Intelligent air conditioner energy-saving control method based on people number perception and load optimization

Abstract

The invention discloses an intelligent air conditioner energy-saving control method based on people number perception and load optimization. The method comprises the steps that static data and dynamic data are collected, wherein the dynamic data obtains the number of people indoors, indoor and outdoor temperatures and the set temperature of an air conditioner in real time through a camera and a temperature sensor; automatically generating an air conditioner turn-off or turn-on instruction according to whether the number of indoor people is zero; in an air conditioner starting state, if the number of people in a room is smaller than a preset number threshold value and the indoor temperature is lower than a preset temperature threshold value, entering a regulation and control mode, starting an air conditioner load model based on a PINN parameter identification method, traversing a plurality of candidate set temperatures, calculating a corresponding load, and selecting the temperature when the load is minimum as an optimal set instruction; and finally, executing a corresponding instruction through the infrared emission device to control the air conditioner to operate. According to the method, unmanned idling energy consumption can be effectively avoided, load minimization is achieved through dynamic optimization, the energy efficiency is remarkably improved while the comfort degree is guaranteed, and the method is suitable for air conditioner refrigeration energy saving in summer.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of intelligent control and building energy saving, and particularly relates to an intelligent air conditioner energy saving control method based on number perception and load optimization. BACKGROUND

[0002] With energy saving and emission reduction becoming a global consensus, energy consumption optimization of public building central air conditioning systems has become a key issue. In public buildings (such as classrooms, office buildings, and shopping malls), air conditioning systems usually adopt a centralized management mode of unified start-stop and fixed temperature setting. This mode is not flexible enough, often causing waste of energy and funds.

[0003] In the prior art, there are some attempts to improve. For example, some schemes use pyroelectric infrared sensors to detect the presence of the human body, but this method cannot accurately count the number of people and is easily disturbed by the ambient temperature; some schemes use WiFi probes or Bluetooth beacons to inversely deduce the number density through signal strength, but there is a risk of privacy leakage and the coverage is limited. In addition, the precision of traditional air conditioning control strategies is limited, for example, simply increasing the temperature by a fixed value cannot dynamically and accurately optimize according to the actual thermal characteristics of the building, the outdoor climate, and the performance of the air conditioner itself, and the energy saving potential is not deep, which belongs to a kind of extensive "empirical" control.

[0004] In summary, the prior art has not effectively solved the following core problems: how to build a non-contact, high-precision personnel dynamic perception, and on this basis, through a load calculation model, to realize dynamic and accurate optimization of air conditioning set points, so as to realize deep and intelligent energy saving control on the premise of ensuring basic comfort. SUMMARY

[0005] To solve the above problems, the present application discloses an intelligent air conditioner energy saving control method based on number perception and load optimization, which can accurately perceive the indoor personnel dynamic through non-contact means, and on this basis, use the air conditioning load model for dynamic optimization, which can effectively avoid the energy consumption of idle running without people, realize load minimization, and improve energy efficiency while ensuring comfort.

[0006] To achieve the above purpose, the technical scheme of the present application is as follows:

[0007] An intelligent air conditioner energy saving control method based on number perception and load optimization, comprising:

[0008] S1: data acquisition step:

[0009] Input static data: building area S, number threshold R1 for starting the control mode, indoor temperature threshold C1 for starting the control mode, and air conditioner historical load data set;

[0010] Collecting dynamic data: air conditioner set temperature CN, current indoor number of people R, indoor temperature C, outdoor temperature CO. S2: basic start-stop control step: when it is identified that the current indoor number of people R = 0, an air conditioner off instruction is generated; when it is identified that the current indoor number of people R changes from zero to at least one person, an air conditioner on instruction is generated.

[0011] S3: load calculation optimization step: when the air conditioner is in the on state, it is judged whether the air conditioner can enter the regulation mode; after entering the regulation mode, the preset air conditioner load model based on the PINN parameter identification method is started, a plurality of candidate set temperatures are traversed, and the load corresponding to each candidate set temperature is calculated using the load model; the smallest one of the calculated loads is selected, and the candidate set temperature corresponding to the smallest one is determined as the optimal set instruction.

[0012] S4: instruction execution step: executing the off or on instruction generated in step S2 or the optimal set instruction generated in step S3 through an infrared emission device, thereby controlling the operation of the air conditioner.

[0013] Further, the step S1 comprises:

[0014] S11: input static data should be set in advance, as preferred, the number of people threshold R1 for starting the regulation mode and the indoor temperature threshold C1 for starting the regulation mode can be set according to actual conditions, generally R1 is not less than 5 and C1 is not less than 23°. The start of the regulation mode of the present application refers to the operation of the air conditioner in the mode, which is excessive refrigeration, and can not greatly affect the comfort of indoor personnel under the condition of increasing the air conditioner set temperature.

[0015] S12: the scene image data of the indoor area is collected by the camera, and the indoor number of people R is obtained by processing through a number of people recognition algorithm.

[0016] S13: the indoor temperature C and the outdoor temperature CO are collected by the temperature sensors installed indoors and outdoors.

[0017] S14: the initial data of the air conditioner set temperature CN is 26°, which will be changed following the generated instruction.

[0018] S15: the collection interval of dynamic data is 2 minutes, and a certain time is needed to process data and generate control instructions after collecting dynamic data each time.

[0019] Further, the step S2 comprises:

[0020] S21: when it is identified that the current indoor number of people R = 0, an air conditioner off instruction (represented by “0”) is generated and directly transmitted to the infrared module.

[0021] S22: When the air conditioner is in the off state, the camera is still collecting data. If it is identified that the current indoor population R changes from zero to at least one person, an air conditioner on command (represented by "1") is generated and directly transmitted to the infrared module.

[0022] Further, the step S3 comprises:

[0023] S31: Determine whether the dynamic data indoor temperature C is less than the indoor temperature threshold C1 for starting the control mode and the current indoor population R is less than the population threshold R1 for starting the control mode. If both conditions are met, it is considered to enter the control mode.

[0024] S32: After entering the control mode, the air conditioner load equivalent heat parameter model is enabled. The input of the model is indoor temperature, outdoor temperature, building area, air conditioner set temperature, and the output is air conditioner load value. By inputting the air conditioner set temperature from the current air conditioner set temperature, increase 0.5° in turn until 30°, get multiple output air conditioner load values.

[0025] S33: By comparing the size algorithm, comparing the size of the above air conditioner load value, selecting the candidate set temperature corresponding to the minimum air conditioner load value to determine the optimal set instruction.

[0026] Further, the step S4 comprises:

[0027] S41: The infrared device can receive the control instruction and emit corresponding infrared rays to operate the air conditioner in real time.

[0028] Further, the air conditioner set temperature and the air conditioner running condition are both cooling conditions. The present application is only applicable to summer air conditioner running energy saving.

[0029] The beneficial effects of the present application are:

[0030] By using non-contact population identification technology (such as camera), the indoor personnel dynamics are monitored in real time, the automatic opening and closing control of the air conditioner is realized, the idle running energy consumption in the unattended state is effectively avoided, and the unnecessary power consumption is significantly reduced. At the same time, by using the air conditioner load equivalent heat parameter model, combining with indoor and outdoor temperature, building area and other parameters, multiple candidate set temperatures are dynamically traversed, and the optimal temperature instruction with the minimum predicted load is selected to ensure that the air conditioner meets the comfort requirement while minimizing the load, further improving the energy efficiency. And based on the population threshold and temperature threshold, it is intelligently judged whether to enter the control mode. When the cooling is excessive (such as fewer people and lower temperature), the set temperature is automatically adjusted to avoid excessive cooling, which not only guarantees the comfort experience of indoor personnel, but also realizes the reasonable use of energy. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 Flowchart of the method described in the present application;

[0032] Figure 2 YOLOv8 model framework diagram;

[0033] Figure 3 Air conditioning load model calculation flowchart;

[0034] Figure 4 First-order equivalent thermal parameter circuit diagram;

[0035] Figure 5 PINN air conditioning load model parameter identification framework diagram. DETAILED DESCRIPTION

[0036] The present application will be further illustrated below in conjunction with the accompanying drawings and specific embodiments, and it should be understood that the following specific embodiments are only used to illustrate the present application and not to limit the scope of the present application.

[0037] As shown in the figure, the intelligent air conditioning energy-saving control method based on people perception and load prediction according to the present application comprises:

[0038] S1: data acquisition step: input static data: building area S, people threshold R1 for starting the control mode, indoor temperature threshold C1 for starting the control mode, air conditioning historical load data set; collect dynamic data: air conditioning set temperature CN, current indoor number of people R, indoor temperature C, outdoor temperature CO. S2: basic start-stop control step: when it is identified that the current indoor number of people R=0, generate an air conditioning off instruction; when it is identified that the current indoor number of people R changes from zero to at least one person, generate an air conditioning on instruction.

[0039] S3: load calculation optimization step: when the air conditioner is in the on state, judge whether the air conditioner can enter the control mode; after entering the control mode, start the pre-set air conditioning load model, traverse multiple candidate set temperatures, and calculate the load corresponding to each candidate set temperature by using the load model; select one with the smallest value in the calculated load, and determine the candidate set temperature corresponding thereto as the optimal set instruction.

[0040] S4: instruction execution step: execute the off or on instruction generated in step S2 or the optimal set instruction generated in step S3 by the infrared emission device, so as to control the operation of the air conditioner.

[0041] As a specific embodiment, the step S1 comprises:

[0042] S11: The static data should be set in advance, and as preferred, the number threshold R1 for starting the control mode and the indoor temperature threshold C1 for starting the control mode can be set according to actual conditions, and generally, R1 is not less than 5 and C1 is not less than 23°. The starting control mode in the application refers to the operation of the air conditioner in the mode, which is excessive refrigeration, and can not greatly affect the indoor comfort under the condition of increasing the air conditioner set temperature.

[0043] S12: The scene image data of the indoor area is collected by the camera, and the indoor number R is obtained by processing through the number recognition algorithm.

[0044] S13: The indoor temperature C and the outdoor temperature CO are collected by the temperature sensor installed indoors and outdoors.

[0045] S14: The initial data of the air conditioner set temperature CN is 26°, which will be changed following the generated instructions.

[0046] S15: The collection interval of dynamic data is 2 minutes, and a certain time is needed to process data and generate control instructions after collecting dynamic data each time.

[0047] The number recognition algorithm in S12 is:

[0048] YOLOv8 is selected as the main deep learning model, and the YOLO series is widely welcomed due to its high speed and high accuracy, and is very suitable for real-time personnel detection tasks.

[0049] PyTorch is selected as the core framework of deep learning, which provides flexible API and powerful GPU acceleration, and is suitable for rapid development and testing of deep learning models.

[0050] PyCharm is used as the integrated development environment, which provides powerful code editing, debugging and project management functions.

[0051] Further, as Figure 2 , the step S11 comprises:

[0052] S111: The main part of the YOLOv8 number recognition algorithm includes a backbone network composed of five convolutional blocks and four C2f modules, a feature fusion network and a detection layer.

[0053] Some important hyperparameters used in the training of the YOLOv8 model and their settings are as follows: learning rate (lr0): 0.01, learning rate decay (lrf): 0.01, momentum: 0.937, weight decay: 0.0005, warm-up training period (warmup_epochs): 3.0, batch size (batch): 16.

[0054] As a specific embodiment, the step S3 comprises:

[0055] S31: When the system detects that the dynamic indoor temperature C is less than the indoor temperature threshold C1 for starting the regulation mode and the current indoor number R is less than the number threshold R1 for starting the regulation mode, the regulation mode is entered.

[0056] S32: The indoor temperature, the outdoor temperature, the building area and the air conditioner setting temperature data are input into the air conditioner load equivalent thermal parameter model, and the air conditioner load value can be obtained.

[0057] S33: The air conditioner load values under different air conditioner setting temperatures are compared, and the air conditioner setting temperature corresponding to the minimum load value is taken as the optimal setting temperature.

[0058] Further, as Figure 3 , 4 , the step S32 comprises:

[0059] S321: An equivalent RC dynamic circuit of the first-order ETP model is established as shown in Figure 4 The equivalent circuit model can be represented by the following differential equation, which is as follows:

[0060] In the formula, is the indoor temperature, is the outdoor temperature, Q is the refrigeration / heating capacity of the air conditioner, is the equivalent heat capacity, and R is the equivalent thermal resistance. The parameters to be identified by the PINN neural network are the equivalent thermal resistance and the equivalent heat capacity.

[0061] S322: The PINN (physical neural prior neural network) model is used to effectively identify the parameters of the above air conditioner load model, and the process is as shown in Figure 5 A variable f(t) measuring the degree of violation of the physical rule constraint of the neural network indoor temperature prediction value output to the air conditioner load model is defined, which can be calculated by the indoor temperature prediction value of the neural network and the differential term , and the calculation formula is as follows:

[0062] The purpose of the model is to gradually approach 0 for f(t) in the network training process, and a corresponding physical regularization term is designed in the loss function, so that the PINN training process can be guided to follow the constraint of the air conditioner load model physical rule by minimizing , and the specific expression is as follows: .

[0063] In the formula, a time point for calculating the physical constraint residual, a number of data points for calculating the physical regularization loss.

[0064] S323: The network structure and the loss function of the PINN are both divided into two parts. The network structure is composed of a traditional feedforward neural network part and a differential layer for automatic differentiation of the output of the NN. The loss function is composed of a conventional mean square error term of the true value and the predicted value and a physical regularization term. Therefore, the training and optimization process can also be divided into two parts of supervised learning and unsupervised learning. The supervised learning part is guided by the historical data set and the conventional loss function term , and the parameters of the NN part are adjusted through the back propagation algorithm. The corresponding unsupervised learning part is guided by the unlabeled data set and the physical regularization loss function term , and the parameters of the NN part are adjusted through the back propagation algorithm, and The value of R and C will also be reduced when better estimates of R and C are obtained. Therefore, the parameters R and C to be identified are set as the optimization parameters, and the optimal estimates of R and C can be obtained through the optimization process of the unsupervised learning part.

[0065] S324: After the identification parameters are obtained, as shown in the flowchart Figure 3 , when the input is indoor temperature, outdoor temperature, building area and air conditioner set temperature, the output is the air conditioner load value.

[0066] Further, S33 includes:

[0067] S331: Read the current air conditioner set temperature value, and add 0.5° based on this value as the new air conditioner set temperature to the load model to obtain the load value at the corresponding temperature. After one round, add 0.5° to the air conditioner set temperature, until the maximum cooling set temperature of the air conditioner.

[0068] S332: Compare the obtained load values, and select the air conditioner set temperature corresponding to the minimum value as the optimal set temperature instruction.

[0069] It should be noted that the above content only illustrates the technical idea of the present application and cannot limit the protection scope of the present application. For ordinary skilled persons in the technical field, they can make several improvements and refinements without departing from the principles of the present application, and these improvements and refinements all fall within the protection scope of the claims of the present application.

Claims

1. A smart air conditioning energy-saving control method based on human presence perception and load optimization, characterized in that, include: S1: Data acquisition steps: Input static data and acquire dynamic data; The static data includes building area S, threshold number of people in control mode R1, indoor temperature threshold number in control mode C1, and historical air conditioning load dataset; the dynamic data includes air conditioning set temperature CN, current number of people indoors R, indoor temperature C, and outdoor temperature CO. S2: Basic start-stop control steps: When the current number of people in the room R=0, generate an air conditioner shutdown command; when the current number of people in the room R changes from zero to at least one person, generate an air conditioner start command. S3: Load calculation optimization steps: When the air conditioner is on, determine whether the air conditioner can enter the control mode; after entering the control mode, start the preset air conditioner load model, traverse multiple candidate set temperatures, and use the load model to calculate the load corresponding to each candidate set temperature; Select the smallest value among the calculated loads and determine its corresponding candidate set temperature as the optimal set command; S4: Command execution steps: The air conditioner is controlled by executing the off or on command generated in step S2 or the optimal setting command generated in step S3 through the infrared transmitter.

2. The intelligent air conditioning energy-saving control method based on human presence perception and load optimization as described in claim 1, characterized in that, Step S1 includes: S11: Input static data, the threshold R1 for the number of people to activate the control mode is not less than 5, and the indoor temperature threshold C1 for activating the control mode is not less than 23℃. S12: Collect scene image data of the indoor area through a camera, and process it through a people recognition algorithm to obtain the current number of people R in the indoor area; S13: Indoor temperature C and outdoor temperature CO are collected via temperature sensors; S14: Set the initial value of the air conditioner set temperature CN to 26℃, and update it dynamically with control commands; S15: The dynamic data acquisition interval is 2 minutes. After each acquisition, the data is processed and control commands are generated.

3. The intelligent air conditioning energy-saving control method based on human presence perception and load optimization as described in claim 2, characterized in that, Step S2 includes: S21: When the number of people in the room is detected to be R=0, an air conditioner shutdown command is generated and transmitted to the infrared transmitter; S22: When the air conditioner is off, the camera is still collecting data. If it detects that the number of people in the room R changes from zero to at least one person, it generates an air conditioner turn-on command and transmits it to the infrared transmitter.

4. The intelligent air conditioning energy-saving control method based on human presence perception and load optimization as described in claim 1, characterized in that, Step S3 includes: S31: Determine whether the indoor temperature C in the dynamic data is less than the indoor temperature threshold C1 for starting the control mode, and whether the current number of people R in the room is less than the number threshold R1 for starting the control mode; if both conditions are met, then enter the control mode. S32: After entering the control mode, the equivalent thermal parameter model of air conditioning load based on the PINN parameter identification method is activated. The input of the model includes indoor temperature, outdoor temperature, building area and air conditioning set temperature, and the output is the air conditioning load value. By increasing the air conditioning set temperature from the current value by 0.5℃ to 30℃, multiple candidate set temperatures and their corresponding air conditioning load values ​​are obtained. S33: Compare the magnitudes of the multiple air conditioning load values ​​and select the candidate set temperature corresponding to the minimum air conditioning load value as the optimal set command.

5. The intelligent air conditioning energy-saving control method based on human presence perception and load optimization as described in claim 4, characterized in that, In step S32, the equivalent thermal parameter model of the air conditioning load is established based on a first-order equivalent thermal parameter circuit, and its differential equation is expressed as: in, Indoor temperature, Where is the outdoor temperature, and Q is the cooling / heating capacity of the air conditioner. R is the equivalent heat capacity, and R is the equivalent thermal resistance; the model identifies parameters through a physical information neural network, including the following steps: S321: Define physical constraint violation variables The calculation formula is as follows: S322: During PINN training, minimize the physical regularization loss function. This ensures that the model output conforms to physical laws; where, The time point used to calculate the physical constraint residuals. This represents the number of data points used to calculate the physical regularization loss. S323: Combining supervised and unsupervised learning, jointly optimize the neural network parameters and the parameters to be identified, R and C; S324: Based on the identified parameters, the model outputs the corresponding air conditioning load value when the indoor temperature, outdoor temperature, building area, and air conditioning set temperature are input.

6. The intelligent air conditioning energy-saving control method based on human presence perception and load optimization as described in claim 2, characterized in that, In step S12, the people recognition algorithm is implemented based on the YOLOv8 model, which includes a backbone network consisting of five convolutional blocks and four C2f modules, a feature fusion network, and a detection layer. The model training hyperparameters include: learning rate lr0=0.01, learning rate decay lrf=0.01, momentum=0.937, weight decay=0.0005, warmup epochs=3.0, and batch size=16.

Images