An air conditioner monitoring and parameter identification method of a self-encoding air conditioner collaborative identification model

Abstract

The application discloses an air conditioner monitoring and parameter identification method based on a self-encoding air conditioner collaborative identification model, which comprises the following steps: obtaining operation sampling data of a plurality of air conditioners, and performing resampling and padding processing on the operation sampling data; dividing the air conditioner sampling and padding data into a plurality of subsequences through a preset sliding window, and constructing an air conditioner monitoring parameter identification dataset; constructing an initial self-encoding air conditioner collaborative identification model, performing model training on the initial self-encoding air conditioner collaborative identification model by dynamically adjusting the task weight of the model parameters according to the air conditioner monitoring parameter identification dataset, and obtaining a self-encoding air conditioner collaborative identification model; performing flow reading on real-time operation data of the current air conditioner through a preset sliding window, inputting the real-time operation data into the self-encoding air conditioner collaborative identification model, and obtaining air conditioner monitoring data and physical model parameter identification results. The embodiment realizes accurate air conditioner load monitoring and physical parameter identification, and improves the efficiency and generalization performance of air conditioner physical model parameter identification.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of air conditioner parameter identification, and in particular to an air conditioner monitoring and parameter identification method based on a self-encoding air conditioner collaborative identification model. BACKGROUND

[0002] With the rapid popularization of advanced measurement systems and the rapid development of intelligent algorithms, massive fine-grained information of household-level load data is gradually playing a role and is expected to be used to mobilize demand-side response capabilities to improve new energy consumption capacity of new power systems. At present, residential energy consumption accounts for 40% of total energy consumption, and air conditioners account for 53% of it, which is one of the most responsive potential electrical equipment. The application research of intelligent algorithms for air conditioners can be divided into two categories: air conditioner load monitoring and air conditioner model parameter identification. Air conditioner load monitoring refers to monitoring the active power data of a single air conditioner from the bus-type data of the household electric meter, mainly using machine learning methods to realize the decomposition of air conditioner power consumption based on filtering algorithms and entropy index constraint competition clustering, or using decision trees and long-short term memory neural networks to construct standard deviation, mean and quantile features to identify air conditioners. Air conditioner model parameter identification refers to identifying the parameters of the physical equivalent model from the operating power of the air conditioner and the indoor and outdoor environmental temperature. At present, intelligent search algorithms are mainly used, such as particle swarm optimization-based online identification of the second-order ETP model parameters of variable frequency air conditioners and genetic algorithm-based search of the second-order equivalent thermal parameter model parameters of fixed frequency air conditioners.

[0003] In existing research, air conditioner monitoring and air conditioner parameter identification are two independent upstream and downstream tasks, i.e. the air conditioner operating power obtained by air conditioner monitoring is used for subsequent parameter identification. The independent air conditioner monitoring model has a large decline in monitoring performance in new brands or new operating modes of air conditioners due to limited training samples of air conditioners. The physical model of the air conditioner has a certain generalization, and different air conditioners usually only change the parameters of the physical model, but existing methods cannot introduce guidance information of the air conditioner physical model in air conditioner monitoring, and cannot guarantee the accuracy of the parameter identification result. Although the current air conditioner physical model parameter identification method can meet the requirements of parameter identification of most air conditioners, the algorithm needs a large number of iterations, resulting in slow identification speed and low efficiency. SUMMARY

[0004] The present application provides an air conditioner monitoring and parameter identification method based on a self-encoding air conditioner collaborative identification model, which realizes accurate air conditioner load monitoring and physical parameter identification, and improves the efficiency and generalization performance of air conditioner physical model parameter identification.

[0005] To solve the above technical problems, the present application embodiment provides a construction method of a self-encoding air conditioner collaborative identification model, comprising:

[0006] Obtain running sampling data of several air conditioners, fill the running sampling data by resampling to obtain air conditioner sampling filled data; wherein, the running sampling data includes total line active power of an electric meter, single active power of the air conditioner, indoor temperature and outdoor temperature;

[0007] Divide the air conditioner sampling filled data into several sub-sequences by a preset sliding window, and construct an air conditioner monitoring parameter identification dataset according to each sub-sequence;

[0008] Construct an initial self-encoding air conditioner collaborative identification model, and perform model training on the initial self-encoding air conditioner collaborative identification model by dynamically adjusting the task weight of the model parameter according to the air conditioner monitoring parameter identification dataset, and stop the model training when a preset training end condition is met, to obtain a self-encoding air conditioner collaborative identification model; wherein, the initial self-encoding air conditioner collaborative identification model includes several one-dimensional convolution layers, an air conditioner load monitoring branch and an air conditioner physical parameter identification branch.

[0009] According to the embodiment of the present application, running sampling data of several air conditioners is obtained, the running sampling data is filled by resampling to obtain air conditioner sampling filled data; wherein, the running sampling data includes total line active power of an electric meter, single active power of the air conditioner, indoor temperature and outdoor temperature; the air conditioner sampling filled data is divided into several sub-sequences by a preset sliding window, and an air conditioner monitoring parameter identification dataset is constructed according to each sub-sequence; an initial self-encoding air conditioner collaborative identification model is constructed, and model training is performed on the initial self-encoding air conditioner collaborative identification model by dynamically adjusting the task weight of the model parameter according to the air conditioner monitoring parameter identification dataset, and the model training is stopped when a preset training end condition is met, to obtain a self-encoding air conditioner collaborative identification model; wherein, the initial self-encoding air conditioner collaborative identification model includes several one-dimensional convolution layers, an air conditioner load monitoring branch and an air conditioner physical parameter identification branch. An air conditioner monitoring and parameter identification model based on a self-encoder (self-encoding air conditioner collaborative identification model) is proposed, which integrates air conditioner load monitoring and air conditioner physical parameter identification into a single model for the first time, fuses air conditioner load monitoring (air conditioner power monitoring) and air conditioner physical parameter identification (model parameter identification), improves the accuracy of air conditioner monitoring and air conditioner parameter identification by using the correlation between the upstream and downstream tasks, effectively improves the generalization performance of the air conditioner monitoring model in air conditioners of different brands and different operating modes, and the air conditioner physical parameter identification branch of the air conditioner monitoring and parameter identification model based on the self-encoder (self-encoding air conditioner collaborative identification model) can directly output the identified parameter results without multiple iterations, greatly improving the efficiency of air conditioner physical model parameter identification.

[0010] As a preferred scheme, the initial self-encoding air conditioner collaborative identification model is constructed, specifically as follows:

[0011] The hidden variables are extracted through the one-dimensional convolutional layers, the hidden variables are input into the air conditioner load monitoring branch to obtain air conditioner load monitoring data, and the hidden variables are input into the air conditioner physical parameter identification branch to obtain a physical model parameter identification result, so as to obtain an air conditioner monitoring parameter identification structure, and an initial self-encoding air conditioner collaborative identification model is constructed according to the air conditioner monitoring parameter identification structure;

[0012] The initial value of the model parameter of the initial self-encoding air conditioner collaborative identification model is set as a numerical value in a preset range.

[0013] The air conditioner load monitoring branch includes a plurality of one-dimensional deconvolutional layers, a first fully connected layer and a second fully connected layer.

[0014] The air conditioner physical parameter identification branch includes a third fully connected layer.

[0015] As a preferred solution, an air conditioner monitoring parameter identification dataset is constructed according to each sub-sequence, specifically:

[0016] The average power of each air conditioner unit in each sub-sequence is counted, and a sub-sequence satisfying a first preset extraction condition and a sub-sequence satisfying a second preset extraction condition are randomly extracted to form an air conditioner monitoring parameter identification dataset.

[0017] The first preset extraction condition is a sub-sequence in which the average power of a first preset number of air conditioner units is greater than an air conditioner opening power threshold.

[0018] The second preset extraction condition is a sub-sequence in which the average power of a second preset number of air conditioner units is less than the air conditioner opening power threshold.

[0019] A preset proportion of the sub-sequences in the air conditioner monitoring parameter identification dataset are used as a training set, and the sub-sequences in the air conditioner monitoring parameter identification dataset that are not in the training set are used as a validation set.

[0020] As a preferred solution, the initial self-encoding air conditioner collaborative identification model is trained by dynamically adjusting the task weight of the model parameter according to the air conditioner monitoring parameter identification dataset, specifically:

[0021] The training set is input into the initial self-encoding air conditioner collaborative identification model, a loss function is minimized as an objective function, the model parameters of the initial self-encoding air conditioner collaborative identification model are iteratively calculated based on a back propagation algorithm through an Adam optimizer, and the model is trained by adjusting the task weight of the model parameter through a dynamic weight averaging method.

[0022] The loss function is calculated according to an air conditioner active power monitoring loss function, a weight value of the air conditioner active power monitoring loss function, an air conditioner parameter identification loss function and a weight value of the air conditioner parameter identification loss function, specifically:

[0023]

[0024]

[0025]

[0026]

[0027] In the formula, For loss function, This is the active power monitoring loss function for air conditioning. Identify the loss function for air conditioning parameters. The weight values ​​for the active power monitoring loss function of the air conditioner are: Identify the weight values ​​of the loss function for the air conditioning parameters. The number of samples in the training set. To preset the output window width of the sliding window, This is the output of the air conditioning load monitoring branch of the initial autoencoded air conditioning collaborative identification model. For the training set The th subsequence Active power value of each meter bus type For the training set The th subsequence Active power value of individual air conditioner unit For the training set The th subsequence An indoor temperature value, The indoor temperature is calculated for the first-order air conditioning model of ETP. For the training set The th subsequence Active power value of each meter bus type For the training set The th subsequence An outdoor temperature value, For the training set The th subsequence An indoor temperature value, The sampling interval is... For energy efficiency ratio, For equivalent thermal resistance, This is the equivalent specific heat capacity.

[0028] As the preferred solution, the dynamic weighted averaging method is as follows:

[0029] According to the iterative formula, the weight values ​​of the air conditioner active power monitoring loss function and the air conditioner parameter identification loss function are iteratively updated; the iterative formula is as follows:

[0030]

[0031]

[0032]

[0033]

[0034] wherein, and are respectively and the weight value in the i-th iteration, is the iteration number, is the decay ratio of in the i-th iteration, is the decay ratio of in the i-th iteration, is the decay ratio of in the i-th iteration, is the decay ratio of in the i-th iteration, is the decay ratio of in the i-th iteration, is the decay ratio of in the i-th iteration, is the decay ratio of in the i-th iteration, is the parameter selected by the validation set, is the air conditioner active power monitoring loss function in the i-th iteration, is the air conditioner active power monitoring loss function in the i-th iteration, is the air conditioner parameter identification loss function in the i-th iteration, is the air conditioner parameter identification loss function in the i-th iteration. As a preferred solution, the air conditioner sampling filling data is divided into a plurality of subsequences by a preset sliding window, specifically: According to the window width of the output window and the window width of the input window of the preset sliding window, the air conditioner sampling filling data is divided into each subsequence in time sequence from back to front;

[0035] Wherein, the sampling time stamps of the electric meter bus type active power, the air conditioner single active power, the indoor temperature and the outdoor temperature in the same subsequence are the same; the front end of the output window consistent with the sliding direction is the same as the front end of the input window.

[0036] As a preferred solution, the running sampling data is resampled and filled to obtain the air conditioner sampling filling data, specifically:

[0037] According to the preset filling rule, the indoor temperature and the outdoor temperature are filled backward to the same frequency as the electric meter bus type active power or the air conditioner single active power;

[0038] As a preferred solution, the running sampling data is resampled and filled to obtain the air conditioner sampling filling data, specifically:

[0039] According to the preset filling rule, the indoor temperature and the outdoor temperature are filled backward to the same frequency as the electric meter bus type active power or the air conditioner single active power;

[0040] ​​The preset filling rule is to repeat the indoor temperature and the outdoor temperature for a preset number of times to form a sequence with the same frequency as the total bus active power of the electric meter or the single active power of the air conditioner.

[0041] The preset number of times is a multiple of the sampling frequency of the total bus active power of the electric meter or the single active power of the air conditioner relative to the indoor temperature or the outdoor temperature.

[0042] As a preferred solution, the running sampling data of several air conditioners are obtained, specifically:

[0043] The running sampling data of the air conditioners are measured by the intelligent terminal;

[0044] The intelligent terminal includes an intelligent electric meter, an intelligent socket and an intelligent air switch.

[0045] The total bus active power of the electric meter is the active power collected by the intelligent electric meter or the intelligent air switch from the household user, and the sampling frequency is not less than 1 Hz.

[0046] The single active power of the air conditioner is the active power collected when only one air conditioner is running, and the sampling frequency of the single active power of the air conditioner is the same as that of the total bus active power of the electric meter.

[0047] The indoor temperature is the indoor environment temperature data collected synchronously with the total bus active power of the electric meter or the total bus active power of the electric meter, and the sampling interval of the indoor temperature is not less than 5 minutes.

[0048] The outdoor temperature is the outdoor environment temperature data collected synchronously with the total bus active power of the electric meter or the total bus active power of the electric meter, and the sampling interval of the outdoor temperature is not less than 5 minutes.

[0049] In order to solve the same technical problem, the embodiment of the present application also provides an air conditioner monitoring and parameter identification method of a self-encoding air conditioner collaborative identification model, comprising: reading the real-time running data of the current air conditioner in a streaming manner through a preset sliding window, inputting the real-time running data into the self-encoding air conditioner collaborative identification model, and obtaining air conditioner monitoring data and physical model parameter identification results.

[0050] The self-encoding air conditioner collaborative identification model is obtained by a self-encoding air conditioner collaborative identification model construction method; the real-time running data includes the total bus active power of the electric meter, the single active power of the air conditioner, the indoor temperature and the outdoor temperature.

[0051] In order to solve the same technical problem, the embodiment of the present application also provides a computer device, comprising a processor and a memory, the memory is used to store a computer program, and the computer program is executed by the processor to realize the carbon efficiency analysis method of the electric device.

[0052] To solve the same technical problems, the embodiment of the present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize an air conditioner monitoring and parameter identification method of a self-encoding air conditioner collaborative identification model. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 FIG. 1 is a flowchart of an embodiment of a construction method of a self-encoding air conditioner collaborative identification model provided by the present application;

[0054] Figure 2 FIG. 2 is an air conditioner monitoring parameter identification structure diagram of an embodiment of the construction method of the self-encoding air conditioner collaborative identification model provided by the present application;

[0055] Figure 3 FIG. 3 is a flowchart of an embodiment of an air conditioner monitoring and parameter identification method of the self-encoding air conditioner collaborative identification model provided by the present application. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0057] Embodiment one

[0058] Please refer to Figure 1 FIG. 1 is a flowchart of an embodiment of a construction method of a self-encoding air conditioner collaborative identification model provided by the present application. The construction method comprises steps 101 to 103, and each step is specifically as follows:

[0059] Step 101: Obtain the running sampling data of several air conditioners, perform resampling padding processing on the running sampling data, and obtain air conditioner sampling padding data; wherein the running sampling data comprises the total line active power of the electric meter, the single active power of the air conditioner, the indoor temperature and the outdoor temperature.

[0060] Optionally, step 101 specifically comprises steps 1011 to 1012, and each step is specifically as follows:

[0061] Step 1011: Measure the running sampling data of each air conditioner through an intelligent terminal;

[0062] In this embodiment, the total line active power of the electric meter, the single active power of the air conditioner, the indoor temperature and the outdoor temperature of the multiple fixed-frequency air conditioners are measured by using the intelligent terminal.

[0063] Optionally, the smart terminal includes a smart meter, a smart socket, and a smart air switch.

[0064] In this embodiment, the smart terminal is a smart meter, a smart socket, and a smart air switch with a sampling capability of active power greater than 1 Hz.

[0065] Optionally, the total bus active power of the meter is the active power collected by the smart meter or the smart air switch from the home user at the entrance, and the sampling frequency is not less than 1 Hz.

[0066] Optionally, the single active power of the air conditioner is the active power collected when only one air conditioner is running, and the sampling frequency of the single active power of the air conditioner is the same as that of the total bus active power of the meter.

[0067] Optionally, the indoor temperature is the indoor environmental temperature data collected synchronously with the total bus active power of the meter or the total bus active power of the meter, and the sampling interval of the indoor temperature is not less than 5 minutes.

[0068] Optionally, the outdoor temperature is the outdoor environmental temperature data collected synchronously with the total bus active power of the meter or the total bus active power of the meter, and the sampling interval of the outdoor temperature is not less than 5 minutes.

[0069] In this embodiment, the indoor temperature and the outdoor temperature are environmental temperature data collected synchronously with the active power, and due to the inertia of temperature, the sampling interval is not less than 5 minutes.

[0070] Step 1012: according to a preset filling rule, the indoor temperature and the outdoor temperature are filled backward to the same frequency as the total bus active power of the meter or the single active power of the air conditioner;

[0071] The preset filling rule is to repeat the indoor temperature and the outdoor temperature for a preset number of times to form a sequence with the same frequency as the total bus active power of the meter or the single active power of the air conditioner, and the preset number of times is the sampling frequency multiple of the total bus active power of the meter or the single active power of the air conditioner relative to the indoor temperature or the outdoor temperature.

[0072] In this embodiment, the collected running sampling data is resampled and filled, that is, the indoor and outdoor temperatures collected in step 1011 are filled backward to the same frequency as the total bus active power of the meter and the single active power of the air conditioner. The preset filling rule is to repeat each indoor and outdoor temperature value k times (preset number of times) to form a sequence with the same frequency as the total bus active power of the meter and the single active power of the air conditioner, wherein the preset number of times k is the sampling frequency multiple of the total bus active power of the meter and the single active power of the air conditioner relative to the indoor and outdoor temperatures.

[0073] Step 102: divide the air conditioner sampling filling data into a plurality of subsequences by a preset sliding window, and construct an air conditioner monitoring parameter identification dataset according to each subsequence.

[0074] In this embodiment, the data is divided into a plurality of subsequences based on a preset sliding window, and an air conditioner monitoring and parameter identification dataset (air conditioner monitoring parameter identification dataset) is constructed.

[0075] Optionally, step 102 specifically includes steps 1021 to 1024, each of which is specifically as follows:

[0076] Step 1021: according to the window width of the output window and the window width of the input window of the preset sliding window, divide the air conditioner sampling filling data into each subsequence in time sequence from back to front;

[0077] Among the same subsequence, the sampling time stamps of the electric meter bus type active power, the air conditioner single active power, the indoor temperature and the outdoor temperature are the same; the front end of the output window consistent with the sliding direction is the same as the front end of the input window.

[0078] In this embodiment, the filled air conditioner sampling filling data is divided into a plurality of subsequences based on a preset sliding window, and the window width of the input window and the window width of the output window are used to divide a plurality of subsequences in time sequence from back to front, and the sampling time stamps of the electric meter bus type active power, the air conditioner single active power, the indoor temperature and the outdoor temperature in the same subsequence are the same, wherein the front end of the output window consistent with the sliding direction is the same as the front end of the input window. As an example of this embodiment, and may be 1024 and 256 respectively.

[0079] Step 1022: count the air conditioner single average power in each subsequence, randomly extract a subsequence satisfying a first preset extraction condition and a subsequence satisfying a second preset extraction condition, and compose an air conditioner monitoring parameter identification dataset;

[0080] Among them, the first preset extraction condition is a subsequence in which the first preset number of air conditioner single average powers is greater than the air conditioner opening power threshold; the second preset extraction condition is a subsequence in which the second preset number of air conditioner single average powers is less than the air conditioner opening power threshold; a preset proportion of subsequences in the air conditioner monitoring parameter identification dataset is used as a training set, and subsequences other than the training set in the air conditioner monitoring parameter identification dataset are used as a validation set.

[0081] In this embodiment, the air conditioner single average power in each subsequence is counted first, and a subsequence in which the first preset number (such as: ) of air conditioner single average powers is greater than the air conditioner opening power threshold and a subsequence in which the second preset number (such as: Clause 2) the average power of the air conditioner is less than the air conditioner start power threshold Clause 3) the air conditioner monitoring and parameter identification dataset (air conditioner monitoring parameter identification dataset) is composed of a plurality of subsequences, wherein, Clause 4) the total number of the subsequences after the sliding window division. A preset proportion (for example, 70%) of the subsequences in the air conditioner monitoring and parameter identification dataset (air conditioner monitoring parameter identification dataset) are randomly extracted as a training set, and the remaining subsequences are used as a validation set. As an example of the present embodiment, the air conditioner start power threshold Clause 5) is selected as 35W.

[0082] Step 103: An initial self-encoding air conditioner collaborative identification model is constructed, and the initial self-encoding air conditioner collaborative identification model is trained according to the air conditioner monitoring and parameter identification dataset by dynamically adjusting the task weight of the model parameter, and when a preset training end condition is met, the model training is stopped, and a self-encoding air conditioner collaborative identification model is obtained. The initial self-encoding air conditioner collaborative identification model includes a plurality of one-dimensional convolutional layers, an air conditioner load monitoring branch, and an air conditioner physical parameter identification branch.

[0083] In the present embodiment, the initial self-encoding air conditioner monitoring and parameter identification model (initial self-encoding air conditioner collaborative identification model) is initialized, which means that the model parameters are initialized to floating-point numbers in the range of 0 to 1. The training set obtained in step 1022 is used for model training, and the dynamic weight average algorithm is used to adjust the task weight of the air conditioner monitoring and parameter identification. In the process of model training, the preset training end condition is that the loss function of the initialized self-encoding air conditioner monitoring and parameter identification model (initial self-encoding air conditioner collaborative identification model) is Clause 6) the minimum on the validation set.

[0084] Optionally, the initial self-encoding air conditioner collaborative identification model is constructed, specifically: hidden variables are extracted through each one-dimensional convolutional layer, the hidden variables are input into the air conditioner load monitoring branch to obtain air conditioner load monitoring data, and the hidden variables are input into the air conditioner physical parameter identification branch to obtain physical model parameter identification results, thereby obtaining an air conditioner monitoring and parameter identification structure, and the initial self-encoding air conditioner collaborative identification model is constructed according to the air conditioner monitoring and parameter identification structure. The initial value of the model parameter of the initial self-encoding air conditioner collaborative identification model is set to a value in a preset range; the air conditioner load monitoring branch includes a plurality of one-dimensional deconvolutional layers, a first full connection layer, and a second full connection layer; and the air conditioner physical parameter identification branch includes a third full connection layer.

[0085] In the present embodiment, the air conditioner monitoring and parameter identification structure is, for example, Figure 2As shown, the air conditioner monitoring parameter identification structure is the structure of the initial self-encoding air conditioner collaborative identification model, the numerical value in the parentheses of the one-dimensional convolution layer means (convolution kernel width, convolution step), the numerical value in the parentheses of the full connection layer means the number of neurons, the numerical value in the parentheses of the one-dimensional deconvolution layer means (deconvolution kernel width, deconvolution step), the input power meter bus type active power sequence (subsequence in the training set) is first extracted to obtain the hidden variable In the air conditioner load monitoring branch, the hidden variable is calculated through the deconvolution layer and the full connection layer to obtain the estimated air conditioner single active power air conditioner load monitoring data (air conditioner monitoring data) as the output, since the output sequence length is less than the input sequence, that is, the model is a sequence-subsequence architecture; and in the air conditioner physical parameter identification branch, the hidden variable is calculated through the full connection layer to obtain the estimated physical model parameter identification result, the physical model parameter identification result includes the equivalent specific heat capacity , the equivalent thermal resistance and the energy efficiency ratio . The air conditioner monitoring parameter identification structure in the embodiment is an example, and other different neural network structures can be used to establish the air conditioner monitoring parameter identification structure in actual application.

[0086] In the embodiment, the air conditioner physical parameter identification branch introduced after the hidden vector improves the original sequence-subsequence load monitoring model based on the self-encoder, proposes the air conditioner monitoring and parameter identification model based on the self-encoder, that is, the self-encoding air conditioner collaborative identification model, first integrates the air conditioner load monitoring and the air conditioner physical parameter identification into a single model, proposes the self-encoding air conditioner collaborative identification model that fuses the air conditioner power monitoring and the model parameter identification task, and uses the correlation between the upstream and downstream tasks to improve the accuracy of the air conditioner monitoring and the air conditioner parameter identification.

[0087] Optionally, according to the air conditioner monitoring parameter identification data set, the initial self-encoding air conditioner collaborative identification model is trained by dynamically adjusting the task weight of the model parameter, specifically:

[0088] The training set is input into the initial self-encoding air conditioner collaborative identification model, the loss function is minimized as the objective function, the model parameters of the initial self-encoding air conditioner collaborative identification model are iteratively calculated based on the back propagation algorithm through the Adam optimizer, and the model is trained by adjusting the task weight of the model parameter through the dynamic weight average method;

[0089] In the embodiment, the model training refers to minimizing the loss function For the minimum objective function, the Adam optimizer is used to iteratively calculate the parameters of the air conditioner monitoring and parameter identification model based on the back propagation algorithm. The dynamic weight average method adjusts the task weight of the model parameters. In addition to the dynamic weight average method, other multi-task weight adjustment algorithms can be used to replace the dynamic weight average algorithm for iterative update. The air conditioner physical parameter identification branch can directly output the identified parameter results without multiple iterations, greatly improving the efficiency of air conditioner physical model parameter identification.

[0090] Optionally, the loss function is calculated according to the air conditioner active power monitoring loss function, the weight value of the air conditioner active power monitoring loss function, the air conditioner parameter identification loss function, and the weight value of the air conditioner parameter identification loss function. Specifically,

[0091]

[0092]

[0093]

[0094]

[0095] In the formula, is the loss function, is the air conditioner active power monitoring loss function, is the air conditioner parameter identification loss function, is the weight value of the air conditioner active power monitoring loss function, is the weight value of the air conditioner parameter identification loss function, is the number of samples in the training set, is the window width of the output window of the preset sliding window, is the output of the air conditioner load monitoring branch of the initial self-encoding air conditioner collaborative identification model, is the th electric meter bus active power value in the th subsequence in the training set, is the th air conditioner single active power value in the th subsequence in the training set, is the th indoor temperature value in the th subsequence in the training set, is the outdoor temperature calculation value of the ETP first-order air conditioner model, i.e., the th outdoor temperature value in the th subsequence in the training set calculated by the ETP first-order air conditioner model based on the air conditioner physical parameter value output by the air conditioner physical parameter identification branch of the air conditioner monitoring and parameter identification model based on the self-encoder, is the The th subsequence Active power value of each meter bus type For the training set The th subsequence An outdoor temperature value, For the training set The th subsequence An indoor temperature value, The sampling interval is... For energy efficiency ratio, For equivalent thermal resistance, This is the equivalent specific heat capacity.

[0096] It should be noted that the ETP first-order air conditioning model is a widely used basic model in the field of air conditioning modeling, as shown in the following equation:

[0097]

[0098] In the formula, For equivalent specific heat capacity, For equivalent thermal resistance, For the outside temperature, Indoor air temperature, For air conditioner power, Let be the cooling efficiency of the air conditioner. After discretization, it is shown in the following formula:

[0099]

[0100] In the formula, and This represents the indoor and outdoor temperatures at time k. express Air conditioner power at all times This represents the sampling time interval.

[0101] In this embodiment, the loss function of the fusion ETP first-order air conditioning model can also adopt different functional forms, such as the loss function based on the L1 norm. The loss function designed by the fusion ETP first-order air conditioning model enables joint training of air conditioning load monitoring and air conditioning physical parameter identification. It integrates multi-source information on electrical quantities and temperature with air conditioning model knowledge. During training, the latent vectors are supplemented with the air conditioning physical model parameter identification task as additional guidance, which can effectively improve the generalization performance of the air conditioning monitoring model on air conditioners of different brands and operating modes.

[0102] Optional, dynamic weighted averaging method, specifically:

[0103] According to the iterative formula, the weight values ​​of the air conditioner active power monitoring loss function and the air conditioner parameter identification loss function are iteratively updated; the iterative formula is as follows:

[0104]

[0105]

[0106]

[0107]

[0108] In the formula, and They are respectively and In the Weight values ​​in the next iteration For the number of iterations, For the first In the next iteration The attenuation ratio, For the first In the next iteration The attenuation ratio, The parameters selected for the validation set. For the first The loss function for monitoring the active power of the air conditioner in the next iteration. For the first The loss function for monitoring the active power of the air conditioner in the next iteration. For the first The air conditioning parameter identification loss function in the next iteration For the first The air conditioning parameter identification loss function in the next iteration.

[0109] In this embodiment, a dynamic weight averaging algorithm is used to iteratively update the weights of the two tasks, air conditioning load monitoring and air conditioning physical parameter identification, to solve the trade-off between the importance of air conditioning load monitoring and physical parameter identification tasks during training. The dynamic weight averaging algorithm is used to dynamically adjust the loss weights, which effectively improves the stability of training and model performance.

[0110] Example 2

[0111] Accordingly, see Figure 3 , Figure 3 This is a flowchart illustrating Embodiment Two of the air conditioning monitoring and parameter identification method based on a self-encoded air conditioning collaborative identification model provided by the present invention. Figure 3 As shown, the real-time operating data of the air conditioner is read in a streaming manner through a preset sliding window. The real-time operating data is then input into the self-encoded air conditioner collaborative identification model to obtain the air conditioner monitoring data and physical model parameter identification results.

[0112] The self-encoding air conditioner collaborative identification model is obtained by a construction method of the self-encoding air conditioner collaborative identification model; and the real-time operation data include total bus active power of an electric meter, single active power of an air conditioner, indoor temperature and outdoor temperature.

[0113] In the embodiment, the total bus active power of the electric meter, the single active power of the air conditioner, the indoor temperature and the outdoor temperature during the operation of multiple fixed-frequency air conditioners are measured by using the intelligent terminal, the collected data are resampled, the data are divided into multiple subsequences based on a sliding window, an air conditioner monitoring and parameter identification dataset is constructed, the self-encoding air conditioner collaborative identification model is initialized, the model is trained by using a training set of the air conditioner monitoring and parameter identification dataset, and a dynamic weight average algorithm is used to adjust the task weight of air conditioner monitoring and parameter identification. After the self-encoding air conditioner collaborative identification model obtained by training is deployed and applied, real-time total bus data, indoor temperature data and outdoor temperature data are read in a streaming mode based on the same sliding window as when the subsequences are divided, the model is input, air conditioner monitoring data are obtained in the air conditioner load monitoring branch, the air conditioner monitoring data include the operating power of the air conditioner, and physical model parameter identification results are obtained in the air conditioner physical parameter identification branch, the physical model parameter identification results include equivalent specific heat capacity, equivalent thermal resistance and energy efficiency ratio.

[0114] The embodiment of the present application proposes a self-encoding air conditioner collaborative identification model that integrates the air conditioner power monitoring and model parameter identification tasks, uses the correlation between the upstream and downstream tasks to improve the accuracy of air conditioner monitoring and air conditioner parameter identification, integrates the multi-source information of electrical quantities and temperature and incorporates air conditioner model knowledge, adds the air conditioner physical model parameter identification task as additional guidance to the hidden vector in the training model, and can effectively improve the generalization performance of the air conditioner monitoring model in air conditioners of different brands and different operating modes. The air conditioner physical parameter identification branch of the self-encoding air conditioner collaborative identification model can directly output the identified parameter results without multiple iterations, greatly improving the efficiency of air conditioner physical model parameter identification. To solve the trade-off problem of the importance of air conditioner load monitoring and physical parameter identification tasks in training, a dynamic weight average algorithm is used to dynamically adjust the loss weight, effectively improving the stability of training and the performance of the model.

[0115] In addition, the embodiment of the present application also provides a computer device, which includes a processor and a memory, the memory is used to store a computer program, and the computer program is executed by the processor to realize the steps in any method embodiment described above.

[0116] The air conditioner monitoring and parameter identification method of the self-encoding air conditioner collaborative identification model described above is based on the construction method of the self-encoding air conditioner collaborative identification model described above. The optional items in the method embodiments described above are also applicable to the present embodiment, which will not be described in detail here. The remaining contents of the present embodiment can be referred to the contents of the method embodiments described above, which will not be described in detail in the present embodiment.

[0117] The above detailed description of the specific embodiments of the present application is provided for the purpose of further explaining the objects, technical solutions and advantages of the present application, and it should be understood that the above is only a specific embodiment of the present application and is not used to limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for constructing a self-encoding air conditioner collaborative identification model, characterized in that, The method comprises the following steps: Obtain running sampling data of several air conditioners, fill the running sampling data by resampling to obtain air conditioner sampling filled data; wherein the running sampling data comprises total bus active power, air conditioner single active power, indoor temperature and outdoor temperature; Divide the air conditioner sampling filled data into several sub-sequences by a preset sliding window, and construct an air conditioner monitoring parameter identification data set according to each sub-sequence; Construct an initial self-encoding air conditioner collaborative identification model, and perform model training on the initial self-encoding air conditioner collaborative identification model by dynamically adjusting the task weight of the model parameter according to the air conditioner monitoring parameter identification data set, stop the model training when a preset training end condition is met, and obtain a self-encoding air conditioner collaborative identification model; wherein the initial self-encoding air conditioner collaborative identification model comprises several one-dimensional convolution layers, an air conditioner load monitoring branch and an air conditioner physical parameter identification branch; Wherein, the training set is input into the initial self-encoding air conditioner collaborative identification model, the loss function is minimized as the objective function, the model parameters of the initial self-encoding air conditioner collaborative identification model are iteratively calculated based on the back propagation algorithm through the Adam optimizer, and the model training is performed by adjusting the task weight of the model parameter through the dynamic weight average method; Wherein, the loss function is calculated according to the air conditioner active power monitoring loss function, the weight value of the air conditioner active power monitoring loss function, the air conditioner parameter identification loss function and the weight value of the air conditioner parameter identification loss function, and specifically: In the formula, Let the loss function be... Let the active power monitoring loss function of the air conditioner be denoted as . Identify the loss function for the air conditioning parameters. The weight values ​​are those of the active power monitoring loss function for the air conditioner. Identify the weight values ​​of the loss function for the air conditioning parameters. The number of samples in the training set. The width of the output window of the preset sliding window. This is the output of the air conditioning load monitoring branch of the initial autoencoded air conditioning collaborative identification model. For the training set, the first The th subsequence Active power value of each meter bus type For the training set, the first The th subsequence Active power value of individual air conditioner unit For the training set, the first The th subsequence An indoor temperature value, The indoor temperature is calculated for the first-order air conditioning model of ETP. For the training set, the first The th subsequence Active power value of each meter bus type For the training set, the first The th subsequence An outdoor temperature value, For the training set, the first The th subsequence An indoor temperature value, The sampling interval is... For energy efficiency ratio, For equivalent thermal resistance, This is the equivalent specific heat capacity.

2. The method of claim 1, wherein the self-encoding air conditioner collaborative recognition model is constructed by, The initial self-encoding air conditioner collaborative identification model is constructed, and specifically: The hidden variables are extracted through each one-dimensional convolution layer, the hidden variables are input into the air conditioner load monitoring branch to obtain air conditioner load monitoring data, and the hidden variables are input into the air conditioner physical parameter identification branch to obtain physical model parameter identification results, thereby obtaining an air conditioner monitoring parameter identification structure, and the initial self-encoding air conditioner collaborative identification model is constructed according to the air conditioner monitoring parameter identification structure; The initial value of the model parameter of the initial self-encoding air conditioner collaborative identification model is set as a value in a preset range; The air conditioner load monitoring branch comprises several one-dimensional deconvolution layers, a first full connection layer and a second full connection layer; The air conditioner physical parameter identification branch comprises a third full connection layer.

3. The method of claim 1, wherein the self-encoding air conditioner collaborative recognition model is constructed by using a convolutional neural network (CNN) and a recurrent neural network (RNN). The air conditioner monitoring parameter identification data set is constructed according to each sub-sequence, and specifically: The air conditioner single average power in each sub-sequence is counted, and the sub-sequences satisfying the first preset extraction condition and the sub-sequences satisfying the second preset extraction condition are randomly extracted to form the air conditioner monitoring parameter identification data set; The first preset extraction condition is that the sub-sequence with the first preset number of air conditioner single average power is greater than the air conditioner opening power threshold; The second preset extraction condition is that the sub-sequence with the second preset number of air conditioner single average power is less than the air conditioner opening power threshold; A preset proportion of the sub-sequences in the air conditioner monitoring parameter identification data set is used as a training set, and the sub-sequences other than the training set in the air conditioner monitoring parameter identification data set are used as a validation set.

4. The method of claim 1, wherein the self-encoding air conditioner collaborative recognition model is constructed by using a convolutional neural network (CNN) and a recurrent neural network (RNN). The dynamic weight average method is specifically: According to the iterative formula, the weight values of the air conditioner active power monitoring loss function and the air conditioner parameter identification loss function are iteratively updated; wherein the iterative formula is specifically: wherein with respectively with in the first iteration, is the iteration number, is the decay ratio of the in the first iteration, is the decay ratio of the in the first iteration, is the validation set selected parameter, is the air conditioner active power monitoring loss function in the first iteration, is the air conditioner active power monitoring loss function in the first iteration, is the air conditioner parameter identification loss function in the first iteration, is the air conditioner parameter identification loss function in the first iteration.

5. The method of claim 1, wherein the self-encoding air conditioner collaborative recognition model is constructed by using a convolutional neural network (CNN) and a recurrent neural network (RNN). The air conditioner sampling filling data is divided into a plurality of subsequences by the preset sliding window, specifically: According to the window width of the output window and the window width of the input window of the preset sliding window, the air conditioner sampling filling data is divided into each subsequence in time sequence from back to front; Wherein, the sampling time stamps of the electric meter bus type active power, the air conditioner single active power, the indoor temperature and the outdoor temperature in the same subsequence are the same; the front end of the output window consistent with the sliding direction is the same as the front end of the input window.

6. The method of claim 1, wherein the self-encoding air conditioner collaborative recognition model is constructed by using a convolutional neural network (CNN) and a recurrent neural network (RNN). The running sampling data is resampled and filled to obtain air conditioner sampling filling data, specifically: According to the preset filling rule, the indoor temperature and the outdoor temperature are filled backward to the same frequency as the electric meter bus type active power or the air conditioner single active power; Wherein, the preset filling rule is to repeat the indoor temperature and the outdoor temperature for a preset number of times to form a sequence with the same frequency as the electric meter bus type active power or the air conditioner single active power; The preset number of times is the sampling frequency multiple of the electric meter bus type active power or the air conditioner single active power relative to the indoor temperature or the outdoor temperature.

7. The method of claim 1, wherein the self-encoding air conditioner collaborative recognition model is constructed by using a convolutional neural network (CNN) and a recurrent neural network (RNN). The running sampling data of several air conditioners is obtained, specifically: The running sampling data of each of the air conditioners is measured through an intelligent terminal; Wherein, the intelligent terminal includes a smart meter, a smart socket and a smart air switch; The electric meter bus type active power is the active power collected by the smart meter or the smart air switch from the household user's entrance, and the sampling frequency is not less than 1Hz; The air conditioner single active power is the active power collected when only one air conditioner is running, and the sampling frequency of the air conditioner single active power is the same as that of the electric meter bus type active power; The indoor temperature is the indoor environment temperature data collected synchronously with the electric meter bus type active power or the electric meter bus type active power, and the sampling interval of the indoor temperature is not less than 5 minutes; The outdoor temperature is the outdoor environment temperature data collected synchronously with the electric meter bus type active power or the electric meter bus type active power, and the sampling interval of the outdoor temperature is not less than 5 minutes.

8. An air conditioner monitoring and parameter identification method of a self-encoding air conditioner collaborative identification model, characterized in that, It includes: Real-time running data of the current air conditioner is read through a preset sliding window, and the real-time running data is input into a self-encoding air conditioner collaborative identification model to obtain air conditioner monitoring data and physical model parameter identification results; Wherein, the self-encoding air conditioner collaborative identification model is obtained by the construction method of the self-encoding air conditioner collaborative identification model in any one of claims 1-7; the real-time running data includes electric meter bus type active power, air conditioner single active power, indoor temperature and outdoor temperature.

9. A computer device, comprising: The air conditioner monitoring and parameter identification method comprises a processor and a memory, the memory is used for storing a computer program, and the computer program is executed by the processor to realize the air conditioner monitoring and parameter identification method of the self-coding air conditioner collaborative identification model.

Images