, expected duration, and prediction deviation will be directly output as the prediction result.

[0075] S3. When a period of directional deviation is predicted, the heat pump's heating power reduction ratio and the heat storage device's heat release power compensation ratio are determined based on the degree of deviation and the status of the heat storage device during the period of directional deviation.

[0076] Specifically, the supply water temperature, return water temperature, and circulating water flow rate of the main heating network are obtained. The difference between the supply water temperature and the return water temperature is multiplied by the circulating water flow rate and the specific heat capacity of the water to obtain the real-time heating load.

[0077] A first-order exponential smoothing filter is applied to the real-time heating load, and the filtered value is used as the reference power for the heating load demand.

[0078] By using the baseline power and deviation of heating load demand, the proportion of heating power deficit required to maintain terminal comfort during periods of directional deviation is calculated.

[0079] The ratio of the maximum allowable heat release power of the heat storage device to the baseline power required for heating load is used as the alternative upper limit ratio; the maximum allowable heat release power of the heat storage device is the factory-calibrated parameter of the heat storage device, which is read from the nameplate of the heat storage device or the equipment parameter database.

[0080] When the heating power deficit ratio is less than or equal to the upper limit of the substitutable ratio, the heat release capacity of the individual heat storage device is sufficient to cover the deficit that needs to be compensated. The heating power reduction ratio is set as the heating power deficit ratio, and the heat release power compensation ratio is set to 1. The heat storage device fully replaces the heating power deficit caused by the reduction in heating power.

[0081] When the heating power deficit ratio is greater than the substitutable upper limit ratio, the maximum heat release capacity of the thermal storage device is insufficient to independently cover the entire deficit. The heating power reduction ratio is set to the substitutable upper limit ratio, while the heat release power compensation ratio is still set to 1. The thermal storage device uses its maximum heat release capacity to replace the compensable part of the heating deficit caused by the reduction in heating power, and the auxiliary heating branch is activated to provide supplementary heat according to the difference between the heating power deficit ratio and the heating power reduction ratio. The operating power of the auxiliary heating branch is required to be equal to the heating load demand base power multiplied by the calculated difference.

[0082] It should be noted that the calculation method for the heating power deficit ratio is as follows: the absolute value of the change in heating performance during the deviation period is divided by the average value of the transient heating coefficient within the preset period before the deviation, to obtain the heating performance decline ratio for each historical deviation event.

[0083] The ratio of the decrease in heating performance to the degree of deviation for each deviation event is used as the single-event conversion ratio, and the median value of all single-event conversion ratios is used as the conversion factor.

[0084] Define the allowable relative estimation error and confidence level of the conversion factor, and calculate the minimum sample size based on the coefficient of variation of the historical single conversion ratio series using the following formula: .

[0085] In the formula, The minimum sample size to be solved. The quantiles of the standard normal distribution at the corresponding confidence level are used. In this embodiment, the confidence level is 95%. , This is the coefficient of variation for a single historical conversion series. The coefficient of variation represents the ratio of the standard deviation to the mean. To allow for relative estimation error, a value of 0.1 is set.

[0086] When the number of statistical samples of historical divergent events is lower than the minimum sample size, the conversion factor is set to 1.

[0087] The formula is used to determine the minimum historical deviation event sample size required to reliably estimate the conversion factor under given confidence levels and allowable errors. Its mathematical structure shows that the required sample size is directly proportional to the square of the product of the coefficient of variation and the confidence level quantile, and inversely proportional to the square of the allowable relative estimation error. When the actual historical sample size is lower than the minimum sample size, the reliability and accuracy of the statistical estimation results cannot meet the preset requirements; therefore, the conversion factor is set to 1. The estimation accuracy of the conversion factor determined in this way increases with the increase of the historical sample size; when the historical sample size is insufficient, the default value of 1 is used, and statistical estimation is not relied upon.

[0088] Multiply the degree of deviation by a conversion factor to obtain the initial heating power deficit ratio.

[0089] Obtain the allowable heating reduction margin ratio for end-user comfort. The heating reduction margin ratio is determined as follows: extract the time period in historical deviation events where the heating power decreases but the end-user temperature does not exceed the allowable fluctuation range, calculate the maximum reduction ratio of the heating power relative to the baseline power before the deviation during this period, form a sequence composed of the maximum reduction ratio of each historical event, sort the sequence in ascending order, take the second value, divide it by the preset safety factor, and take it as the heating reduction margin ratio.

[0090] The preset safety factor is determined as follows: the maximum allowable overshoot of the terminal temperature is obtained, and the maximum allowable overshoot is the difference between the upper limit of the terminal user's comfortable temperature range and the actual set temperature; the upper limit of the terminal user's comfortable temperature range is the preset upper limit value of the user's comfortable temperature interval, which is set by the user through the heating system control panel or determined by the factory default configuration of the heating system.

[0091] Extract the end-temperature overshoot sequence from historical divergence events, and take the 95th percentile of the overshoot sequence as the historical maximum overshoot.

[0092] The ratio of the historical maximum overshoot to the maximum allowable overshoot is used as the preset safety factor. If the calculated preset safety factor is less than 1, the preset safety factor is directly set to 1.

[0093] Subtract the heating capacity reduction margin from the initial heating power deficit ratio. If the difference is greater than 0, use the difference as the heating power deficit ratio. If the difference is less than or equal to zero, set the heating power deficit ratio to 0.

[0094] S4. In the preparatory stage before entering the directional divergence period, control the heat pump to operate at compensating power and introduce the surplus heat exceeding the real-time heating load into the heat storage device.

[0095] In this embodiment, controlling the heat pump to operate with compensated power includes: accumulating the baseline power of the heating load demand, the percentage decrease in heating power, and the predicted duration of the directional deviation period, and calculating the total heat output reduced due to the decrease in heating power during the directional deviation period.

[0096] The total heat output is calculated based on the heat loss rate during the heat release process of the heat storage device. Specifically, the difference between 1 and the heat loss rate is used as the denominator, and the total heat output is used as the numerator. The ratio is then used to obtain the effective heat that should be pre-stored. The heat loss rate is the ratio of the remaining heat after the heat storage device has been left to stand for a unit time in a fully stored state to the full stored heat. This is a calibration parameter of the heat storage device at the time of manufacture and is provided by the heat storage device manufacturer.

[0097] The effective heat is allocated to the available heat storage time in the preparatory stage. The available heat storage time is the time span from the current time to the predicted start time of the deviation period. The basic compensation power is obtained by dividing the effective heat to be pre-stored by the available heat storage time.

[0098] Extract the upper limit of the short-term overload capacity of the heat pump, the maximum heat storage power converted from the remaining storage capacity during the heat storage period, and the basic compensation power, and take the minimum value among the three as the actual compensation power.

[0099] Among them, the upper limit of the short-time overload capacity of the heat pump is the maximum allowable input power for short-time as indicated on the heat pump nameplate; the remaining storable capacity refers to the difference between the rated heat storage capacity of the heat storage device in the current state and the heat storage capacity already stored, which represents the upper limit of heat that the heat storage device can continue to store; the maximum heat storage power converted from the remaining storable capacity during the heat storage period is specifically obtained by dividing the remaining storable capacity of the heat storage device by the available heat storage time.

[0100] During the preparation phase, the heat pump is controlled to operate at the total output of the benchmark power for heating load demand plus the actual compensation power. Here, "plus" specifically refers to the addition operation. The heat energy corresponding to the compensation power is introduced into the heat storage device through the heat exchange cycle. The compensation operation is terminated when the energy storage level of the heat storage device reaches the preset high energy storage indicator or reaches the start time of the deviation period.

[0101] It should be noted that the preset high energy storage indicator is determined in the following way: the maximum allowable operating temperature of the heat storage device is obtained, and the highest heat storage temperature recorded by the heat storage device during the heat storage stage is obtained.

[0102] When the maximum heat storage temperature does not exceed the maximum allowable operating temperature, the preset high energy storage indicator position is set to 90% to 95% of the rated heat storage capacity.

[0103] When the maximum heat storage temperature exceeds the maximum allowable operating temperature, the ratio of the heat storage temperature corresponding to the maximum allowable operating temperature to the rated heat storage temperature is used as the preset high energy storage indicator. The heat storage temperature corresponding to the maximum allowable operating temperature is determined by the product of the specific heat capacity of the heat storage medium, the total mass of the heat storage medium, and the maximum allowable operating temperature.

[0104] S5. When entering the period of directional deviation, the heating power of the heat pump is reduced according to the ratio of the decrease in heating power, and the heat storage device is controlled to release heat to compensate for the deficiency according to the heat release power compensation ratio. When the heat storage device has insufficient heat release capacity, it is supplemented by the auxiliary heating branch.

[0105] In this embodiment, the control during the period of directional deviation includes: the control unit obtains the actual heating power after the heat pump heating power decreases in real time through the heat pump operating parameter feedback interface, and determines the instantaneous heating deficit power based on the difference between the heating load demand benchmark power and the actual heating power.

[0106] The product of the instantaneous heating deficit power and the heat release power compensation ratio is taken as the target compensation power of the heat storage device.

[0107] The heat release regulating mechanism of the thermal storage device is adjusted to make the heat release power track the target compensation power. Specifically, for sensible thermal storage devices, the regulating mechanism is a variable frequency drive of the circulating water pump or an electric regulating valve of the heat exchange pipeline, which adjusts the heat release power by changing the circulation flow rate of the thermal storage medium; for phase change thermal storage devices, the regulating mechanism is a flow regulating valve of the heat exchange medium circuit; for electrothermal thermal storage devices, the regulating mechanism is a group contactor of the electrothermal element or a thyristor power regulator, which adjusts the heat release power by changing the number of engaged groups or the conduction angle. The control unit uses the target compensation power as a reference value and drives the regulating mechanism through a proportional-integral control algorithm to make the actual heat release power track the target value. The specific tracking method is a mature existing technology and will not be elaborated here.

[0108] During the heat release compensation process, the remaining energy storage capacity of the heat storage device is continuously monitored. When the remaining energy storage capacity drops to the preset safety reserve limit, the heat release of the heat storage device is stopped, and the auxiliary heating branch is activated to provide the remaining portion. The preset safety reserve limit is determined based on the minimum energy storage density that does not affect the cycle life of the heat storage device under deep heat release conditions. It is obtained from the durability test report of the heat storage device at the factory, specifically the minimum energy storage capacity corresponding to a heat storage capacity decay of no more than 20% in 100 deep charge and deep discharge cycle tests.

[0109] After detecting the end of the deviation period, if the auxiliary heating branch has been put into operation, it will be shut down first; if the remaining energy storage capacity is higher than the safety reserve limit, the heat storage device will be restored to the heat release state; otherwise, the heat release will be stopped until the recharge process is completed.

[0110] It should be added that the detection and judgment conditions for the end of the deviation period are any one of the following: the current time coincides with the predicted end time of the deviation period, or the control unit confirms that the two are no longer in a deviation state by real-time monitoring of the directional relationship between carbon emission factor and heating performance, and the directional markers of carbon emission factor and heating performance return to the same or the directional difference disappears.

[0111] S6. When the period of directional deviation ends, control the heat pump to resume normal heating and perform recharging on the heat storage device.

[0112] In this embodiment, the recharging of the heat storage device includes: obtaining the total heat released by the heat storage device during the period of directional deviation by integrating the heat release power of the heat storage device over time, that is, by multiplying the heat release power in each sampling interval by the sampling interval duration and then summing them up.

[0113] Setting a recharge allowable time window: Obtain time-of-use electricity price data for the current day and the following day from the electricity market information disclosure interface, and locate the nearest low-price period after the end of the divergence period as the initial candidate interval for the recharge window. Simultaneously, continuously run the divergence prediction relationship. If a new directional divergence period is predicted within the candidate interval, move the end time of the recharge window forward to a safety margin time one step before the prediction start time of the new divergence period. This safety margin is half the usable heat storage duration during the preparation phase, and does not exceed 30 minutes.

[0114] Within the recharge allowable time window, the heat pump is subjected to normal load tracking control, and the natural power margin between the actual output of the heat pump and the heating load demand is collected in real time. The method of obtaining the actual output of the heat pump is the same as that of obtaining the actual heating power after the heat pump heating power decreases in S5, and both are read through the heat pump operating parameter feedback interface.

[0115] The net heat to be recharged is obtained by subtracting the natural power margin within the recharge window from the total heat released. The net heat to be recharged is then amortized over the remaining time of the recharge allowable time window to generate real-time additional recharge power.

[0116] The heat pump output target is to add additional real-time recharge power on top of meeting the heating load demand, and the heat energy corresponding to the additional output is introduced into the heat storage device until the energy storage level rises to the preset high energy storage indicator or the recharge allowable time window ends.

[0117] Specifically, the complete implementation of the complementary control of the present invention is as follows: In the time dimension, the pre-heat storage before the deviation period, the replacement of the decrease in heating power during the deviation period, and the recharge recovery after the deviation period are connected in series to form a time-series complementary relationship between heat pump pre-heat storage and heat storage replacing heat pump; In the power dimension, the actual heating power after the decrease in the heating power of the heat pump during the deviation period and the target compensation power of the heat storage device jointly supply the heating load, wherein the output of the heat pump is reduced according to the proportion of the decrease in heating power and bears the reduced basic heating share, and the output of the heat storage device bears the compensation share in the heating deficit according to the heat release power compensation ratio. The two constitute a power complementary relationship between the basic output of the heat pump and the compensation output of the heat storage device.

[0118] Reference Figure 3 As shown, the second embodiment of the present invention provides a heating-complementary control system based on AI technology, including: a prediction training module, a prediction output module, a power matching determination module, a heat storage compensation preparation module, a heat storage compensation execution module, and a recovery and recharge control module.

[0119] The prediction training module is connected to the prediction output module, the prediction output module is connected to the power matching determination module, the power matching determination module is connected to the heat storage compensation preparation module, the heat storage compensation preparation module is connected to the heat storage compensation execution module, and the heat storage compensation execution module is connected to the recovery and recharge control module.

[0120] The prediction training module identifies periods within the historical operating range of the heat pump where the heating performance and carbon emission factor change directions diverge. It extracts the environmental condition sequence and carbon emission factor sequence corresponding to the divergent periods and trains the divergence prediction relationship based on a mapping network constructed by recursive quantile partitioning.

[0121] The prediction output module collects current environmental conditions and current carbon emission factors, and uses the deviation prediction relationship to output the prediction results and the degree of deviation for the directional deviation period within a preset future time period.

[0122] The power matching determination module determines the reduction ratio of the heat pump's heating power and the compensation ratio of the heat storage device's heat release power during the predicted period of directional deviation, based on the degree of deviation and the status of the heat storage device.

[0123] The heat storage compensation preparation module controls the heat pump to operate at compensating power during the preparation stage before entering the directional deviation period, and introduces the surplus heat exceeding the real-time heating load into the heat storage device.

[0124] The heat storage compensation execution module reduces the heat pump's heating power according to the ratio of the decrease in heating power when entering the period of directional deviation. At the same time, it controls the heat storage device to release heat to compensate for the deficiency according to the heat release power compensation ratio. When the heat storage device's heat release capacity is insufficient, it is supplemented by the auxiliary heating branch.

[0125] When the recharge control module is restored and the period of directional deviation ends, the heat pump is controlled to resume normal heating and the heat storage device is recharged.

[0126] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A heating complementary control method based on AI technology, characterized in that, include: Identify the periods in the historical operation range of heat pumps where the heating performance and carbon emission factor change direction deviate from each other, extract the environmental condition sequence and carbon emission factor sequence corresponding to the periods of deviation, and train the deviation prediction relationship based on the mapping network constructed by recursive quantile partitioning. Collect current environmental conditions and current carbon emission factors, and use the deviation prediction relationship to output the prediction results and the degree of deviation for the directional deviation period within the future preset period. When a period of directional deviation is predicted, the heat pump's heating power reduction ratio and the heat storage device's heat release power compensation ratio are determined based on the degree of deviation and the status of the heat storage device during the directional deviation period. In the preparatory stage before entering the directional divergence period, the heat pump is controlled to operate at a compensating power level, and the surplus heat exceeding the real-time heating load is introduced into the heat storage device. When entering the period of directional deviation, the heat pump heating power is reduced according to the ratio of the decrease in heating power, and the heat storage device is controlled to release heat to compensate for the deficiency according to the heat release power compensation ratio. When the heat storage device has insufficient heat release capacity, it is supplemented by the auxiliary heating branch. When the period of directional deviation ends, the heat pump is controlled to resume normal heating and the heat storage device is recharged. The period during which the heating performance and carbon emission factor changes diverge within the historical operating range of the heat pump include: A sliding time window is applied to the transient heating coefficient sequence and carbon emission factor sequence of the historical operating range, and continuous segments are extracted. Within each sub-segment, a pruning least-squares regression is performed using a trend extraction operator to obtain a trend scalar. The trend scalar is then compared with an upper and lower bound determined based on the local fluctuation amplitude of the sub-segment: when the trend scalar is greater than the upper bound, an upward flag is assigned; when it is less than the lower bound, a downward flag is assigned; and when it is between the upper and lower bounds, a direction flag of the immediately preceding sub-segment is assigned. After traversing all sub-segments, the heating performance direction marker sequence and the carbon emission factor direction marker sequence are compared point by point. Continuous time intervals with directional differences are merged into directional deviation events. The start time, end time, degree of deviation, and environmental condition sequence and carbon emission factor sequence during the deviation period are recorded. The degree of deviation is the relative proportion of the change in heating performance to the change in carbon emission factor during the deviation period.

2. The heating complementary control method based on AI technology according to claim 1, characterized in that, The process of using trend extraction operators to perform trimmed least-squares regression to obtain the propensity scalar includes: Perform an initial linear fit on the sequence points within the current sub-segment, and take the slope of the fitted line as the tendency scalar; Calculate the residuals of each sequence point relative to the fitted line, and calculate the standard deviation of the residuals; Outliers exceeding a preset multiple of the residual standard deviation are removed, and the retained sequence points are refitted. Iteratively perform residual calculation, outlier removal and refitting until no new outliers are added or the preset maximum number of iterations is reached; The slope of the straight line obtained from the last fitting is used as the final tendency scalar output of the current sub-segment.

3. The heating complementary control method based on AI technology according to claim 1, characterized in that, The training deviation from the predicted relationship includes: For each historical deviation event, extract the environmental condition sequence and carbon emission factor sequence within a preset backtracking width before the event starting point, calculate the multi-order statistical characteristics, trend slope, extreme value difference of the environmental conditions within the preset backtracking width, as well as the mean, cumulative rate of change and fluctuation of the carbon emission factors, and form a condition attribute vector; The result attributes are composed of the lag time and the degree of deviation of the event relative to the current moment; A mapping network consisting of hierarchical condition decision nodes is established. Each decision node corresponds to a condition attribute and a comparison branch based on the condition attribute quantile interval is set. The input data is guided to the leaf node after being judged at each level. The leaf node stores the corresponding values ​​of the predicted deviation time period offset and the degree of deviation. The mapping network is obtained by performing recursive quantile partitioning on the correspondence between all condition attribute vectors and result attributes.

4. The AI-based heating complementary control method according to claim 3, characterized in that, The method of using the divergence prediction relationship to output the prediction results and degree of deviation for the directional divergence period within a preset future time period includes: Collect current environmental conditions and current carbon emission factors, and obtain the current condition attribute vector using the same window width and feature extraction method as the constructed condition attribute vector; The current conditional attribute vector is input into the mapping network, and after step-by-step branching and judgment, it reaches the corresponding leaf node. The output is the predicted start offset, predicted duration, and prediction deviation of the predicted deviation period relative to the current time. Obtain the current environmental conditions' near-term trend. When the near-term trend deviates from the typical historical trend and the rate of change exceeds the trend reinforcement reference value, adjust the expected initial offset towards the direction closer to the current moment, and amplify the prediction deviation proportionally to the proportion of the rate of change exceeding the trend reinforcement reference value.

5. The heating complementary control method based on AI technology according to claim 1, characterized in that, The determination process for the heat pump's heating power reduction ratio and the heat storage device's heat release power compensation ratio during the period of directional deviation includes the following steps: Using the baseline power and deviation of heating load demand, calculate the proportion of heating power deficit required to maintain terminal comfort during periods of directional deviation. The ratio of the maximum allowable heat release power of the thermal storage device to the baseline power required for the heating load is used as the alternative upper limit ratio. When the heating power deficit ratio is less than or equal to the upper limit of the substitutable ratio, the heating power reduction ratio is set to the heating power deficit ratio, and the heat release power compensation ratio is set to 1. When the heating power deficit ratio is greater than the substitutable upper limit ratio, the heating power reduction ratio is set to the substitutable upper limit ratio, the heat release power compensation ratio is set to 1, and the auxiliary heating branch is activated to provide supplementary heat according to the difference between the heating power deficit ratio and the heating power reduction ratio.

6. The heating complementary control method based on AI technology according to claim 1, characterized in that, The control of the heat pump to operate with compensated power includes: Based on the percentage decrease in heating power and the predicted duration of the directional deviation period, calculate the total cumulative reduction in heating capacity due to the decrease in heating power during the directional deviation period. The total heat output is calculated by converting the heat loss rate during the heat release process of the heat storage device to obtain the effective heat that should be pre-stored. The effective heat is allocated to the available heat storage time in the preparation stage to obtain the basic compensation power; Extract the upper limit of the short-term overload capacity of the heat pump, the maximum heat storage power converted from the remaining storage capacity during the heat storage period, and the basic compensation power, and take the minimum value among the three as the actual compensation power; During the preparation phase, the heat pump is controlled to operate at the total output of the benchmark power for heating load demand plus the actual compensation power. The heat energy corresponding to the compensation power is introduced into the heat storage device through heat exchange circulation. When the energy storage level of the heat storage device reaches the preset high energy storage indicator or reaches the start time of the deviation period, the compensation operation is terminated.

7. The heating complementary control method based on AI technology according to claim 1, characterized in that, Control measures during periods of directional divergence include: The actual heating power after the heat pump heating power decreases is obtained in real time, and the instantaneous heating deficit power is determined based on the difference between the heating load demand benchmark power and the actual heating power. The product of the instantaneous heating deficit power and the heat release power compensation ratio is taken as the target compensation power of the heat storage device. Adjust the heat release regulating mechanism of the heat storage device to make the heat release power track the target compensation power; During the heat release process, the remaining energy storage capacity of the heat storage device is continuously monitored. When the remaining energy storage capacity drops to the preset safety reserve limit, the heat release of the heat storage device is stopped, and the auxiliary heating branch is activated to provide the remaining portion. After detecting the end of the deviation period, if the auxiliary heating branch has been put into operation, it will be shut down first; if the remaining energy storage capacity is higher than the safety reserve limit, the heat storage device will be restored to the heat release state; otherwise, the heat release will be stopped until the recharge process is completed.

8. The heating complementary control method based on AI technology according to claim 1, characterized in that, The recharging of the thermal storage device includes: The total heat released by the thermal storage device during the period of directional deviation is statistically analyzed, and a recharge allowable time window is set. Within the allowed recharge time window, the heat pump is subjected to normal load tracking control, and the natural power margin between the actual output of the heat pump and the heating load demand is collected in real time. The net heat to be recharged is obtained by subtracting the natural power margin within the recharge window from the total heat released. The net heat to be recharged is then amortized over the remaining time of the recharge allowable time window to generate real-time additional recharge power. The heat pump output target is to add additional real-time recharge power on top of meeting the heating load demand, and the heat energy corresponding to the additional output is introduced into the heat storage device until the energy storage level rises to the preset high energy storage indicator or the recharge allowable time window ends.

9. A heating-complementary control system based on AI technology, characterized in that, include: The prediction training module identifies periods in the historical operation range of the heat pump where the heating performance and the direction of change in carbon emission factors deviate. It extracts the environmental condition sequence and carbon emission factor sequence corresponding to the deviation period, and trains the deviation prediction relationship based on the mapping network constructed by recursive quantile partitioning. The period during which the heating performance and carbon emission factor changes diverge within the historical operating range of the heat pump include: A sliding time window is applied to the transient heating coefficient sequence and carbon emission factor sequence of the historical operating range, and continuous segments are extracted. Within each sub-segment, a pruning least-squares regression is performed using a trend extraction operator to obtain a trend scalar. The trend scalar is then compared with an upper and lower bound determined based on the local fluctuation amplitude of the sub-segment: when the trend scalar is greater than the upper bound, an upward flag is assigned; when it is less than the lower bound, a downward flag is assigned; and when it is between the upper and lower bounds, a direction flag of the immediately preceding sub-segment is assigned. After traversing all sub-segments, the heating performance direction marker sequence and the carbon emission factor direction marker sequence are compared point by point. Continuous time intervals with directional differences are merged into directional deviation events. The start time, end time, degree of deviation, and environmental condition sequence and carbon emission factor sequence during the deviation period are recorded. The degree of deviation is the relative proportion of the change in heating performance to the change in carbon emission factor during the deviation period; The prediction output module collects current environmental conditions and current carbon emission factors, and uses the deviation prediction relationship to output the prediction results and the degree of deviation for the future preset time period. The power matching determination module determines the heating power reduction ratio of the heat pump and the heat release power compensation ratio of the heat storage device during the predicted period of directional deviation, based on the degree of deviation and the status of the heat storage device. The heat storage compensation preparation module controls the heat pump to operate at compensation power during the preparation stage before entering the directional deviation period, and introduces the surplus heat exceeding the real-time heating load into the heat storage device. The heat storage compensation execution module reduces the heat pump heating power according to the heating power reduction ratio when entering the directional deviation period, and simultaneously controls the heat storage device to release heat to compensate for the deficiency according to the heat release power compensation ratio. When the heat storage device's heat release capacity is insufficient, it is supplemented by the auxiliary heating branch. When the recharge control module is restored and the period of directional deviation ends, the heat pump is controlled to resume normal heating and the heat storage device is recharged.

Images