AI-based dynamic threshold energy-saving regulation algorithm execution system

Abstract

The application relates to the technical field of industrial and building energy-saving control, and particularly discloses an AI-based dynamic threshold energy-saving regulation and control algorithm execution system, which collects working condition characteristic data, dynamic threshold sequences and energy efficiency feedback values of multiple source regulation and control systems; calculates local heterogeneity indexes, working condition boundary envelopes and risk overflow coefficients of the source regulation and control systems; obtains initial working condition characteristic data of a target regulation and control system, calculates the heterogeneity weighted distance between the initial working condition characteristic data and the boundary envelopes of the source regulation and control systems, and accumulates the total heterogeneity distance value; generates a nonlinear attenuation base number in sequence according to the value, obtains an inertia damping coefficient through weighting, and then obtains a safety barrier compression factor through inverse tangent stretching; multiplies the factor with original dynamic thresholds and regulation and control change amplitudes to obtain conservative correction thresholds and gradual regulation and control limiting amplitudes and outputs; the application can automatically correct and migrate thresholds when the working conditions are greatly different, thereby guaranteeing safety and improving regulation and control stability.

Description

Technical Field

[0001] This invention relates to the field of industrial and building energy-saving control technology, specifically to an AI-based dynamic threshold energy-saving control algorithm execution system. Background Technology

[0002] Currently, AI-based dynamic threshold energy-saving control methods have been applied in scenarios such as central air conditioning, data center cooling, and centralized heating. These methods collect real-time equipment operating parameters and environmental parameters, and use machine learning models to dynamically adjust the upper or lower threshold values ​​of the controlled object, enabling the equipment to operate in a low-energy consumption range while meeting demand. Compared to traditional fixed threshold control methods, this approach offers better adaptability to changing operating conditions.

[0003] When federated transfer learning is applied to dynamic threshold energy-saving regulation, the existing technology lacks a method that can quantify the degree of heterogeneity between the target regulation system and multiple source regulation systems, and perform safety-oriented nonlinear compression of the transfer threshold based on the degree of heterogeneity while maintaining the smoothness of the time-series regulation. This results in new sites being forced to abandon the use of the transfer strategy because they cannot assess the risks, and the engineering value of transfer learning cannot be realized. Summary of the Invention

[0004] The purpose of this invention is to provide an AI-based dynamic threshold energy-saving control algorithm execution system to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions: An AI-based dynamic threshold energy-saving control algorithm execution system includes: The source domain feature acquisition module is used to collect the real-time operating parameters and environmental parameters of multiple source control systems as source operating condition feature data, and to collect the dynamic threshold sequence and corresponding energy efficiency feedback value output by each source control system in the historical control cycle. The heterogeneity and boundary extraction module calculates the local heterogeneity index of each source control system based on the degree of deviation between the source operating condition feature data and the global benchmark gradient direction. At the same time, it extracts the operating condition boundary envelope of each source control system in the feature space and calibrates its risk spillover coefficient based on the historical energy efficiency feedback value of each source control system. The target domain distance accumulation module acquires the initial operating condition feature data of the target control system, calculates the heterogeneous weighted distance of the initial operating condition feature data falling within the corresponding operating condition boundary envelope of each source control system, and accumulates the heterogeneous weighted distances of all source control systems to obtain the total heterogeneous distance value. The safety compression factor generation module generates a safety barrier compression factor based on the total heterogeneous distance value. The safety barrier compression factor decreases monotonically as the total heterogeneous distance value increases. The conservative correction output module multiplies the compression factor with the original dynamic threshold and the original single-step control change amplitude to obtain the conservative correction threshold and the gradual control limit, and uses the conservative correction threshold and the gradual control limit as the initial execution parameters output of the target control system.

[0006] As a further aspect of the present invention: the calculation of the local heterogeneity index of each source control system specifically includes: Calculate the cosine of the deviation angle between the gradient update direction of each source control system and the global baseline gradient direction, and then calculate the coefficient of variation of the threshold adjustment step size of each source control system within a continuous control period. The product of the cosine of the deviation angle and the coefficient of variation is used as the local heterogeneity index of the source control system.

[0007] As a further aspect of the present invention: the step of calibrating the risk spillover coefficient based on the historical energy efficiency feedback values ​​of each source control system specifically includes: For each source control system, its historical operating condition characteristic data is reduced in dimensionality through principal component analysis, and the convex hull vertices are extracted as the set of operating condition boundary envelope points. Then, the proportion of the duration during which the historical energy efficiency feedback value of each source control system is lower than the preset energy efficiency benchmark is counted as the basic risk coefficient. The basic risk coefficient is multiplied by the corresponding local heterogeneity index and normalized to obtain the risk spillover coefficient.

[0008] As a further aspect of the present invention: the process of obtaining the set of boundary points of the working condition is as follows: Calculate the local reachability density of each historical working condition feature data point after dimensionality reduction, retain points with local reachability density greater than the global density mean as candidate boundary points, and then use the convolution wrapping method to connect the candidate boundary points in counterclockwise order to form an initial convex hull vertex sequence. Calculate the Euclidean distance between every two adjacent vertices in the initial convex hull sequence, and perform linear interpolation to fill in the points between adjacent vertex pairs whose distance is greater than three times the overall average spacing, to obtain the encrypted working condition boundary envelope point set.

[0009] As a further aspect of the present invention: obtaining the total heterogeneous distance value specifically includes: For each source control system, calculate the shortest Euclidean distance from the initial operating condition characteristic data points to the operating condition boundary envelope of the source control system; Multiplying the shortest Euclidean distance by the product of the local heterogeneity index and the risk spillover coefficient of the corresponding source control system yields the heterogeneous weighted distance of the source control system. After sorting the heterogeneous weighted distances of all source control systems in ascending order of their values, the maximum and minimum values ​​in the sequence are removed, and the remaining heterogeneous weighted distances are summed. The sum is taken as the total heterogeneous distance value.

[0010] As a further aspect of the present invention: the generation of the security barrier compression factor based on the total heterogeneous distance value specifically includes: The nonlinear decay base is obtained by convolving the reciprocal of the total heterogeneous distance value with the negative exponent of the base of the natural constant. The inertial damping coefficient is obtained by weighting the nonlinear decay base with the historical compression factor output from the first three control cycles and correcting it based on the total heterogeneous distance. An arctangent stretching transformation is performed on the inertial damping coefficient so that the output value decreases monotonically as the total heterogeneous distance increases, and the result is used as the safety barrier compression factor.

[0011] As a further aspect of the present invention: the formation of the inertial damping coefficient specifically includes: Obtain the historical compression factor of the output of the previous three consecutive control cycles, assign a decay weighting coefficient that is inversely proportional to the duration of the current cycle to the historical compression factor and normalize it. The inertia estimate is obtained by multiplying the nonlinear attenuation base with the three normalized attenuation weighting coefficients and summing the results. Calculate the absolute difference between the preliminary inertia estimate and the inertia damping coefficient output in the previous control cycle. Multiply the absolute difference by the hyperbolic tangent of the total heterogeneous distance value as a correction, and add it to the preliminary inertia estimate. The resulting value is the inertia damping coefficient for this control cycle.

[0012] As a further aspect of the present invention: the use of conservative correction threshold and gradual control limit as the initial execution parameter output of the target control system specifically includes: The compression factor is subjected to square root and cube root operations respectively. The original dynamic threshold is multiplied by the square root operation result to obtain the first correction value. The original dynamic threshold is multiplied by the cube root operation result to obtain the second correction value. The arithmetic mean of the first correction value and the second correction value is calculated as the conservative correction threshold. The temporary amplitude limit is obtained by multiplying the original single-step control change amplitude by the product of the safety barrier compression factor and the arctangent function value of the total heterogeneous distance. The temporary control limit is weighted and averaged with the gradual control limit output from the previous control cycle, where the weight of the current temporary control limit is greater than the weight of the output from the previous cycle. The resulting weighted average is used as the gradual control limit output for the current control cycle.

[0013] The beneficial effects of this invention are: (1) By calculating the heterogeneous weighted distance between the target control system and each source control system and generating a safety barrier compression factor that monotonically decreases as the distance increases, the dynamic threshold of the migration to the new site can be automatically conservatively corrected towards the lower safety limit when the operating conditions are significantly different. This avoids the risk of equipment operating parameters exceeding the limit that may be caused by directly adopting the aggressive strategy of the source site, provides the operation and maintenance team with a quantifiable safety margin basis, and improves the acceptability of the federated transfer learning strategy in actual engineering.

[0014] (2) By weighting the nonlinear attenuation base with the historical compression factor to form the inertial damping coefficient, and performing an arctangent stretching transformation on the inertial damping coefficient, the safety barrier compression factor has an inertial lag characteristic between multiple control cycles, avoiding threshold mutation caused by the fluctuation of the total heterogeneous distance value in a single operation, ensuring the smoothness and stability of the initial execution parameter output of the target control system, and reducing the frequent action loss of the actuator caused by the drastic threshold jump in the control system. Attached Figure Description

[0015] The invention will now be further described with reference to the accompanying drawings.

[0016] Figure 1 This is a system block diagram of the present invention; Figure 2 This is a flowchart of the process for generating the safety barrier compression factor in this invention. Detailed Implementation

[0017] 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.

[0018] Please see Figure 1 As shown, the present invention is an AI-based dynamic threshold energy-saving control algorithm execution system, comprising: The source domain feature acquisition module is used to collect the real-time operating parameters and environmental parameters of multiple source control systems as source operating condition feature data, and to collect the dynamic threshold sequence and corresponding energy efficiency feedback value output by each source control system in the historical control cycle. The heterogeneity and boundary extraction module calculates the local heterogeneity index of each source control system based on the degree of deviation between the source operating condition feature data and the global benchmark gradient direction. At the same time, it extracts the operating condition boundary envelope of each source control system in the feature space and calibrates its risk spillover coefficient based on the historical energy efficiency feedback value of each source control system. The target domain distance accumulation module acquires the initial operating condition feature data of the target control system, calculates the heterogeneous weighted distance of the initial operating condition feature data falling within the corresponding operating condition boundary envelope of each source control system, and accumulates the heterogeneous weighted distances of all source control systems to obtain the total heterogeneous distance value. The safety compression factor generation module generates a safety barrier compression factor based on the total heterogeneous distance value. The safety barrier compression factor decreases monotonically as the total heterogeneous distance value increases. The conservative correction output module multiplies the compression factor with the original dynamic threshold and the original single-step control change amplitude to obtain the conservative correction threshold and the gradual control limit, and uses the conservative correction threshold and the gradual control limit as the initial execution parameters output of the target control system.

[0019] In the source domain feature acquisition module, real-time operating parameters and environmental parameters of multiple source control systems are collected as source operating condition feature data. The module also collects the dynamic threshold sequences and corresponding energy efficiency feedback values ​​output by each source control system within historical control cycles, specifically including: For each source control system participating in data acquisition, the following sensors are connected to its local controller via a data bus interface: temperature sensors, pressure transmitters, flow meters, and power meters for detecting equipment operating parameters, and outdoor temperature and humidity sensors and barometers for detecting environmental parameters. These sensors sample real-time values ​​at a frequency of once per second, converting them from analog to digital and storing them in a local historical database. Operating parameters include at least the inlet and outlet water temperatures, pipeline pressure, medium flow rate, and equipment input power of the controlled object; environmental parameters include at least outdoor dry-bulb temperature, relative humidity, and atmospheric pressure. The data acquisition period is no less than 30 consecutive calendar days and covers at least three typical load conditions. These real-time operating parameters and environmental parameters together constitute the source operating condition characteristic data.

[0020] Dynamic threshold sequences corresponding to the aforementioned data collection periods are extracted from the historical database of each source control system. These sequences consist of the upper and lower threshold values ​​output by the source control system cycle-by-cycle during its historical control periods, with each control period lasting from 5 to 15 minutes. The sequences are arranged chronologically. Simultaneously, energy efficiency feedback values ​​are collected for each control period. These feedback values ​​represent the ratio of the actual output energy to the input electrical energy of the controlled object within that period, obtained by synchronously reading and calculating data from calorimeters installed on the energy output side and smart meters on the input side. The dynamic threshold sequences and energy efficiency feedback values ​​are correlated one-to-one according to the control period, forming time-aligned data pairs.

[0021] All source control system data, including source operating condition characteristics, dynamic threshold sequences, and energy efficiency feedback values, are aggregated to a central data storage node via a wired local area network (LAN) or encrypted wireless network. Each source control system's data is accompanied by a unique identifier and timestamp and stored in a structured table format for subsequent calculations of local heterogeneity indices, extraction of operating condition boundary envelopes, and calibration of risk spillover coefficients. This data acquisition process is performed offline, without interfering with the normal control operation of each source control system.

[0022] In the heterogeneity and boundary extraction module, the local heterogeneity index of each source control system is calculated based on the degree of deviation between the source operating condition feature data and the global benchmark gradient direction. Simultaneously, the operating condition boundary envelope of each source control system in the feature space is extracted, and its risk spillover coefficient is calibrated based on the historical energy efficiency feedback values ​​of each source control system. Specifically, this includes: Calculate the local heterogeneity index for each source control system. This index is obtained by multiplying the cosine of the deviation angle by the coefficient of variation. The cosine of deviation is calculated as follows: for each source control system, perform a dot product between its local gradient update direction vector and the global reference gradient direction vector, then divide the result by the product of the magnitudes of these two vectors; the quotient is the cosine of deviation. The global reference gradient direction is the arithmetic mean of the local gradient vectors uploaded by all source control systems. The coefficient of variation is calculated as follows: continuously extract the dynamic threshold adjustment step size of a source control system for each of the last 50 control cycles. First, calculate the arithmetic mean of these 50 step sizes. Then, calculate the square root of the sum of the squares of the differences between each step size and the mean, divide by the number of steps, and finally divide the standard deviation by the arithmetic mean to obtain the coefficient of variation. The result of multiplying the cosine of deviation by the coefficient of variation is used as the local heterogeneity index for that source control system. If the local gradient direction of a certain source control system is completely consistent with the global reference direction and the step size is stable, the index will approach 0; if the direction deviates greatly and the step size fluctuates drastically, the index value will be large.

[0023] It should be noted that each source control system incorporates a Deep Q-Network (DQN) as its threshold decision model. This model takes current operating condition feature data (an 8-dimensional vector) as input and outputs a dynamic threshold adjustment. During model training, a mean squared error loss function is used. The loss is calculated each cycle based on the actual energy efficiency feedback value, and backpropagation is performed to obtain the gradient of the loss function with respect to the model parameters. The gradient vector of the loss function with respect to the weights of the last fully connected layer is used as the local gradient update direction for this system. The arithmetic mean of the local gradients uploaded from all source control systems is then used to obtain the global baseline gradient direction.

[0024] Extract the boundary envelope point set of each source control system in the feature space. The specific process is as follows: First, for all historical source control system feature data collected, each data point contains 8 dimensions including operating parameters and environmental parameters. Principal component analysis is used to reduce the 8-dimensional data to 2-dimensionality, retaining the first two principal component components with a cumulative variance contribution rate greater than 85%, resulting in a 2D coordinate point set after dimensionality reduction. Second, calculate the local reachability density of each data point after dimensionality reduction: For each point, find its 15 nearest neighboring points, calculate the Euclidean distance from the point to each neighboring point, and take the arithmetic mean of the reciprocals of these 15 distances as the local reachability density of the point. Third, calculate the global density mean, which is the arithmetic mean of the local reachability densities of all data points. Points with local reachability densities greater than the global density mean are retained as candidate boundary points. Fourth, use the convolute method to process the candidate boundary points: Starting from the leftmost and bottommost point, each time the point with the smallest counterclockwise polar angle is selected as the next vertex, and so on, forming an initial convex hull vertex sequence. The fifth step involves calculating the Euclidean distance between every two adjacent vertices in the initial convex hull sequence, and then calculating the arithmetic mean of all adjacent vertex distances to obtain the overall average spacing. For adjacent vertex pairs whose distance is greater than three times the overall average spacing, linear interpolation is performed to fill in the gaps: several points are inserted at equal intervals between the lines connecting the two vertices, ensuring that the distance between any two adjacent points after interpolation does not exceed 1.5 times the overall average spacing. After interpolation, the final set of boundary envelope points for the working condition is obtained.

[0025] The risk spillover coefficient of each source control system is calibrated. The calculation of the risk spillover coefficient involves multiplying the basic risk coefficient by the local heterogeneity index and then normalizing. The basic risk coefficient is obtained by statistically analyzing the energy efficiency feedback values ​​of each source control system for all control cycles within the historical data collection period. A preset energy efficiency benchmark value is established, which is the rated energy efficiency ratio of the source control system under design conditions. The proportion of the duration during which the energy efficiency feedback value is lower than the benchmark value for each control cycle is calculated, and the arithmetic mean of this proportion is taken for all cycles to obtain the basic risk coefficient. The basic risk coefficient is then multiplied by the previously calculated local heterogeneity index of the source control system to obtain a product value. This product value is then normalized: the product values ​​of all source control systems are summarized, and the maximum and minimum values ​​are identified. For each source control system, the minimum value is subtracted from its product value, and the result is divided by the difference between the maximum and minimum values; the quotient is the risk spillover coefficient of that source control system. The coefficient ranges from 0 to 1. The larger the value, the more frequently the regulation strategy of the source regulation system deviates from the energy efficiency benchmark in history and the stronger its heterogeneity.

[0026] In the target domain distance accumulation module, the initial operating condition characteristic data of the target control system is acquired, the heterogeneous weighted distance of the initial operating condition characteristic data falling within the corresponding operating condition boundary envelope of each source control system is calculated, and the heterogeneous weighted distances of all source control systems are accumulated to obtain the total heterogeneous distance value, specifically including: Acquire initial operating condition characteristic data of the target control system. Before connecting the target control system, allow it to operate safely in a fixed threshold mode for 72 hours. During this period, collect operating parameters and environmental parameters of the same type as those of the source control system through sensors connected to its local controller, including inlet water temperature, outlet water temperature, pipeline pressure, medium flow rate, equipment input power, outdoor dry-bulb temperature, relative humidity, and atmospheric pressure. The sampling frequency is once per second, and all collected data points constitute the target initial operating condition characteristic dataset.

[0027] For each source control system, the shortest Euclidean distance from each data point in the target initial operating condition feature dataset to the operating condition boundary envelope of that source control system is calculated. Specifically, for a data point in the target initial operating condition feature dataset, it is first reduced to two-dimensional coordinates using principal component analysis (PCA), with the same principal component projection matrix used when extracting the operating condition boundary envelope of the source control system. Then, the Euclidean distance between this reduced-dimensional point and each vertex in the point set of the operating condition boundary envelope of the source control system is calculated, and the minimum value is taken as the shortest distance from this data point to the operating condition boundary envelope of the source control system. This process is repeated for all data points in the target initial operating condition feature dataset. After obtaining the shortest distance for each point, the arithmetic mean of these shortest distances is calculated and taken as the average shortest Euclidean distance from the target control system to the operating condition boundary envelope of the source control system.

[0028] Multiplying the aforementioned average shortest Euclidean distance by the product of the local heterogeneity index and the risk spillover coefficient of the corresponding source control system yields the heterogeneous weighted distance of that source control system. The local heterogeneity index and the risk spillover coefficient are calculated using the steps described above. For each source control system, the local heterogeneity index and the risk spillover coefficient are first multiplied to obtain a combination coefficient, and then the average shortest Euclidean distance is multiplied by this combination coefficient; the resulting product is the heterogeneous weighted distance of that source control system. If the target data point falls within the boundary envelope of the operating condition, the average shortest Euclidean distance is zero, and the heterogeneous weighted distance of the source control system is also zero.

[0029] The heterogeneous weighted distances of all source control systems are collected and sorted in ascending order of value to obtain an ordered sequence. The first minimum value and the last maximum value are removed from this sequence, retaining the remaining intermediate values. The sum of these intermediate values ​​is the total heterogeneous distance. This total heterogeneous distance is used to generate the safety barrier compression factor, whose magnitude reflects the degree of heterogeneity between the target control system and the overall operating condition distribution of all source control systems, as well as the weighted accumulation of potential risks.

[0030] Please see Figure 2 As shown, in the safety compressibility factor generation module, the safety barrier compressibility factor is generated based on the total heterogeneous distance value. The safety barrier compressibility factor monotonically decreases as the total heterogeneous distance value increases, specifically including: Obtain the total heterogeneous distance value, denoted as numerical value. The first sub-step in generating the security barrier compression factor is to calculate the nonlinear attenuation base. Specifically, this involves first calculating the total heterogeneous distance. The reciprocal of 1, that is, 1 divided by 1 Then take the common logarithm (base 10) of the reciprocal to obtain the common logarithmic value; then, using the natural constant e as the base, calculate the total heterogeneous distance value. The opposite number (i.e., negative) Let e ​​be the exponent, and calculate the value of the exponential function, i.e., the negative of e. The product of the commonly used logarithmic value and the exponential function value is defined as the nonlinear decaying base, expressed by the following mathematical formula: ;in, Indicates the nonlinear decay base. This represents the total heterogeneous distance. Represents the common logarithm to base 10. This represents the natural constant (approximately 2.71828). When When it approaches 0, It approaches positive infinity, but in practical applications... The minimum value is limited to no less than 0.001 through data preprocessing to ensure computational stability. This nonlinear decay base varies with... The rapid increase followed by a rapid decrease reflects the physical meaning that the greater the difference between the target operating condition and the source operating condition, the stronger the degree of base attenuation.

[0031] The inertial damping coefficient is formed. This step is specifically divided into three sub-steps: The first sub-step involves obtaining the historical compression factors output from the previous three consecutive control cycles, denoted as the compression factor of the previous cycle, the compression factor of the previous two cycles, and the compression factor of the previous three cycles, respectively. These three historical compression factors are assigned decay weighting coefficients inversely proportional to the duration from the current cycle; that is, the coefficient for the previous cycle is 1, the coefficient for the previous two cycles is half, and the coefficient for the previous three cycles is one-third. The sum of these three coefficients is calculated, and then each coefficient is divided by the sum to obtain three normalized decay weighting coefficients, denoted as the first weighting coefficient, the second weighting coefficient, and the third weighting coefficient, respectively. The second sub-step involves multiplying the nonlinear decay base by the first weighting coefficient, multiplying the compression factor of the previous cycle by the second weighting coefficient, and multiplying the compression factor of the previous two cycles by the third weighting coefficient. These three products are then added together to obtain the preliminary inertial estimate. The third sub-step involves calculating the absolute difference between the preliminary inertial estimate and the inertial damping coefficient output from the previous control cycle. Simultaneously, the total heterogeneous distance is calculated. hyperbolic tangent (i.e. Multiply the absolute difference by the hyperbolic tangent, and use the product as the correction. Finally, add the correction to the preliminary inertia estimate to obtain the inertial damping coefficient for this control cycle. The calculation process in the third sub-step above is expressed by the following mathematical formula: ;in, This represents the inertial damping coefficient for this control cycle. This indicates a preliminary estimate of inertia. This represents the inertial damping coefficient of the output in the previous control cycle. Represents the hyperbolic tangent function. This represents the total heterogeneous distance. The purpose of this formula is to add a correction factor when the initial estimate deviates significantly from the historical value and the total heterogeneous distance is large. This causes the damping coefficient to lag behind sudden changes, preventing drastic fluctuations in the compressibility factor.

[0032] Performing an arctangent stretching transformation on the above inertial damping coefficient yields the safety barrier compressibility factor. Specifically, this involves calculating the arctangent function value of the inertial damping coefficient (i.e.,...). Then, the arctangent value is divided by half of pi (π / 2), stretching the output value to between 0 and 1. Since the arctangent function is monotonically increasing in its domain, and when... As the value increases from 0 to positive infinity, the arctangent value approaches π / 2, therefore the stretched output value follows... It increases monotonically. And... The value of the total heterogeneous distance itself The value increases and then decreases monotonically (because the nonlinear decay base R increases with the value of R). Increases and decreases, and in the correction term Although the overall effect is dominated by R, the final safety barrier compression factor increases accordingly. It increases and then monotonically decreases. The result after the above stretching transformation is output as the safety barrier compression factor for this control cycle.

[0033] It should be noted that for the first control cycle, if there is no historical compression factor, the default value of the historical compression factor will be set to 0.5; the initial value of the inertial damping coefficient will be set to 0.5.

[0034] In the conservative correction output module, the compression factor is multiplied by the original dynamic threshold and the original single-step control change amplitude to obtain the conservative correction threshold and the asymptotic control limit. These conservative correction threshold and asymptotic control limit are then used as the initial execution parameters output for the target control system, specifically including: Obtain the safety barrier compression factor and the original dynamic threshold. Perform square root and cube root operations on the safety barrier compression factor to obtain the square root result and cube root result, respectively. Multiply the original dynamic threshold by the square root result to obtain the first correction value; then multiply the original dynamic threshold by the cube root result to obtain the second correction value. Calculate the arithmetic mean of the first and second correction values, i.e., add them together and divide by two. The result is used as the conservative correction threshold. This conservative correction threshold is used to replace the original dynamic threshold as the upper or lower limit of the target control system in the initial stage. Its value lies between the original dynamic threshold and the safety lower limit, and it approaches the safety lower limit as the safety barrier compression factor decreases.

[0035] Obtain the original single-step control change amplitude. Multiply the safety barrier compression factor by the arctangent function value of the total heterogeneous distance to obtain an intermediate product. The arctangent function value of the total heterogeneous distance is obtained by calculating the arctangent of the total heterogeneous distance and then dividing it by half of pi to normalize the result to between 0 and 1. Multiply the original single-step control change amplitude by the above intermediate product to obtain the temporary limit. This temporary limit is used to restrict the maximum permissible change of the controlled parameter within each control cycle.

[0036] Obtain the asymptotic control limit output from the previous control cycle. Calculate a weighted average of the temporary limit and the asymptotic control limit output from the previous control cycle, where the weight of the current temporary limit is set to 0.7, and the weight of the asymptotic control limit output from the previous cycle is set to 0.3. Calculate the weighted average: first multiply the temporary limit by 0.7, then multiply the asymptotic control limit output from the previous cycle by 0.3, and finally add the two products. The result is used as the asymptotic control limit output for the current control cycle. This asymptotic control limit, together with the conservative correction threshold, constitutes the initial execution parameters of the target control system and is written into the register of the local controller for subsequent cycle-by-cycle energy-saving control execution.

[0037] The working principle of this invention is as follows: Real-time operating parameters and environmental parameters of multiple source control systems are collected as source operating condition characteristic data. The dynamic threshold sequences and corresponding energy efficiency feedback values ​​output by each source control system within historical control cycles are also collected. Based on the deviation between the source operating condition characteristic data and the global benchmark gradient direction, the local heterogeneity index of each source control system is calculated. Simultaneously, the operating condition boundary envelope of each source control system in the feature space is extracted, and the risk spillover coefficient of each source control system is calibrated based on historical energy efficiency feedback values. The initial operating condition characteristic data of the target control system is obtained, and the initial operating condition characteristics are calculated. The heterogeneous weighted distances falling within the boundary envelope of each source control system's operating conditions are calculated, and all heterogeneous weighted distances are summed to obtain the total heterogeneous distance value. Then, based on this total heterogeneous distance value, the nonlinear attenuation base is calculated sequentially, and the inertial damping coefficient is formed through weighting. An arctangent stretching transformation is performed on the inertial damping coefficient to generate a safety barrier compression factor that monotonically decreases as the total heterogeneous distance value increases. This safety barrier compression factor is multiplied by the original dynamic threshold and the original single-step control change amplitude to obtain the conservative correction threshold and the asymptotic control limit, which are used as the initial execution parameters output of the target control system.

[0038] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An AI-based dynamic threshold energy-saving control algorithm execution system, characterized in that, include: The source domain feature acquisition module is used to collect the real-time operating parameters and environmental parameters of multiple source control systems as source operating condition feature data, and to collect the dynamic threshold sequence and corresponding energy efficiency feedback value output by each source control system in the historical control cycle. The heterogeneity and boundary extraction module calculates the local heterogeneity index of each source control system based on the degree of deviation between the source operating condition feature data and the global benchmark gradient direction. At the same time, it extracts the operating condition boundary envelope of each source control system in the feature space and calibrates its risk spillover coefficient based on the historical energy efficiency feedback value of each source control system. The target domain distance accumulation module acquires the initial operating condition feature data of the target control system, calculates the heterogeneous weighted distance of the initial operating condition feature data falling within the corresponding operating condition boundary envelope of each source control system, and accumulates the heterogeneous weighted distances of all source control systems to obtain the total heterogeneous distance value. The safety compression factor generation module generates a safety barrier compression factor based on the total heterogeneous distance value. The safety barrier compression factor decreases monotonically as the total heterogeneous distance value increases. The conservative correction output module multiplies the compression factor with the original dynamic threshold and the original single-step control change amplitude to obtain the conservative correction threshold and the gradual control limit, and uses the conservative correction threshold and the gradual control limit as the initial execution parameters output of the target control system.

2. The AI-based dynamic threshold energy-saving control algorithm execution system according to claim 1, characterized in that, The calculation of the local heterogeneity index for each source control system specifically includes: Calculate the cosine of the deviation angle between the gradient update direction of each source control system and the global baseline gradient direction, and then calculate the coefficient of variation of the threshold adjustment step size of each source control system within a continuous control period. The product of the cosine of the deviation angle and the coefficient of variation is used as the local heterogeneity index of the source control system.

3. The AI-based dynamic threshold energy-saving control algorithm execution system according to claim 1, characterized in that, The process of calibrating the risk spillover coefficient based on the historical energy efficiency feedback values ​​of each source control system specifically includes: For each source control system, its historical operating condition characteristic data is reduced in dimensionality through principal component analysis, and the convex hull vertices are extracted as the set of operating condition boundary envelope points. Then, the proportion of the duration during which the historical energy efficiency feedback value of each source control system is lower than the preset energy efficiency benchmark is counted as the basic risk coefficient. The basic risk coefficient is multiplied by the corresponding local heterogeneity index and normalized to obtain the risk spillover coefficient.

4. The AI-based dynamic threshold energy-saving control algorithm execution system according to claim 3, characterized in that, The process of obtaining the boundary envelope point set of the working condition is as follows: Calculate the local reachability density of each historical working condition feature data point after dimensionality reduction, retain points with local reachability density greater than the global density mean as candidate boundary points, and then use the convolution wrapping method to connect the candidate boundary points in counterclockwise order to form an initial convex hull vertex sequence. Calculate the Euclidean distance between every two adjacent vertices in the initial convex hull sequence, and perform linear interpolation to fill in the points between adjacent vertex pairs whose distance is greater than three times the overall average spacing, to obtain the encrypted working condition boundary envelope point set.

5. The AI-based dynamic threshold energy-saving control algorithm execution system according to claim 1, characterized in that, The process of obtaining the total heterogeneous distance value specifically includes: For each source control system, calculate the shortest Euclidean distance from the initial operating condition characteristic data points to the operating condition boundary envelope of the source control system; Multiplying the shortest Euclidean distance by the product of the local heterogeneity index and the risk spillover coefficient of the corresponding source control system yields the heterogeneous weighted distance of the source control system. After sorting the heterogeneous weighted distances of all source control systems in ascending order of their values, the maximum and minimum values ​​in the sequence are removed, and the remaining heterogeneous weighted distances are summed. The sum is taken as the total heterogeneous distance value.

6. The AI-based dynamic threshold energy-saving control algorithm execution system according to claim 1, characterized in that, The generation of the safety barrier compression factor based on the total heterogeneous distance specifically includes: The nonlinear decay base is obtained by convolving the reciprocal of the total heterogeneous distance value with the negative exponent of the base of the natural constant. The inertial damping coefficient is obtained by weighting the nonlinear decay base with the historical compression factor output from the first three control cycles and correcting it based on the total heterogeneous distance. An arctangent stretching transformation is performed on the inertial damping coefficient so that the output value decreases monotonically as the total heterogeneous distance increases, and the result is used as the safety barrier compression factor.

7. The AI-based dynamic threshold energy-saving control algorithm execution system according to claim 6, characterized in that, The formation of the inertial damping coefficient specifically includes: Obtain the historical compression factor of the output of the previous three consecutive control cycles, assign a decay weighting coefficient that is inversely proportional to the duration of the current cycle to the historical compression factor and normalize it. The inertia estimate is obtained by multiplying the nonlinear attenuation base with the three normalized attenuation weighting coefficients and summing the results.

8. The AI-based dynamic threshold energy-saving control algorithm execution system according to claim 7, characterized in that, The formation of the inertial damping coefficient further includes: calculating the absolute difference between the preliminary inertial estimate and the inertial damping coefficient output in the previous control cycle, multiplying the absolute difference by the hyperbolic tangent of the total heterogeneous distance as a correction, and superimposing it on the preliminary inertial estimate. The resulting value is used as the inertial damping coefficient for the current control cycle.

9. The AI-based dynamic threshold energy-saving control algorithm execution system according to claim 1, characterized in that, The method of using a conservative correction threshold and a gradual control limit as the initial execution parameters output for the target control system specifically includes: The compression factor is subjected to square root and cube root operations respectively. The original dynamic threshold is multiplied by the square root operation result to obtain the first correction value. The original dynamic threshold is multiplied by the cube root operation result to obtain the second correction value. The arithmetic mean of the first correction value and the second correction value is calculated as the conservative correction threshold. The temporary amplitude limit is obtained by multiplying the original single-step control change amplitude by the product of the safety barrier compression factor and the arctangent function value of the total heterogeneous distance.

10. The AI-based dynamic threshold energy-saving control algorithm execution system according to claim 9, characterized in that, The method of using a conservative correction threshold and a gradual control limit as the initial execution parameters output for the target control system also includes: The temporary control limit is weighted and averaged with the gradual control limit output from the previous control cycle, where the weight of the current temporary control limit is greater than the weight of the output from the previous cycle. The resulting weighted average is used as the gradual control limit output for the current control cycle.

Images