Building energy consumption optimization method based on artificial intelligence

Abstract

The invention relates to the field of building energy automation control, and discloses a building energy consumption optimization method based on artificial intelligence, which comprises the following steps: acquiring an energy loop state, generating a reference control instruction, superposing a detection signal of which the amplitude is lower than the upper limit of a mechanical dead zone, and sending the superposed detection signal to an execution mechanism; extracting a phase difference of the feedback response relative to the detection signal by using orthogonal correlation logic; according to the matching relation between the instruction gradient direction and the phase difference curvature evolution trend, an asymmetric damping rule is configured for a loop, and control parameters are corrected, the asymmetric physical characteristics of building thermal inertia are matched, phase mismatch between the instruction and physical response is eliminated, energy backflow and overshoot oscillation generated in the adjusting process are restrained, and the stability of the building thermal inertia is improved. Therefore, energy consumption adjustment is converted from oscillation approximation to smooth cut-in, and on-line identification of the deterioration state of the actuating mechanism is realized on the premise that hardware cost is not increased.

Description

Technical Field

[0001] This invention relates to a building energy consumption optimization method based on artificial intelligence, belonging to the field of building energy automation control technology. Background Technology

[0002] Current building automation systems employ proportional-integral-derivative (PID) control algorithms to maintain the thermodynamic stability of the controlled environment, while industrial control systems utilize adaptive models to adjust control parameters online for energy efficiency configuration. However, the increased frequency of personnel movement and transient thermal radiation within the controlled space leads to dynamic nonlinear drift characteristics in building thermal inertia. The thermodynamic sensitivity of the controlled object evolves with the physical boundary. Existing adaptive control methods, based on quasi-static model assumptions, presuppose that the thermodynamic sensitivity of the controlled object remains constant within the sampling period. Phase lag uncertainty exists between the actuator's command output and energy feedback reception, causing a mismatch between control commands and physical responses. The system increases the proportional gain to compensate for environmental deviations, but the physical medium conduction hysteresis effect causes control energy to be released at incorrect times, resulting in energy return and frequent actuator movements, increasing mechanical losses.

[0003] In addition to hardware constraints, software modeling and control logic also attempt to incorporate artificial intelligence solutions, but there are limitations in practical engineering applications. For example, Chinese invention patent CN113048626A discloses a method and device for optimizing building energy consumption and a readable storage medium. It uses a biased ReLU network to model environmental parameters and combines a genetic algorithm to solve the control sequence. The prediction mechanism based on the black box neural network depends on the coverage of training samples. When facing non-stationary load conditions, the static model is difficult to capture real-time thermal inertial phase drift. The global optimization algorithm consumes a lot of computing power in the embedded controller. The model update lag is prone to induce phase trajectory divergence, causing the feedback loop to oscillate under extreme disturbances, which restricts the stability of the system throughout its entire life cycle.

[0004] Therefore, the technical problem to be solved by this invention is how to accurately observe the real-time thermodynamic sensitivity of the energy loop and correct the control parameters to eliminate the effects of phase lag while maintaining the simplicity of the control architecture. Summary of the Invention

[0005] To address the problems mentioned in the background art, the technical solution of the present invention is as follows: A building energy consumption optimization method based on artificial intelligence, comprising the following steps: Step S1: Obtain the real-time operating status of the building energy loop and generate a reference control command as the main excitation flow; Step S2: Superimpose an excitation signal with a preset frequency characteristic and an amplitude lower than the upper limit of the mechanical dead zone of the actuator into the reference control command to generate a composite control command and send it to the actuator. Step S3: Synchronously collect feedback response data generated by the energy circuit under the excitation of composite control commands, and use orthogonal correlation logic to extract the response phase difference of the feedback response data relative to the excitation signal; Step S4: Calculate the first derivative of the reference control command over time to determine the command gradient direction, and calculate the second derivative of the response phase difference over time to characterize the dynamic sensitivity characteristics of the controlled object. Step S5: Execute the asymmetric damping control rule based on the dynamic sensitivity characteristics constrained by the command gradient direction: If the sign of the command gradient direction is consistent with the sign of the second derivative, it is determined that the energy loop is in the high inertia region. The dynamic damping parameter is increased according to the second derivative to offset the overshoot kinetic energy of the actuator through damping compensation; If the sign of the command gradient direction is opposite to the sign of the second derivative, it is determined that the energy loop is in the high sensitivity region. The dynamic damping parameter is decreased and the proportional gain is increased synchronously to compensate for the physical response lag of the controlled object. Step S6: Using the adjusted dynamic damping parameters and the proportional gain correction reference control command, an adaptive optimization decision is output to suppress regulation fluctuations.

[0006] Preferably, adjusting the dynamic damping parameter based on the second derivative includes the following steps: Step S21, extracting historical operating data of the energy circuit and setting a damping compensation benchmark based on the mechanical loss performance of the actuator at different load points; Step S22, in the high inertia region, using the second derivative as a nonlinear mapping input, dynamically adjusting the damping compensation benchmark based on the magnitude of the second derivative to generate a damping correction amount; Step S23, using the damping correction amount to perform asymmetric weighting processing on the damping ratio of the control circuit, increasing the damping before the actuator reaches the target set value, and suppressing the overshoot of the feedback response data.

[0007] Preferably, sending the composite control command to the actuator in step S2 includes the following steps: step S31, assigning a coprime modulation frequency to each energy circuit according to the physical topology distribution of each energy circuit; step S32, generating a unique excitation pulse for each energy circuit according to the modulation frequency as an excitation signal; step S33, performing coherent demodulation on the feedback response data at the same frequency as the modulation frequency to remove the same-frequency interference components caused by the physical coupling between energy circuits.

[0008] Preferably, after step S3, the following steps are also included: Step S41, extracting the waveform symmetry of the feedback response data on the time axis; Step S42, calculating the distortion rate D of the waveform symmetry deviating from the preset response model, using the formula: Where D is the distortion rate. The rising edge slope of the feedback response data, The falling edge slope of the feedback response data, The standard slope of the linear response is preset; Step S43: If the distortion rate D exceeds the preset threshold, it is determined that the actuator has entered the nonlinear dead zone, and the width of the nonlinear dead zone is determined based on the value of the distortion rate D; Step S44: The compensation bias corresponding to the width of the nonlinear dead zone is superimposed on the reference control command.

[0009] Preferably, after step S4, the following steps are also included: step S51, calculating the rate of change acceleration of the response phase difference; step S52, when the rate of change acceleration exceeds a preset mutation threshold, stopping the correction of the gain parameter of the reference control command; step S53, locking the current output offset of the actuator and entering the safety limit mode until the rate of change acceleration recovers to below the mutation threshold.

[0010] Preferably, step S3 further includes the following steps: step S61, analyzing the high-order harmonic characteristic values ​​of the corresponding excitation signal frequency in the feedback response data; step S62, if the high-order harmonic characteristic values ​​deviate from the preset linear range, determining that the actuator of the energy circuit has deteriorated in operating condition; step S63, reducing the control priority weight of the energy circuit in the global optimization decision according to the degree of deterioration in operating condition.

[0011] Preferably, extracting the response phase difference of the feedback response data relative to the excitation signal includes the following steps: Step S71, extracting the main frequency component of the feedback response data in the frequency domain; Step S72, calculating the time delay distribution of the main frequency component relative to the original frequency of the excitation signal; Step S73, converting the time delay distribution into a response phase difference characterizing the thermal hysteresis features of the controlled object.

[0012] Preferably, the method further includes the following steps: Step S81, monitoring the evolution curvature of the response phase difference over time; Step S82, when the evolution curvature reaches a preset threshold, switching the current control logic to the amplitude limiting damping mode; Step S83, in the amplitude limiting damping mode, suppressing control oscillations caused by physical boundary saturation by forcibly constraining the action amplitude of the actuator.

[0013] Preferably, the method further includes the following steps: Step S91, acquiring environmental data of the controlled object and identifying the type of environmental disturbance; Step S92, if the type of environmental disturbance is transient disturbance, pausing the update logic of the proportional gain; Step S93, using the phase mean value at the moment before the transient disturbance occurs to perform adaptive optimization decision output maintenance.

[0014] Preferably, the adaptive optimization decision output in step S6 includes the following steps: step S101, converting the corrected reference control command into a pulse width modulation signal; step S102, using the pulse width modulation signal to drive the corresponding frequency converter and regulating valve.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. In building energy consumption using artificial intelligence, the baseline control command is superimposed with the detection signal and the feedback response is collected synchronously to realize the transformation from passive deviation adjustment to active system identification; the orthogonal correlation algorithm is used to extract the phase characteristic offset of the feedback response relative to the detection signal, and the phase offset curvature is combined with the real-time thermodynamic sensitivity of the energy loop; the active detection captures the thermal inertial drift caused by transient changes in the building environment such as personnel flow or thermal radiation, so that the control command is in a phase leading state, and the damping ratio is pre-adjusted before the system approaches the equilibrium point to suppress the common oscillation overshoot of objects with large lag, so that the energy consumption adjustment process is transformed from oscillating approximation to smooth entry.

[0016] 2. By utilizing the coprime characteristics of prime frequencies to allocate the modulation frequency of each energy loop, a logic-level control channel segmentation is constructed to solve the coherent interference generated by multiple loops sharing the physical medium. After collecting feedback data, the interference components generated by physical coupling between energy loops are removed through coherent integral calculation at the same frequency, enabling independent observation under conditions where multiple controlled units share a cold source or air supply duct. This method improves the real-time performance and concurrency efficiency of multi-loop parallel control, ensures the accuracy of global optimization decisions, and enhances the stability of the system under severe full-load fluctuation conditions.

[0017] 3. By matching the control command change direction sensitive operator with the phase characteristic evolution trend, a heterogeneous dynamic damping operator is configured to solve the problem of control energy asymmetry caused by the unidirectionality of heat conduction; when the control direction is consistent with the inertial drift trend, the damping is increased to suppress actuator overshoot, and when the two are opposite, the damping is decreased to improve response sensitivity; for the heterogeneous adjustment of the energy gradient direction, the control law is matched with the differential thermal resistance characteristics of the heating or cooling process, eliminating low-frequency oscillations caused by thermodynamic sensitivity asymmetry, and reducing ineffective jitter and mechanical wear of the actuator. Attached Figure Description

[0018] Figure 1 This is a flowchart of the active detection and asymmetric damping adaptive optimization algorithm of the present invention; Figure 2 This is a graph showing the evolution of control parameters and temperature response under dynamic thermal inertia conditions according to the present invention. Figure 3 This is a closed-loop architecture diagram of the control logic of the present invention, which balances interference suppression and boundary safety. Detailed Implementation

[0019] The present invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0020] This invention provides an artificial intelligence-based building energy consumption optimization method, applied to the closed-loop control loop of industrial control systems or building automation systems. It addresses the phase mismatch between control commands and physical responses caused by thermal inertia-induced nonlinear drift in HVAC systems. The method primarily comprises five core processing stages: reference command generation, probe signal injection, orthogonal phase feature extraction, dynamic sensitivity mapping, and asymmetric damping correction. It acquires the real-time operating status of the building energy loop, including but not limited to return air temperature data collected by temperature sensors or supply and return water pressure difference data collected by pressure sensors. This real-time operating status is compared with a preset target setpoint, and a reference control command is generated using a proportional-integral-derivative algorithm. In the reference control command A composite control command is generated by superimposing an excitation signal with a preset frequency characteristic. The excitation signal is sent to the actuator. It is a sine wave or pseudo-random pulse sequence with constant amplitude and a single frequency. Its amplitude is set below the upper limit of the actuator's mechanical dead zone and does not cause fluctuations in the controlled environment temperature; for example, it is set to 1% to 3% of the valve's full opening. The actuator responds to the composite control command. The action causes a fluctuation in the controlled physical quantity. The sensor collects this fluctuation and feeds it back to the controller, forming feedback response data Y(t). Orthogonal correlation logic is used to extract the phase difference between the feedback response data Y(t) and the excitation signal. The processor internally constructs two frequencies related to the excitation signal frequency. The synchronous reference signals are the in-phase reference signals. and quadrature reference signals The real-time collected feedback response data Y(t) is compared with... and Multiplication is performed, and the product is integrated over the complete signal period T to obtain the in-phase component I and the quadrature component Q. This integration operation uses the principle of trigonometric orthogonality to filter and detect frequencies. Inconsistent environmental noise components, according to the formula Calculate the response phase difference This parameter characterizes the degree of thermal hysteresis and energy transfer delay characteristics of the controlled object under the current operating conditions.

[0021] Based on the matching relationship between the command gradient direction and the evolution trend of phase difference curvature, an asymmetric damping rule is configured for the loop and the control parameters are modified to calculate the reference control command. First derivative with time To determine the direction of the command gradient and simultaneously calculate the response phase difference. Second derivative with time To characterize the dynamic sensitivity features of the controlled object, when symbols and When the signs are consistent, the energy loop is determined to be in the high inertia region, according to... Numerical adjustment of dynamic damping parameters That is, at the original damping ratio Add to the basis of A proportional compensation term is used to offset the overshoot kinetic energy of the actuator through damping compensation. symbols and When the signs are opposite, the energy circuit is determined to be in the high-sensitivity region, and the dynamic damping parameter is lowered. And simultaneously increase the proportional gain To compensate for the physical response lag of the controlled object, the adjusted dynamic damping parameters are used. and proportional gain Correct reference control command The system outputs adaptive optimization decisions. For multi-loop parallel operation, a channel isolation mechanism based on prime frequency distribution is adopted, allocating coprime modulation frequencies to each energy loop according to the physical topology distribution of each energy loop. When generating composite control commands, each loop operates according to its assigned modulation frequency. A unique excitation pulse is generated as the excitation signal. When processing the feedback response data, the modulation frequency is adjusted accordingly. Coherent demodulation at the same frequency utilizes the principle of orthogonality to remove interference components at the same frequency caused by physical coupling between energy circuits.

[0022] The waveform symmetry of the feedback response data on the time axis is extracted to identify the nonlinear dead zone state of the actuator. The rising edge slope of the feedback response data waveform is calculated respectively. and the slope of the falling edge Calculate the distortion rate D of the waveform symmetry deviating from the preset response model, using the following formula: ,in The standard slope for the linear response is preset. If the distortion rate D exceeds a preset threshold, the actuator is determined to have entered a nonlinear dead zone. The width of the nonlinear dead zone is determined based on the value of D, and the reference control command is applied. The compensation bias corresponding to the width of the nonlinear dead zone is superimposed in the middle. Drive the actuator across the dead zone; calculate the rate of change of the response phase difference acceleration. To monitor the stability boundary of the system, when When the preset mutation threshold is exceeded, the correction of the gain parameter of the reference control command is stopped, the current output offset of the actuator is locked, and a safety limit mode is entered until... The mechanism restores the system to below the mutation threshold to prevent the adaptive algorithm from diverging under physical boundary saturation or extreme disturbance conditions, ensuring the safe operation of the control system. It analyzes the high-order harmonic characteristic values ​​of the corresponding excitation signal frequency in the feedback response data. If the high-order harmonic characteristic values ​​deviate from the preset linear range, it determines that the actuator of the energy loop has deteriorated in operating status, and reduces the control priority weight of the energy loop in the global optimization decision according to the degree of deterioration. It identifies the type of environmental disturbance. If it is a transient disturbance, it suspends the update logic of the proportional gain, uses the phase mean value of the moment before the transient disturbance to maintain the output, and converts the corrected reference control command into a pulse width modulation signal to drive the corresponding frequency converter and regulating valve.

[0023] Example 1: In a large commercial complex central air conditioning water system application scenario with three parallel-operating variable frequency chilled water pumps and multiple air handling units, the system faces the condition of drastic load fluctuations in summer. When the return water temperature rises non-linearly and rapidly due to the influence of personnel density and solar radiation during morning start-up or midday load peak, the system executes the following control flow: real-time return air temperature data is collected using a temperature sensor installed on the main return water pipe, and this data is compared with the target setpoint of 24°C. A reference control command is generated through a proportional-integral-derivative algorithm. The controller generates a sinusoidal detection signal with a frequency of 0.11Hz and an amplitude of 2% of the valve's full stroke, and superimposes it onto the reference control command. Generate composite control commands The system drives the chilled water valve to generate a slight vibration near its current opening degree. A differential pressure sensor installed on the supply and return water pipelines synchronously collects the network differential pressure fluctuation data caused by this slight vibration as a feedback response Y(t). The processor receives the feedback response data Y(t) and, using a 0.11Hz sine and cosine reference signal synchronized with the detection signal, performs orthogonal correlation demodulation on the feedback response data Y(t) to calculate the in-phase component I and the quadrature component Q. The system then uses the formula... Solve the phase difference between the feedback response and the probe signal. ,in The system characterizes the thermal hysteresis of the controlled object and simultaneously calculates the first derivative of the reference control command with time. To determine the direction of the command gradient and calculate the second derivative of the response phase difference over time. To characterize the dynamic rate of change of thermal inertia.

[0024] When the system detects Positive value and When the value is positive, the controlled object is determined to be in a high-inertia boost zone. This state corresponds to the physical process where an increase in valve opening does not immediately cause a drop in return water temperature, and energy accumulates due to the inertia of the water's thermal capacity. The adaptive controller executes asymmetric damping correction logic based on... Numerical linear adjustment of dynamic damping parameters A damping term proportional to inertial acceleration is introduced into the control loop. This damping reduces the growth rate of the control command and suppresses the accumulation of kinetic energy in the actuator at the moment of inertial release. When the valve opening approaches the target position and the rate of change of the feedback phase difference tends to level off, the system dynamically adjusts the damping parameters. The system reverts to the baseline value to maintain steady-state operation.

[0025] Example 2: To verify the control accuracy and channel isolation effectiveness of the artificial intelligence-based building energy consumption optimization method of the present invention under multi-loop physical coupling conditions, as well as the stability of the asymmetric damping correction mechanism in actual thermal inertia scenarios, a dedicated hardware-in-the-loop simulation verification platform was built. The test platform consists of a real-time simulator, a programmable logic controller (PLC), and a physical hydraulic loop simulation device. The real-time simulator runs a high-precision building thermal environment dynamic model, which integrates nonlinear heat transfer equations and fluid network hydraulic coupling modules to simulate the temperature response of three parallel air conditioning terminal loops under different load disturbances. The physical simulation device includes three controlled regulating valve actuators and corresponding fluid resistance components. The position feedback signals of the actuators are connected to the simulator to form a hardware-in-the-loop closed-loop test system. To simulate the signal interference and environmental disturbances of real industrial sites, the test environment is preset with Gaussian white noise with a signal-to-noise ratio of 20dB, and random temperature fluctuations with an amplitude of 0.5℃ are superimposed.

[0026] In the experiment, the sampling period was set to 1 second to satisfy Shannon's sampling theorem and take into account the controller's computational load. To verify the multi-loop channel isolation mechanism, coprime detection frequencies were assigned to the three parallel loops: 0.11Hz for loop 1, 0.13Hz for loop 2, and 0.17Hz for loop 3. The amplitude of the detection signal for each loop was set to 2% of the valve's full stroke. After the experiment started, the system ran for 30 minutes in the baseline PID mode without activating the optimization algorithm of this invention, and then switched to the adaptive optimization mode of this invention for 30 minutes. During this period, a step load disturbance was injected into each loop through the simulator, and the dynamic response curve of the return water temperature and the phase difference data demodulated from each loop were recorded. To compare and verify the effectiveness of the scheme of this invention, three sets of experimental conditions were set: control group: using the traditional PID control strategy, No detection signal was superimposed; adjustment was based solely on the return water temperature deviation. The experimental group employed the optimized method of this invention, namely, superimposing detection signals with coprime frequencies and activating an asymmetric damping correction mechanism based on phase difference. The interference group used a traditional PID control architecture but forcibly superimposed detection signals, with all three loops using the same detection frequency of 0.11Hz to simulate co-frequency interference without channel isolation. Table 1 shows the comparison data of key performance indicators for each loop under different control modes. Temperature overshoot refers to the maximum deviation of the return water temperature from the target value after a step change in load. Stabilization time refers to the time required for the return water temperature to re-enter and remain within the ±0.2℃ error band of the target value from the start of the disturbance. Phase demodulation signal-to-noise ratio refers to the ratio of the amplitude of the demodulated detection signal's phase characteristic to the amplitude of the background noise.

[0027] Table 1: Example Table of Key Performance Indicator Comparison Data

[0028] Experimental data shows that in the control group (traditional PID mode), the temperature overshoot reached 1.8℃ due to thermal inertia hysteresis, and the settling time was relatively long (420s). In the interference group, when all three loops used the same-frequency detection signal of 0.11Hz, the phase demodulation signal-to-noise ratio was only 3.2dB due to crosstalk caused by physical coupling. This resulted in the controller being unable to correctly identify the thermal inertia characteristics of each loop, causing the system to oscillate and diverge, and the temperature overshoot further deteriorated to 2.1℃. In contrast, the experimental group using the method of this invention achieved effective channel isolation through prime frequency distribution, and the phase demodulation signal-to-noise ratio was improved to 18.5dB. This indicates that each loop can extract its own physical response characteristics in a high-noise and mutually coupled environment. Based on the asymmetric damping correction mechanism triggered by this characteristic, the temperature overshoot was reduced to 0.4℃, and the settling time was shortened to 150 seconds. The data confirms that this invention, through the synergistic effect of active detection and asymmetric damping, can suppress multi-loop crosstalk while improving the control quality and response speed of the system for nonlinear thermal inertial objects.

[0029] Example 3: This example combines Figures 1 to 3 This section describes an artificial intelligence-based method for optimizing building energy consumption, such as... Figure 1 As shown, starting from acquiring the real-time operating status of the energy loop, by collecting return air temperature and supply and return water pressure difference data, the process enters the stage of superimposing the reference control command and the detection signal to generate an excitation flow below the upper limit of the mechanical dead zone. This process is supported by the coprime frequency channel isolation module. This module eliminates coupling interference based on prime frequency allocation and calculates the phase difference between the feedback response and the excitation through orthogonal correlation logic phase extraction, and performs dynamic sensitivity characteristic calculation, that is, calculates the second derivative of the command gradient direction and the phase difference. Based on the calculation of the compensation bias by the waveform distortion rate using the actuator dead zone compensation module, the asymmetric damping rule correction is performed. The parameters are adjusted according to the high inertia or high sensitivity zone, and finally the adaptive optimization decision is output and converted into a pulse width modulation signal to drive the execution.

[0030] like Figure 2 As shown in the figure, the horizontal axis represents time (seconds), the left vertical axis represents temperature (°C), and the right vertical axis represents control parameters. The legend shows that solid lines represent temperature response, dashed lines represent changes in damping parameters, and dotted lines represent changes in proportional gain. This figure illustrates that within a time range of 0 to 288 seconds, as the temperature response fluctuates from 24.0°C and eventually stabilizes, the system dynamically adjusts the initial values ​​of the damping parameter and proportional gain to 1.0, causing the control parameter value corresponding to the right vertical axis to dynamically evolve within the range of 0 to 2.0. Figure 3 As shown, the logic architecture takes the smoothing of building energy consumption as its core objective and optimizes system performance by suppressing overshoot oscillations and energy backflow. Its main functional pillars include active detection signal injection and orthogonal correlation phase extraction under the thermodynamic feature recognition dimension; dead zone self-compensation bias and feedback waveform symmetry identification under the actuator nonlinear compensation dimension; coherent integral calculation and prime frequency coprime allocation under the multi-loop interference suppression dimension; and amplitude limiting damping safety mode and phase acceleration mutation monitoring under the stability boundary protection dimension.

[0031] Example 4: Targeting dynamic damping parameters in a control system and proportional gain In engineering applications, relying on empirical values ​​for tuning can lead to insufficient adaptive accuracy and divergence risks. This embodiment provides a standardized calibration procedure based on physical parameter identification and stability boundary constraints. This transforms the parameter determination process from an empirical black box into a deterministic path driven by objective physical characteristics. It clarifies the initial system state definition and baseline parameter calibration process. During the system's first run or quarterly maintenance, the controller executes an offline calibration program, placing the actuator in a 50% linear operating range and injecting a step excitation signal with an amplitude of 5% of the full stroke. The sensor records the step response data of the return water temperature at a sampling frequency of 100Hz. The processor uses the least squares method to perform system identification on the response data and obtain the static gain of the controlled object. Time constant and pure lag time The system automatically calculates the reference proportional gain based on the improved Ziegler-Nichols tuning rule. and reference damping ratio , serving as the initial anchor point for online adaptive adjustment.

[0032] A parameter adaptive logic based on dynamic sensitivity characteristics is established, and the system calculates the first derivative of the reference control command in real time. Second derivative of the phase difference with response When the system is in the high inertia boost region At that time, dynamic damping parameters Real-time calculations are performed based on the following physical constraint model: ,in, The physical limit value of phase acceleration measured by offline calibration. The model establishes a linear mapping relationship between the damping compensation strength and the normalized index of inertial acceleration, using a preset damping enhancement coefficient ranging from 0.5 to 1.0. This ensures that the damping intervention strength is limited by the physical boundaries of the system, and is applied when the system is determined to be in a high-sensitivity demand zone. At that time, proportional gain Follows the adaptive law based on phase lag compensation: in, For real-time monitoring of the response phase difference, As the gain compensation coefficient, this logic compensates for physical transmission delay by enhancing the proportional gain, thereby improving the system's dynamic response bandwidth. A parameter-safe envelope mechanism based on Lyapunov stability theory is constructed, with the system pre-setting a stability domain centered on the reference parameter. Defined as: In each control cycle, if the parameter value output by the optimization algorithm exceeds the stability region... The controller clamps it to the boundary value.

[0033] Example 5: During the engineering configuration phase before the system is officially put into closed-loop operation, a standardized actuator characteristic mapping procedure is executed to establish a defined functional relationship between the feedback waveform distortion rate and the mechanical dead zone width. The controller outputs a 0.05Hz triangular wave scanning signal covering the entire stroke to the actuator, and simultaneously records valve stem position feedback data and pipeline flow response data. By analyzing the hysteresis loop area and the rate of change of the slope of the velocity zero-crossing point in the response curve, the system quantifies the inherent mechanical hysteresis error of the current actuator and generates a stored waveform distortion rate D and compensation offset. The lookup table of the mapping relationship, this discrete dataset serves as a static reference benchmark for subsequent online dead zone compensation logic, ensuring that the compensation values ​​generated by the system implicitly contain a precise definition of the hardware physical boundary rather than a random estimate. After the hardware mapping is completed, the system enters a 24-hour continuous background monitoring mode to construct a statistical baseline of phase trajectory stability boundary based on the target building load characteristics.

[0034] During this period, the controller injects a probe signal but does not perform control parameter correction; it only records the response phase difference under normal load fluctuation conditions. and its second derivative The time-series data, the processor processes the acquired phase acceleration dataset Perform statistical analysis and calculate its arithmetic mean. and standard deviation And the phase trajectory mutation threshold used for the safety triggering logic. Set as This statistical setting method anchors the trigger boundary of the safety mechanism above the probability limit of normal operation of a specific energy circuit, preventing the protection logic from being falsely triggered by normal environmental noise.

[0035] Example 6: To ensure the adaptive accuracy and control stability of the system under different physical objects and environmental conditions, a standardized pre-deployment calibration procedure was performed before formal operation. This aimed to obtain key physical characteristic parameters of the controlled object through offline engineering experiments, thereby determining the initial state baseline of the adaptive control algorithm. In the initial stage of system deployment, actuators, such as chilled water regulating valves, were placed in manual control mode, and three operating points were set with opening degrees of 20%, 50%, and 80%. After stabilization at each operating point, a step excitation signal with an amplitude of 5% of the full stroke was injected into the actuator, and the step response curve of the controlled physical quantity, such as return water temperature, was recorded at a sampling frequency of 100Hz. By performing system identification on the response data, the static gain under different operating conditions was analyzed. Time constant and pure lag time .

[0036] Based on the acquired physical parameters, the key initial parameters of the control algorithm, including the reference proportional gain, are determined using a pre-defined engineering optimization logic. Set to meet the gain margin and phase margin The critical value is given by the formula: Reference damping ratio Based on the critical damping condition, the value is typically set between 0.707 and 1.0 to ensure that the initial response of the system has no significant overshoot. To eliminate the risk of resonance between the probe signal frequency and the system's natural frequency, a frequency sweep test is required. The system output frequency range covers a sinusoidal probe signal from 0.01 Hz to 1.0 Hz, and the amplitude-frequency response characteristics of the controlled physical quantity are monitored simultaneously. By analyzing the amplitude-frequency response curve, the resonant frequency point of the system is identified. The frequency of the active excitation signal used for online detection To set a safe value that avoids the resonant frequency band, it is usually selected as [value missing]. This ensures the effectiveness of the detection while avoiding inducing unstable oscillations in the system.

[0037] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0038] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An artificial intelligence-based building energy consumption optimization method, characterized by, The method comprises the following steps: Step S1, obtaining the real-time running state of the building energy loop, generating a reference control instruction as a main excitation flow; Step S2, superimposing an excitation signal with a preset frequency characteristic and an amplitude lower than the upper limit of the mechanical dead zone of the actuator on the reference control instruction, generating a composite control instruction and sending it to the actuator; Step S3: Synchronously acquire feedback response data generated by the energy circuit under the excitation of the composite control command, and extract the phase difference of the feedback response data relative to the excitation signal using orthogonal correlation logic; establish a phase-thermal hysteresis physical characteristic mapping model; and synchronously acquire the detection frequency. Phase difference with response Calculate the physical transmission delay time The inherent time constant of the loop, obtained by least squares identification during the offline calibration phase, is used. Define the hysteresis sensitivity operator ,like A deviation from the steady-state range of 0.8 to 1.2 triggers an adaptive clamping mechanism for the control gain, utilizing... The evolution gradient of the asymmetric damping weight is corrected by the first-order rate of change, so as to achieve quantitative alignment between the damping compensation intensity and the thermodynamic transport hysteresis. Step S4, calculating the first derivative of the reference control instruction with respect to time to determine the instruction gradient direction, and calculating the second derivative of the response phase difference with respect to time to represent the dynamic sensitivity characteristic of the controlled object; Step S5, executing an asymmetric damping control rule based on the instruction gradient direction to constrain the dynamic sensitivity characteristic: if the sign of the instruction gradient direction is consistent with the sign of the second derivative, it is determined that the energy loop is in a high inertia zone, and the dynamic damping parameter is adjusted according to the second derivative to compensate for the overshoot of the actuator through damping compensation; if the sign of the instruction gradient direction is opposite to the sign of the second derivative, it is determined that the energy loop is in a high sensitivity zone, and the dynamic damping parameter is adjusted and the proportional gain is simultaneously adjusted to compensate for the physical response lag of the controlled object; Step S6, using the adjusted dynamic damping parameter and the proportional gain to correct the reference control instruction, and outputting an adaptive optimization decision for suppressing regulation fluctuations.

2. The building energy consumption optimization method based on artificial intelligence according to claim 1, characterized in that, Adjusting the dynamic damping parameter according to the second derivative comprises the following steps: Step S21, extracting historical running data of the energy loop, and setting a damping compensation reference according to the mechanical loss performance of the actuator at different load points; Step S22, in the high inertia zone, inputting the second derivative as a nonlinear mapping, dynamically adjusting the damping compensation reference according to the size of the second derivative, and generating a damping correction amount; Step S23, using the damping correction amount to perform asymmetric weighting processing on the damping ratio of the control loop, and increasing the damping before the actuator reaches the target set value.

3. The building energy consumption optimization method based on artificial intelligence according to claim 1, characterized in that, Sending the composite control instruction to the actuator in step S2 comprises the following steps: Step S31, according to the physical topology distribution of each energy loop, allocating a co-prime modulation frequency to each energy loop; Step S32, generating an excitation pulse unique to each energy loop according to the modulation frequency as the excitation signal; Step S33, coherently demodulating the feedback response data at the same frequency as the modulation frequency to remove the same frequency interference components generated due to the physical coupling between the energy loops.

4. The building energy consumption optimization method based on artificial intelligence according to claim 1, characterized in that, After step S3, the following steps are further included: step S41, extracting waveform symmetry of the feedback response data on the time axis; step S42, calculating a distortion rate D of the waveform symmetry deviating from a preset response model, with a formula being: wherein D is the distortion rate, is a rising edge slope of the feedback response data, is a falling edge slope of the feedback response data, is a preset linear response standard slope; step S43, if the distortion rate D exceeds a preset threshold, determining that the actuator enters a nonlinear dead zone, and determining a width of the nonlinear dead zone based on a magnitude of the distortion rate D; and step S44, superimposing a compensation bias corresponding to the width of the nonlinear dead zone in the reference control instruction.

5. The building energy consumption optimization method based on artificial intelligence according to claim 1, characterized in that, After step S4, the following steps are further included: Step S51, calculating the acceleration of the change rate of the response phase difference; Step S52, when the acceleration of the change rate exceeds a preset mutation threshold, stopping the correction of the gain parameter of the reference control instruction; Step S53, locking the current output offset of the actuator, and entering a safety limiting mode until the acceleration of the change rate returns to below the mutation threshold.

6. The building energy consumption optimization method based on artificial intelligence according to claim 1, characterized in that, Step S3 further comprises the following steps: Step S61, analyzing the high-order harmonic characteristic value corresponding to the frequency of the excitation signal in the feedback response data; Step S62, if the high-order harmonic characteristic value deviates from a preset linear interval, it is determined that the actuator of the energy loop has running state degradation; Step S63, according to the running state degradation degree, reducing the control priority weight of the energy loop in the global optimization decision.

7. The building energy consumption optimization method based on artificial intelligence according to claim 1, characterized in that, The response phase difference of the feedback response data relative to the excitation signal comprises the following steps: step S71, extracting a main frequency component of the feedback response data in a frequency domain; step S72, calculating a time delay distribution of the main frequency component relative to an original frequency of the excitation signal; and step S73, converting the time delay distribution into the response phase difference representing a thermal hysteresis characteristic of the controlled object.

8. The building energy consumption optimization method based on artificial intelligence according to claim 7, characterized in that, Further comprising the following steps: Step S81, monitoring an evolution curvature of the response phase difference over time; step S82, switching the current control logic to a limit damping mode when the evolution curvature reaches a preset threshold; and step S83, in the limit damping mode, suppressing control oscillation caused by saturation of a physical boundary by forcibly restricting a motion amplitude of an actuator. 9.The building energy consumption optimization method based on artificial intelligence of claim 1, wherein, Further comprising the following steps: Step S91, acquiring environmental data of the controlled object and identifying an environmental disturbance type; step S92, if the environmental disturbance type is a transient disturbance, suspending an update logic of a proportional gain; and step S93, executing output retention of adaptive optimization decision making by using a phase mean value at a time instant before the transient disturbance occurs.

10. The building energy consumption optimization method based on artificial intelligence according to claim 1, characterized in that, The output adaptive optimization decision making in step S6 comprises the following steps: step S101, converting the modified reference control instruction into a pulse width modulation signal; and step S102, driving a corresponding variable frequency controller and a valve by using the pulse width modulation signal.

Images