An energy consumption optimization method and data acquisition system for central air conditioners
By using convolutional calculation and empirical mode decomposition in the central air-conditioning system to extract the eigenmode function, combining the deep residual network for multi-step energy consumption prediction, and optimizing operating parameters through improved particle swarm optimization algorithm and association rules, the problems of poor energy consumption prediction and insufficient correlation of optimization parameters in the existing technology are solved, and more efficient energy consumption prediction and optimization effects are achieved.
Patent Information
- Application Number
- CN202210540529.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-05-17
- Publication Date
- 2025-05-27
- Estimated Expiration
- 2042-05-17
AI Technical Summary
The prior art has problems of poor multi-step prediction results and insufficient correlation of optimization parameters in terms of energy consumption prediction and optimization of central air-conditioning systems.
Convolutional calculation combined with empirical modal decomposition is used to extract multiple eigenmode functions, and a multi-channel data set is constructed for deep residual networks for multi-step energy consumption prediction. At the same time, niche law and asynchronous updates are introduced to improve the particle swarm optimization algorithm, and the correlation rules for operating parameters are obtained through the frequent mode growth algorithm as constraints for the optimization algorithm.
The accuracy and optimization effect of energy consumption prediction of central air conditioning system is improved, more accurate energy consumption prediction and more effective energy consumption optimization are achieved, and system energy consumption is reduced.
Smart Images
Figure CN114997044B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of energy prediction and optimization of central air-conditioning systems, and relates to an energy consumption prediction and real-time optimization method applicable to central air-conditioning. Background Art
[0002] As a typical air-conditioning device, the central air-conditioning system has a very high proportion of energy consumption in building energy consumption due to its large volume and long working hours. Therefore, energy-saving research on the central air-conditioning system is of great significance for energy conservation and emission reduction.
[0003] For energy-saving research on the central air-conditioning system, an accurate model is required first. Currently, there are mainly two methods: identifying and fitting with a few key parameters or building a neural network. For the complex system parameter identification method of central air-conditioning, it is often necessary to separately identify and fit multiple sub-devices and then integrate the results, which will lead to the accumulation of identification errors of each sub-device. The neural network generally uses the backpropagation neural network, and its fully connected characteristic makes it difficult to model for more variables. In addition, both are generally single-step prediction methods. Affected by the large time delay of the central air-conditioning, their prediction effects in actual working conditions are average. For system energy consumption optimization, as a strongly coupled system, the current optimization algorithms generally do not have the relevance of optimizing parameters, which will also cause the finally optimized parameters to deviate from the actual operating characteristics and affect the optimization effect. Summary of the Invention
[0004] Aiming at the problems of poor single-step prediction effect with few variables and poor actual optimization effect due to lack of parameter relevance in the background art, the present invention proposes a system energy consumption prediction method capable of multi-step long-time prediction, and integrates the correlation characteristics of operating parameters into the optimization algorithm.
[0005] The technical solution adopted to solve its technical problems is: an energy consumption optimization method for central air-conditioning. As a complex multi-variable non-linear system, the central air-conditioning adopts the convolution calculation method, which can effectively alleviate the calculation amount of the data set composed of multi-variables and numerous time steps. Limited by the receptive field, the ability of a single-channel convolution kernel to initially extract feature correlations is limited. Using empirical mode decomposition to extract multiple intrinsic mode functions and jointly constituting a multi-channel data set with the original data set is helpful for extracting feature correlations between parameters during convolution calculation. For the non-linear problem of multi-step prediction, the depth of the neural network is expanded through residual connection, which can effectively improve the prediction accuracy. For the particle swarm optimization algorithm, the niche rule is introduced and the asynchronous update method is used for improvement. At the same time, the frequent pattern growth algorithm is used to obtain the association rules between operating parameters, and these are added to the optimization as the constraint conditions of the particle swarm algorithm, endowing the optimization algorithm with parameter relevance. The built central air-conditioning operating parameter acquisition system is as Figure 1As shown in the figure, the flow chart of the overall system energy consumption prediction and optimization technology route is as Figure 2 shown. Therefore, the technical solution of the present invention is an energy consumption optimization method for a central air conditioner, and the method includes:
[0006] Step 1: Collect the inlet and outlet temperatures of the evaporator, the inlet and outlet temperatures of the condenser, the total chilled water flow rate, the total cooling water flow rate, the load rate of the chiller, the indoor ambient temperature, and the total system power in the central air conditioning system respectively. Among them, the indoor temperature is the average value of the temperatures on five floors, and the frequencies of the chilled water pump, the cooling water pump, and the cooling tower fan are obtained according to the equipment frequency converter, and a total of Q operating parameters are obtained. The specific value of Q is determined according to the number of equipment, because there may be multiple chilled water pumps, cooling water pumps, and cooling tower fans in the central air conditioner;
[0007] Step 2: Preprocess the Q operating parameters collected in Step 1;
[0008] Step 2.1: Normalize the data set;
[0009]
[0010] In the formula: is the i-th operating data in the j-th parameter after normalization, is the corresponding data before normalization, and are the minimum value and the maximum value in the j-th parameter respectively;
[0011] Step 2.2: Combine the operating ranges of the equipment parameters to eliminate the outlier data in each variable, and then use the density-based clustering method to cluster the data set in the Q-dimensional space to obtain an irregular cluster, and remove the remaining operating data points;
[0012] Step 2.3: In view of the asynchronous problem existing in the sensor acquisition, integrate the interval of all parameter data at one-minute time intervals, take the average value of multiple acquisition data within the same minute, and merge all parameters according to the time series. For the interval with missing acquisition data in the obtained data set, forward filling or backward filling is used;
[0013] Step 3: Model the data set preprocessed in Step 2 with a multi-channel deep residual network and predict the energy consumption;
[0014] Step 3.1: Combine the continuous 59-minute operating parameters with the operating parameters separated by 1 hour as the input part of one piece of data. The operating parameters separated by 1 hour are understood as the system set parameters, and the power for 60 consecutive minutes with the last time point of the input as the end point is used as the output part of this piece of data, and a mode of predicting the system power for the subsequent 60 minutes by combining 59-minute operating data with the set parameters is constructed;
[0015] Step 3.2: Decompose the data set obtained in Step 3.1 by using the empirical mode decomposition method to obtain the first three intrinsic mode function components. Each intrinsic mode function component contains local feature data of different time scales. Combine the data set obtained in Step 3.1 with the three intrinsic mode functions to construct a new data set with 4 channels;
[0016] Step 3.3: Divide the data set into a training set and a validation set according to a ratio of 9:1. The deep residual network is specifically adjusted based on the ResNet18 model: Remove the pooling layer in the middle of the ResNet18 model, and replace the last average pooling layer with a fully connected layer, which can avoid too fast dimensional reduction at the parameter level and make it difficult to effectively extract features; Considering the complexity of the model, a randomly deleted connection node layer is added to this fully connected layer to avoid overfitting of the model; And while ensuring a small convolution kernel, traverse and optimize the model layer parameters to obtain the deep residual network model structure, which uses a total of 19 non-linear operation layers. Figure 3 The dotted part in it is the downsampling convolution calculation with a stride of 2, keeping the same dimension as the main road;
[0017] Step 4: Mine potential association rules for the data set preprocessed in Step 2;
[0018] Step 4.1: Use the K-means clustering algorithm to perform data clustering on each feature parameter in the data set obtained in Step 2. The number of clusters is set according to the maximum average silhouette coefficient S and the smallest possible sum of squared errors SSE:
[0019]
[0020]
[0021] Where:
[0022]
[0023]
[0024] In the formula: n represents the total length of the data, S i represents the silhouette coefficient of the sample point x i , a i is the average distance between the sample point x i and other sample points in the cluster, called the cohesion; b i is the average distance between the sample point x i and all samples in the nearest cluster, called the separation; p is a sample point in a certain cluster C k , m i is the cluster center coordinate;
[0025] Step 4.2: Number the Q parameters in Step 1 in alphabetical order, and number the multiple clustering clusters obtained in each parameter numerically. For example, "a1" represents the first clustering interval of the evaporator inlet temperature, and convert each piece of data into the number of its respective interval;
[0026] Step 4.3: Process the numbered set obtained in Step 4.2 using the frequent pattern growth algorithm, set the minimum support to 5% of the dataset length, and the minimum confidence to 85%, to obtain a set of association rules with different lengths;
[0027] Step 5: Use an optimization model to optimize the energy consumption of the collected real-time data;
[0028] Step 5.1: Improve the particle swarm optimization algorithm, use the niche rule for population division. For particle x i , its Euclidean distance from other particles is:
[0029] d ik =||x i -x k ||, k = 1, 2, 3..., n
[0030] According to the set parameter ε 0 , when d ik <ε 0 , then add this particle to the niche group X c . During the process of updating particle attributes, the sharing of the entire population information degrades to sharing within their respective niche groups. The update formula for particles is:
[0031]
[0032]
[0033] In the formula: and are the velocity and position attributes of the i-th particle in the j-th population at the t-th iteration, ω is the inertia coefficient, c 1 and c 2 are the learning factors, and gBest j (t) are the current particle's optimal position and the population's optimal position respectively;
[0034] The particle population update method uses asynchronous update, that is, after each particle completes the attribute update, the individual optimal information and the population optimal information are updated, so that the next particle can obtain a better fitness value and position information when it is updated;
[0035] Step 5.2: Build an energy consumption optimization model for the central air-conditioning system. Use the deep residual network model trained in Step 3.3 as the objective function of the optimization algorithm, and use the normal range of the collected parameters combined with the association rules obtained in Step 4.3 as the constraint conditions of the optimization algorithm. The association rules are divided into antecedents and consequents. When the value range of the antecedent is satisfied during the parameter optimization process, the corresponding parameter value of the consequent must also be within the corresponding range of the consequent.
[0036] Step 5.3: Use the optimization model to predict the energy consumption of the real-time acquired data. On the premise of ensuring that the indoor temperature and the load rate of the chiller remain unchanged, adjust the remaining set values through the optimization algorithm, and find the working point with the lowest energy consumption in the next hour of the central air-conditioning under the condition of meeting the cooling demand.
[0037] Furthermore, the clustering method in Step 2.2 is as follows:
[0038] Step 2.2.1: Set the minimum number of points parameter Minpts and the neighborhood radius parameter Eps;
[0039] Step 2.2.2: Access an unprocessed point. If there is none, go to Step 2.1.4. Otherwise, find all nearby points within the distance of Eps from it. If the number is greater than Minpts, set this point as a core point and go to Step 2.1.3. Otherwise, set it as a noise point and repeat Step 2.1.2;
[0040] Step 2.2.3: Access an unprocessed point within the neighborhood Eps of the core point. If there is none, go back to Step 2.1.2. Otherwise, find all nearby points within the distance of Eps from it. If the number is greater than Minpts, set this point as a core point. Otherwise, set this point as a boundary point and repeat Step 2.1.3;
[0041] Step 2.2.4: Connect a line between all core points within the distance of Eps. Each group of connected core points forms a cluster;
[0042] Step 2.2.5: Add each boundary point to the cluster of the core point whose distance from it does not exceed Eps, and the process ends;
[0043] Furthermore, the empirical mode decomposition method in Step 3.2 is as follows: Determine the local extreme points of the input data x(t), fit the upper and lower envelope lines by cubic spline interpolation, and calculate the mean value m of the envelope lines 1 , subtract this mean value from the original signal to obtain the difference signal y 1 (t):
[0044] y 1 (t) = x(t) - m 1
[0045] Determine whether the difference signal satisfies the Intrinsic Mode Function (IMF) conditions:
[0046] (1) The number of local extreme points and zero-crossing points must be equal or differ by at most one;
[0047] (2) At any given time, the average value of the upper envelope of the local maxima and the lower envelope of the local minima is zero;
[0048] If the above conditions are not met, then use the difference signal y 1 (t) as the processing signal and continue with empirical mode decomposition until the IMF conditions are satisfied. At this time, the first IMF component obtained is denoted as c 1 (t), and separate this component from the original signal:
[0049] r 1 (t) = x(t) - c 1 (t)
[0050] Use the difference signal r 1 (t) as the initial signal and repeat the above steps to obtain the subsequent two IMF components c 2 (t) and c 3 (t).
[0051] A data acquisition system for the above energy consumption optimization method. The system includes: a Lora wireless sensing system, a Lora gateway, and a host computer attached to the central air-conditioning system; among them, the Lora wireless sensing system includes: 4 + X Lora temperature wireless sensing modules and 2 Lora wireless flow sensing modules; 4 of the 4 + X Lora temperature wireless sensing modules respectively collect: the outlet water temperature and inlet water temperature of the evaporator, the inlet water temperature and outlet water temperature of the condenser, and the remaining X Lora temperature wireless sensing modules are installed on the household side for collecting indoor temperature, where X corresponds to the number of households; the 2 Lora wireless flow sensing modules respectively collect: the total pipeline flow of chilled water and the total pipeline flow of cooling water; each module in the Lora wireless sensing system uses Lora wireless communication to transmit to the Lora gateway, and then the Lora gateway forwards it to the host computer, and the host computer uses the above-mentioned energy consumption optimization method for a central air-conditioning system to calculate and adjust the operating parameters of the central air-conditioning.
[0052] The present invention uses empirical mode decomposition to extract multiple intrinsic mode functions and jointly constructs a multi-channel dataset with the original dataset, which helps to extract the characteristics between parameters during convolutional calculation. For the non-linear problem of multi-step prediction, the depth of the neural network is expanded through residual connection, which can effectively improve the data processing accuracy and achieve the purpose of adjusting the operating parameters of the air conditioner to reduce the energy consumption of the air conditioner. Description of the Drawings
[0053] Figure 1 Schematic diagram of the operation parameter acquisition system for the central air - conditioning system
[0054] Figure 2 Technical flow chart of energy consumption prediction and optimization for the central air - conditioning system
[0055] Figure 3 Depth residual network structure after traversing and optimizing adjustment
[0056] Figure 4 Comparison chart of the central air - conditioning system's energy consumption before and after optimization Specific implementation method
[0057] The central air - conditioning data acquisition system will be further described below with reference to the attached drawings
[0058] Refer to Figure 1 , the central air - conditioning data acquisition system includes various Lora wireless sensor modules, a Lora gateway, and an upper - level storage computer, which is attached to the central air - conditioning system. Among them, the sensors include: 9 Lora temperature wireless sensor modules and 2 Lora wireless flow sensor modules; in the figure, Lora temperature wireless sensor modules 1, 2, 3, and 4 respectively collect the outlet temperature and inlet temperature of the evaporator, and the inlet temperature and outlet temperature of the condenser. The remaining 5 are installed on the user side to collect the indoor temperature. The Lora wireless flow sensor modules respectively collect the total pipeline flow of the chilled water and the total pipeline flow of the cooling water. The frequencies of 3 chilled - water pumps, 3 cooling - water pumps, and 2 cooling - tower fans are read through the motor frequency converters, and the total power is read through the unit controller. The data of all Lora wireless sensor modules are transmitted to the Lora gateway through Lora wireless communication, and then forwarded by the Lora gateway to the computer database
[0059] There are a total of 32084 existing data collected in the database. Each piece of data includes the inlet and outlet temperatures of the evaporator, the inlet and outlet temperatures of the condenser, the total chilled - water flow, the total cooling - water flow, the load rate of the chiller, the indoor environmental temperature, the total system power, the frequencies of 3 chilled - water pumps, the frequencies of 3 cooling - water pumps, and the frequencies of 2 cooling - tower fans. The total data set is normalized in each parameter layer according to step 2.1, and the density - based clustering method in step 2.2 is used to obtain an excellent data set with outliers removed, as shown in Table (1). Among them, the clustering results corresponding to the clustering parameters with all clusters being 1 are summarized. To ensure the quality of the data set, the combination of the smallest neighborhood radius Eps and the largest minimum number of points Minpts is selected. Corresponding to Eps = 0.26 and Minpts = 14 in No. 51, 66 outliers are removed, and the data is merged according to the one - minute interval to fill in the missing values, obtaining 32018 excellent data
[0060]
[0061] The dataset obtained in Step 2 is reorganized using Step 3.1 to obtain 31,840 combined data. In each piece of data, the input includes the operating parameters for the first 59 minutes and one operating parameter with an interval of 1 hour, and the output is the continuous system power value for the subsequent 60 minutes. After the input data is decomposed by empirical mode decomposition in Step 3.2, the first 3 IMF components obtained and the original dataset form a new dataset with 4 channels. The input dimension of each piece of data is (60, 16, 4), and the output dimension is (60, 1). After dividing the dataset, the training set has 28,656 pieces, and the validation set has 3,184 pieces. The ResNet18 residual network model is trained with an iteration number of 200 and a learning rate of 0.0001. The root mean square error of the obtained ResNet model on the validation set is 0.00499, and the average percentage error is 1.5765%, indicating a good prediction effect.
[0062] Association rule mining is performed on the dataset obtained in Step 2. First, the K-means algorithm is used to cluster the data of each feature. By adjusting the number of clusters, the corresponding average maximum silhouette coefficient S and sum of squared errors SSE are calculated. Based on the case where the silhouette coefficient S is as large as possible, the number of clusters corresponding to the smaller sum of squared errors SSE is selected to obtain the clustering interval situation of each feature, as shown in Table (2) below. Since the total system power consumption is the output result of the prediction model, it is not within the scope of mining, and the discretization and association rule mining are performed on the data of the remaining feature variables.
[0063]
[0064]
[0065] The dataset is discretely numbered according to the above intervals, and the frequent pattern growth algorithm in Step 4.3 is used for association rule mining. The minimum support is set to 5% of the dataset length, and the minimum confidence is set to 85% to obtain a set of association rules as shown in Table (3).
[0066]
[0067]
[0068] Taking the improved particle swarm algorithm in Step 5.1 as the optimization algorithm, the ResNet18 prediction model as the objective function, and the association rules in Table (2) as the constraint conditions, the optimization model in Step 5.2 is constructed. Arbitrarily select one piece of operating data and use the optimization model to adjust the operating parameters to obtain the optimization comparison results as shown in Figure 4 shown. The energy consumption in the entire interval is reduced by 4%, and the corresponding energy efficiency ratio improvement rate is 4.63%.
Claims
1. An energy consumption optimization method for central air conditioning, the method comprises: Step 1: Collect the operating parameters of the central air conditioning system respectively, including: the inlet and outlet temperatures of the evaporator, the inlet and outlet temperatures of the condenser, the total chilled water flow rate, the total cooling water flow rate, the load rate of the chiller, the indoor ambient temperature, and the total system power. The indoor temperature is the average value of the temperatures on five floors, and the frequencies of the chilled water pump, the cooling water pump, and the cooling tower fan are obtained according to the equipment frequency converter; Step 2: Preprocess the operating parameters collected in Step 1; Step 2.1: Normalize the data set; Where: is the i-th running data in the j-th parameter after normalization, is the corresponding data before normalization, and are the minimum and maximum values in the j-th parameter, respectively; Step 2.2: Combine the operating ranges of the equipment parameters to eliminate the outlier data in each variable, and then use the density-based clustering method to cluster the data set to obtain an irregular cluster and remove the remaining operating data points; Step 2.3: Integrate all parameter data in segmented intervals, take the average value of multiple collected data within the same time period, and combine all parameters according to the time series. For the intervals with missing collected data in the obtained data set, forward filling or backward filling is used; Step 3: Model and predict the energy consumption of the data set preprocessed in Step 2 using a multi-channel deep residual network; Step 3.1: Combine the operating parameters within consecutive N time periods with the operating parameters separated by a time interval T as the input part of a piece of data. The operating parameters at the time interval T are understood as the system set parameters, and the power for the consecutive time T with the end point at the last moment of the input is used as the output part of this piece of data, and a mode of predicting the system power for the subsequent time T by combining the operating data within consecutive N time periods with the set parameters is constructed; Step 3.2: Use the empirical mode decomposition method to decompose the data set obtained in Step 3.1 to obtain the first 3 intrinsic mode function components. Each intrinsic mode function component contains local feature data of different time scales, and the data set obtained in Step 3.1 is combined with the 3 intrinsic mode functions to construct a new 4-channel data set; Step 3.3: Divide the data set into a training set and a validation set according to 9:
1. The deep residual network is adjusted specifically based on the ResNet18 model: remove the pooling layer in the middle of the ResNet18 model, replace the last average pooling layer with a fully connected layer, add a randomly deleted connection node layer in this fully connected layer, and traverse and optimize the model layer parameters while ensuring that the convolution kernel is small to obtain the deep residual network model structure; Step 4: Mine potential association rules from the data set preprocessed in Step 2; Step 4.1: Use the K-means clustering algorithm to cluster each feature parameter in the data set obtained in Step 2. The number of clusters is set according to the maximum average silhouette coefficient S and the smallest possible sum of squared errors SSE: Where: Where: n represents the total length of the data, S i represents the silhouette coefficient of the sample point x i , a i is the average distance between the sample point x i and other sample points within the cluster, which is called the cohesion degree; b i is the average distance between the sample point x i and all samples in the nearest cluster, which is called the separation degree; p is a sample point in a certain cluster C k , m i is the cluster center coordinate; Step 4.2: Sequentially number the parameters in Step 1. The multiple clusters obtained in each parameter are numbered according to numbers, and each piece of data is converted into the number of its respective interval; Step 4.3: Use the frequent pattern growth algorithm to process the numbered set obtained in Step 4.2 to obtain a set of association rules with different lengths; Step 5: Use the optimization model to optimize the energy consumption of the collected real-time data; Step 5.1: Improve the particle swarm optimization algorithm. Use the niche rule to divide the population. For particle x i , its Euclidean distance from other particles is as follows: d ik = ||x i - x k ||, k = 1, 2, 3..., n According to the set parameter ε 0 , when d ik < ε 0 , then add this particle to the niche population X c . During the update process of particle attributes, the sharing of the entire population information decreases to sharing within each niche population, and the update formula of the particle is as follows: Wherein: and are the velocity and position attributes of the i-th particle in the j-th population at the t-th iteration, ω is the inertia coefficient, c 1 and c 2 are the learning factors, and are the optimal position of the current particle and the optimal position of the population, respectively; The particle swarm update method adopts asynchronous update, that is, after each particle completes the attribute update, the individual optimal information and the group optimal information are updated, so that a better fitness value and position information can be obtained when the next particle is updated; Step 5.2: Construct an energy consumption optimization model for the central air-conditioning system. Take the deep residual network model trained in Step 3.3 as the objective function of the optimization algorithm, and take the normal range of the collected parameters combined with the association rules obtained in Step 4.3 as the constraint conditions of the optimization algorithm. The association rules are divided into antecedents and consequents. When the value range of the antecedent is satisfied during the parameter optimization process, the corresponding parameter value of the consequent must also be within the corresponding range of the consequent; Step 5.3: Use the optimization model to predict the energy consumption of the real-time acquired data. On the premise of ensuring that the indoor temperature and the load rate of the chiller remain unchanged, adjust the remaining set values through the optimization algorithm, and find the working point with the lowest energy consumption of the central air-conditioning in the next 1 hour under the condition of meeting the cooling demand.
2. A method for optimizing the energy consumption of a central air-conditioning system according to claim 1, characterized in that the clustering method in the said Step 2.2 is: Step 2.2.1: Set the minimum number of points parameter Minpts and the neighborhood radius parameter Eps; Step 2.2.2: Visit an unprocessed point. If there is none, go to Step 2.1.
4. Otherwise, find all nearby points within the distance of Eps from it. If the number is greater than Minpts, set this point as a core point and go to Step 2.1.
3. Otherwise, set it as a noise point and repeat Step 2.1.2; Step 2.2.3: Visit an unprocessed point within the neighborhood Eps of the core point. If there is none, go back to Step 2.1.
2. Otherwise, find all nearby points within the distance of Eps from it. If the number is greater than Minpts, set this point as a core point. Otherwise, set this point as a boundary point and repeat Step 2.1.3; Step 2.2.4: Connect a line between all core points within the distance of Eps. Each group of connected core points forms a cluster; Step 2.2.5: Add each boundary point to the cluster of the core point whose distance from it does not exceed Eps, and the process ends.
3. A method for optimizing the energy consumption of a central air-conditioning system according to claim 1, characterized in that The empirical mode decomposition method in step 3.2 is as follows: determine the local extreme points of the input data x(t), fit the upper and lower envelope lines by cubic spline interpolation, and calculate the mean value m of the envelope lines 1 , subtract the mean value from the original signal to obtain the difference signal y 1 (t): y 1 (t) = x(t) - m 1 judge whether the difference signal satisfies the IMF condition: (1) The number of local extreme points and zero-crossing points must be equal or differ by at most one; (2) At any time, the average value of the upper envelope of the local maximum value and the lower envelope of the local minimum value is zero; If the above conditions are not met, the difference signal y 1 (t) is used as the processing signal to continue empirical mode decomposition until the IMF conditions are satisfied. At this time, the first IMF component obtained is denoted as c 1 (t), and this component is separated from the original signal: r 1 r(t) = x(t) - c 1 r(t) Using the difference signal r 1 (t) as the initial signal, repeat the above steps to obtain the subsequent two IMF components c 2 (t) and c 3 (t).
4. A data acquisition system for the energy consumption optimization method described in claim 1, the system includes attached to the central air-conditioning system: Lora wireless sensing system, Lora gateway and upper computer ； wherein the Lora wireless sensing system includes: 4 + X Lora wireless temperature sensing modules and 2 Lora wireless flow sensing modules; 4 of the 4 + X Lora wireless temperature sensing modules respectively collect the outlet water temperature and inlet water temperature of the evaporator, the inlet water temperature and outlet water temperature of the condenser, and the remaining X Lora wireless temperature sensing modules are installed on the household side for collecting the indoor temperature, where X corresponds to the number of households; the 2 Lora wireless flow sensing modules respectively collect the total pipeline flow of chilled water and the total pipeline flow of cooling water; Each module in the Lora wireless sensing system uses Lora wireless communication to transmit to the Lora gateway, and then the Lora gateway forwards it to the upper computer.